BI 211 COGITATE: Testing Theories of Consciousness

[00:00:03] Speaker A: It didn't come out perfectly. So there are things to account for with respect to the orientation, with respect to the maintenance and the synchrony between the areas. So I look at it as an overall win in terms of both theories. Have some homework to do. [00:00:24] Speaker B: I think you could be pessimistic and say, well, this theory, you know, they made a wrong prediction and therefore I'm a proprietor and I think the theory completely wrong and we just can't. We need to invent a completely new theory. I don't think that's a very valid approach. I think it's more trial and error and theories are working process. [00:00:43] Speaker C: Even if you do this adversarial collaboration, people rarely if never change their mind. Oh, right, right. I don't think the aim of this project was to change Stan or Julio's mind. One project is not enough. But we are providing questions and I'm sure there will be many, many PhD students that will start new projects just from what we did. And I hope that will be the case for what we did with Cogitate. [00:01:17] Speaker D: This is brain inspired, powered by the transmitter. Those are the voices of Roni Hirschhorn, Alex Lepov and Oscar Ferrante, three of the many scientists that comprise the Cogitate group. Cogitate is an adversarial collaboration project to test theories of consciousness in humans. In this case testing the integrated information theory of consciousness and the global neuronal workspace theory of consciousness. I said it's an adversarial collaboration. So what does that mean? It's adversarial in that two theories of consciousness are being pitted against each other experimentally. It's a collaboration in that the proponents of the two theories had to agree on what experiments could be performed that could possibly falsify the claims of each of their theories. The group has just published the results of the first round of experiments in a paper titled Adversarial Testing of Global Neuronal Workspace and Integrated Information Theories of Consciousness. And this is what Roni, Alex and Oscar discussed with me today. The short summary is that they used a simple task which they describe and measured brain activity with three different methods, Intracranial eeg, MEG and fmri. And they made predictions about where in the brain correlates of consciousness should be, how that activity should be maintained over time, and what kind of functional connectivity patterns should be present between brain regions. So the take home the results are a mixed bag, with neither theory being fully falsified, but with a ton of data and results for the world to ponder and build on to hopefully continue to refine and develop theoretical accounts of how brains and consciousness are related. So we discussed the project itself, many of the challenges they faced along the way, and continue to face their experiences and reflections, working on it and coming together as a team. The nature of working on an adversarial collaboration when so much is at stake for the proponents of each theory. And as you heard last episode with Dean Guanamano, when one of the theories, integrated Information Theory, is surrounded by a bit of controversy itself regarding whether it should even be considered a scientific theory. Okay, thank you for being here. I hope you enjoy this conversation. Thank you to my Patreon supporters who get full episodes, extra content and access to the full archive of brain inspired episodes. And here are Roni, Oscar and Alex. So this is going to be a real. I was just rereading the manuscript and oh my God, it's so much. Okay, so welcome. Roni, Oscar, Alex, thank you for being here. Let's start by just giving an overview of what you've been up to, what this project is, what Cogitate is. Alex, would you like to begin or anyone can jump in here. [00:04:27] Speaker B: Right, yeah, sure. Thanks a lot for having us. It's very exciting. Yeah. So I'm Alex Lepov and I've been working for the past, I guess, four or five years on the Cogitate project, which is an adversarial collaboration trying to arbitrate between two different theories of consciousness, the Integrated Information Theory and the Global Neural Workspace theory. And there's actually quite a few things that are specific about the project that I'm sure we'll go into. But I guess the first thing to say, it's not only me, Ronnie and Oscar, but it's also a lot of other collaborators. So we have, I think, a couple different first authors on that paper. So we have Ursula Goska, Klimovska, Simon Henin, Aya Hallaf, Ling Liu, David Richter and Yamile Vidal, who aren't here today. And we have also three leading PIs, Lucia Meloni, Liard Mutryk and Michael Pitts. And so what that means is it was special because it was an adversarial collaboration project, but also because it was a massively collaborative team science enterprise, so to speak. [00:05:30] Speaker D: And just to sort of hash out what an adversarial collaboration project is, it is kind of, in this case, two sides pitted against each other who have decided to come together and try to come up with experiments, testable experiments, with predictions that would fit into their various theories. And have to agree on those experiments. And then I'll leave it to you guys to say what your role is to be sort of impartial. Not jurors, but experimenters, researchers, in this case, Oscar Roni. Did Alex miss anything that you would add? [00:06:11] Speaker A: It's actually even bigger than that. It's even bigger than Cogitate. So Cogitate is one of a series of five such adversarial collaborations. So it's not just these two theories, it's more theories that are being tested. And yes, the idea is to sort of maybe level the playing field following this fragmentation of each of the theor working on its own, with its own methods and its own sets of predictions that don't really correspond when you look at the literature, you don't really know how to even begin to compare them and so on. So the idea was, from the Templeton World Charity foundation, is to start this series of collaborations bringing several theories to the table at the same time, agreeing on sets of experiments that can test their predictions at the same time. So quadratates as big as it is even bigger than Alex mentioned, because this is just like the center PIs and the first authors. We have like, I think 50 people on the experiment one paper. So that's even just like one out of. So yeah, it's even bigger than that. But the idea is the Kahneman idea of solving disagreements by working together instead of just. [00:07:32] Speaker D: Who's I? The Kahneman? [00:07:34] Speaker A: Daniel Kahanman. [00:07:36] Speaker D: Oh, he invented adversarial collaboration. [00:07:39] Speaker A: Well, he talked about it in the beginning of the 2000. And Lucia Meloni really likes to mention that. So we are all affected by his impact on our project, specifically. Right, yeah. [00:07:56] Speaker D: Oscar, it looked like you wanted to add something there. [00:07:58] Speaker C: Yes, sorry, Yeah, I was just to say that we big group of people and many other projects similar to Cogitate, but this was like the first one. So we are very proud of what we started. But also we want to mention that this is just the beginning of something that hopefully will be bigger than our paper and what we've done so far. And also in Cogitate there are. So this is what we want to talk today is a paper that we have published on Nature. But there's also other projects associated with Cogitate and another experiment that we're working on. So there is a lot of things and we would like to talk about what we did so far, but also to keep track for what other people are doing at the same time. [00:08:53] Speaker D: Well, you said that this was just the beginning, but in A prior conversation that we were having. You guys have already sort of gone through academic levels from the beginning of the project. So maybe by way of introducing yourselves more. So talk a little bit about. More about who you are and what you do and how you came to this project and maybe your role in the project. Because everyone, you're all working together as a group, but you all have sort of specialized roles, if I have it that correctly. Yeah. [00:09:28] Speaker C: Who was the first one to join among us? [00:09:32] Speaker B: Yeah, Ronnie, probably. [00:09:33] Speaker A: So. [00:09:33] Speaker B: Yeah. [00:09:36] Speaker A: It'S hard to join, I think, because Alex and Oscar like you. Maybe I should have been last on this. I am a student. My name is Voni. I'm a student at Liad Mudryk's lab at Tel Aviv University. I'm now in my PhD. When I first heard of Cogitate, it was when Liad came back from Seattle. It was March 2018. I was in the second year of my bachelor's in computer science and psychology. [00:10:11] Speaker D: So naive. [00:10:12] Speaker A: Yes, so naive. I was an RA in her lab. I was working on another project. She came back, she was extremely excited and. Yeah. And then she said she's going to need some hands on deck with this. With this new. With a new thing. By the third year of my bachelor's, I was already piloting Experiment two. I was already fully incogitate. So that was. Yeah, that was the end of 2018. [00:10:42] Speaker D: Yeah, it's just the beginning. I like that. [00:10:45] Speaker A: That was just the beginning. [00:10:46] Speaker D: That was. Well, sure, that was the beginning for you. Right? But. But then. Yeah. Anyway. All right, Oscar, maybe you want to go next. [00:10:54] Speaker C: Yeah. And I joined in 2019 after another meeting. Also this meeting was in the US and that meeting was in Chicago and was like the starting meeting of the project. When the project officially started and we were deciding the name of the group and then we end up with Cogitate. And I went there with Professor Ole Jensen. He works at the University of Oxford at the moment. But before he was at the University of Birmingham and I was postdoc with him and I joined the project. So he presented the project to me as an opportunity to use my skills in MEG magnetogephalography, which is the technique that I use in Cogitate in my research. And yeah, then that day I had the chance to meet in person most of the people in Cogitate, some of them were online, like Alex. So I had a chat to exchange the chance to exchange some words with Alex online. And then. And it was really exciting. And after that, I wouldn't have imagined how things would have ended. It was unpredictable. [00:12:04] Speaker D: Right. But it sounds like so far, two of the three of you were kind of recruited, but also excited to join. So was it like a recruitment thing or. And then Rony, you mentioned there was an advertisement for the project. So were you guys recruited or did you volunteer or was it sort of in between? [00:12:22] Speaker C: Yeah, I can. In my case, I was already in Ole Jensen's lab and we were thinking how to continue to work together. And then there was this opportunity and he talked about the project and everything. And I had a quick chat with Lucia Meloni, who's the main PI, together with Liad Mudryk and Michael Pittsburgh. And I liked the idea. It was completely different from what I did before. So I joined as a postdoc. So compared to Alex and Roni, I was more experienced. So I had more expectations probably. And then. [00:13:00] Speaker D: But what do you mean? More expectations about what? [00:13:04] Speaker C: Like, I joined not only for the science, but also because I thought it was a good way to go ahead with my career in academia, because there is all these people involved and a super interesting topic and a big project and something that I was excited to see how that would have affected my career in academia. [00:13:29] Speaker D: I didn't even think about that. But that is kind of a shrewd move because you're automatically and integrated into a vast network of people that you're going to be joined with. And we'll talk about how you guys have become friends through the process soon. But yeah, I hadn't thought about the networking aspect, the career aspect of it, because there's a separate way that it could maybe hurt your career, even potentially. Right. [00:13:55] Speaker C: I guess all talk about that if you. [00:13:57] Speaker D: Yeah, yeah, we'll talk about that later. But so then. So then, Alex, how did you come to the project and what's your role? [00:14:02] Speaker B: Yeah, I guess in my case that was a bit a similar story to Ronnie. So at the time, I think 2019, I joined the lab of Lucia Meloni. I just finished my master's in neurobiology. It was more like the, you know, the biology aspect of neuroscience, so molecular and cellular neuroscience type of things. But then I also got really interested in, you know, things like consciousness research and system neuroscience and EEG and so on. So then I started looking for labs and there was the opportunity in Lucia's lab to be a lab manager. So it wasn't really an advertisement for the Cogitate product. It was more just to take care of daily operations at the lab. And I was lucky enough to get the job. But right at the time I started that was also when Cogitate started, meaning there was a lot of things that needed to be done, which as a lab manager, I was kind of like just free hands that could take on any task that was floating around. So I just started working a lot on the Cogitate. And that was also at the time where they were starting to put together all the teams. So I think to go back perhaps to the structure of the project itself. So it all started in one meeting in Seattle where the funding agency, so the Templeton One Charity foundation, kind of brought together a lot of PIs, including two of the founders of two leading theories of consciousness. And they kind of came together and discussed kind of a think tank of coming up with ideas for experiments, which then later became the Cogitate project. And then as part of that project, they also had the ambition of having various specific experiments that brings about contradicting prediction of the theories. But they also wanted to test them in the most detailed and precise way that was available. And to do that, the strategy was to use basically all the best recording tools that exist right now in system neuroscience. So it would be eeg, meg, fmri as well as intracranial eeg. And so then they had the experiment pretty much set up. And then they started to also think of who could be leading the data collection and data analysis of each of these modalities. And yeah, and there they hired PIs that were supposed to be theory neutral and that would also be in charge of hiring postdocs or PhD students that would be theory neutral. And at the time, one of the roles that needed to be fulfilled was for the IEG team that Lucia was kind of in charge of putting together. And she thought of me as a. Then finishing up my time as a lab manager to become a PhD student specialized in the intracranial electrons and photography for the COGITI project. So I guess for the recruitment process, I'm not too sure. I guess it depends on each of the labs, kind of each of the PIs of each of the lab that was involved in each of the modality data collection was kind of in charge of deciding whom to hire for that and how to go about it. [00:16:57] Speaker A: I'd say I think it's pretty similar. Think Alex and I are misrepresenting the rest of the. Because most of all of the rest of the first authors are postdocs like Oscar. So after this call that Alex mentioned, they decided the theory neutral labs that will centralize the data collection, the analysis, and each of their respective PIs hired postdocs. Alex and I, starting as working in Liad and Lucia's labs were kind of like, well, you're already here. And then Alex came to the IEG team and I came into the data monitoring team, which is supposed to be also theory neutral and also modality neutral in the sense of overseeing all of them at the same time. [00:17:42] Speaker D: Do you guys have a sense of. So they had the initial meeting, right, where the idea was brewed up and then they had to agree on experiments. Do you have a sense of that process, how long and difficult a process that was? Because it just, it seems like such a difficult thing to do. And we'll get into sort of the predictions in the experiments that were made. But I mean, I'll just jump the gun here and say like even in the manuscript, I'm not sure how the one that is going to eventually be published words it. But you don't give judgment on the outcomes, partially because I think you could poke holes in any of the results, right? And there's commentary from the proponents of each of the theories in the manuscript, where of course, they support the results that support their theory and then they can poke holes in the ones that don't. And so all of this is up for interpretation. And so just given that you can always poke holes in any results because everything's open to interpretation, it seems so difficult to agree on experiments that should be run to test the hypotheses with clean predictions. So do you have a sense of how long that took or what, you know, how that, how that process was? And then I really want to know also is what did you think was going to happen? [00:19:11] Speaker A: It did take long again from the perspective of like maybe also Alex came like from making the stimuli themselves to piloting each of the experiments to actually getting to run them. And that's after the proponents and the center PI's were already technically done with the conceptualization of the paradigms themselves. So that was even months before. So the entire thing was very iterative and took. [00:19:37] Speaker D: Took a long time because do the proponents have to agree. So then you have to. You make the stimuli, you have to send the stimuli to everyone and they say, oh, it's too bright on the left. That sort of. Was it that sort of iterative process? [00:19:50] Speaker A: So, so not, not to that level with the proponents every time. But like with again, who Lucia. Mike centralized this from, from both ends. But like the design itself, the paradigm itself, the fact that they can make meaningful predictions to the theories using these paradigms, for example. Yeah, they had to sign in on that. They had to physically sign on that. By the way, like to say that these, that these experiments are good to test their predictions on these things. And I will say that it didn't start with only i18 Gen W. There were more theories in this project. And that was exactly the stage where if you didn't agree that it can test your theory, you would opt out instead of opting in. Right. So in that sense, yes, it took a long time from that meeting that they had in Seattle until to bring it to the actual running it with predictions everyone agrees on. But I think that to your question, yes, at least back then when they signed it and when we started running it, everyone agreed that these theories can be meaningfully tested using these experiments. That was the whole point. [00:21:13] Speaker B: Right. And I guess maybe if I can jump in on that one, I guess it's the question like what took a long time. And what we realized along the way is, you know, there's many different layers in agreeing on something. So I think by the time we started, so as you said, everything related to the conceptual idea of what are the experiments going to look like, there's two experiments, where did that come from? And so on, that all happened before we actually joined the project. So I have an impression of what happened, but I have actually no idea if that's true because that's something none of us were there in that meeting. But I guess my impression is that the conceptual idea was a lot of the conceptual work, so to speak, was done at that one meeting where they kind of, you know, got the project started back in Seattle in 2018 or at some point. But then, you know, you have sort of the broad idea of how we're going to go about testing the theory, but then actually going down to the implementation and operationalizes. [00:22:12] Speaker A: Oh, that's sorry, operationalization. [00:22:14] Speaker B: Yeah. Thank you. It's like that is basically a lot of the time what we did is, oh, we have all these predictions, let's go about and just do it. And then you realize, oh, but we forgot to specify that to the level of detail that would be necessary. And then whenever we hit such a roadblock, then we'd actually have to go to the theory again and ask like, oh, we didn't specify that to the level of details that would actually enable us to do a proper analysis. Therefore we decided this and that would that be okay with you? And that's kind of the back and forth process that took a really long time. [00:22:49] Speaker C: And also the project evolved from the original idea because the original idea is, okay, we have two theories. They are both theories of consciousness, but they have different predictions. So we should find ways to say, okay, we have these questions. We could have two possible outcomes, one in favor of theory A, in favor of iit, and the other one in favor of gnw. But then working while we start working on the project, and then when the theoretical questions became some practical hypothesis to test, we also realized that some questions were slightly different for the two theories. So then what we were actually testing were not exclusive. So there were some questions where both could be right. Another question where the answer, the response, the outcome for one theory was irrelevant for the outcome of the other theory. So it's also evolved from going from the theoretical idea to the actual practic. [00:23:47] Speaker D: Yeah, so it went from. That's the way all scientific projects go, right? Where you have like a couple, couple alternative hypotheses that you want to test. You set out to test them, and then day, hour, minute 12 of day one, you realize, oh, no, it's. It's not going to be that clean. Right. And so this seems like that, you know, times 20, you know, so I'm not sure what that. What I want to ask you is like, has it, has it all been a series of headaches or is that, has it all been worth it so far? [00:24:23] Speaker C: I'm happy that you're not asking us which theory one, because I know, I. [00:24:29] Speaker D: Know not to ask you that. But you're. You can't, you're impartial, right? You can't, you can't say. We'll talk about it offline. Don't worry. No, but I mean, did you so. So kind of immediately, you know, to my earlier question, how did you think it was going to work out? Not in terms of which theory would win, but just would this work as an endeavor and would we get a clean answer? Right. Because the goal is to falsify. Not to, like, not to approve of either of the theories, but to falsify those, the predictions, essentially. So did you think it was all just going to work out cleanly in the beginning? [00:25:11] Speaker A: I will say that, like, to do it a little justice to the unclean of the project. It is, it is unique also in the aspect of. We have works showing, including works from my colleagues in my lab, for example, Italian, showing that actually most of the experiments do not lay out the hypotheses and predictions so nicely in advance with respect to falsifying theories of consciousness. What they do is they do experiments, and then when they get the results they post hoc interpret them as supporting, for example, GNW or providing support for iit. That's actually how most of them go. So even the attempt to go in advance and detail it out as messy it ends up being. I think that's already put some things into boxes or into frameworks that make it more streamlined, even if the boxes inside are like full of clutter and mess. So I think even that makes it more helpful and maybe more hopeful in terms of how clean it can be if we continue with that approach of detailing these things beforehand. Exactly. Because of what you said of finding out at minute 12 that oh, it's not really detailed enough or specific enough or not really what I meant when I said I predict that this would happen, stuff like that. [00:26:35] Speaker D: Right. [00:26:36] Speaker B: I guess I was also trying to think back and I guess one of the things we're also very worried about. I don't remember if we end up mentioning it in the paper, but there's this effect of the 15 IQ points as well that afterwards you kind of imagine yourself that you knew that that's what's going to happen all along. And therefore it's difficult to go back in time and imagine what I was thinking back then. And even more so because I think like as me being a very junior researcher, just basically starting my PhD or barely when I first encountered all the prediction, I really didn't know much about the theories themselves. And so it's been a really long kind of learning along the way of, you know, how things. I think it took me a little bit of time to reach out a point where I could sort of have an expectation of how it will pan out. But. [00:27:24] Speaker D: Oh, has this ruined your, your outlook on science in general? Just. No, actually it's such a beautiful idea. Right. And then, you know, the implementation is what it is and you guys did a fantastic job it seems, of sticking to it and getting through those obstacles. [00:27:42] Speaker B: Right? Yeah, but I guess, but I think one thing that I would say that maybe we. That might contradict a bit what you said, Ronnie. I think like there was, you know, messiness in the sense of when you go to the conceptual ideal, to the actual implementation, there's so many degrees of freedom that you can always go back and forth endlessly. But I think one thing, if I remember correctly, I would need to go back to the actual first version of the pre registration which came out. I don't remember exactly when, but actually pretty early in the project. Yeah, I think it was 2019. Right. I think already there we kind of had I think the conceptual idea of how the theories are going to be tested didn't change all that much. That in the sense. So in the one experiment we are talking about represent stimuli for different durations. And there was always this idea. GenW says there's going to be decoding in the prefrontal cortex versus it says there's going to be decoding in posterior region. And we have different, similar durations. So according to gnw, the activation of PFC is going to scale up with the duration of the stimulus and it makes a similar prediction in posterior cortex. So all of that kind of has been there since the very beginning. Now it's about like, you know, well, when you say decoding, which classifier do you want to use? And when you say this and that, what's the actual method? And what is actually the brain region that counts and doesn't count? Like, this is where there's a lot of, you know, back and forth. But the broad idea, I would say, has been pretty consistent from the beginning of the project. And I think I remember distinctively by the time I'd kind of like reached the maturity of understanding what Genevieve was saying. I think one of my. I think I was expecting. Yeah, I was like, oh, yeah, this prediction seems very reasonable to me and I wouldn't be too surprised if, you know, this pans out. [00:29:23] Speaker A: But having it like being predictable and making sense when you read it is one thing, and making it in advance is another thing. Like, that's the thing that I find myself telling people the most when I talk about the project is like, think that they had to come up with that beforehand. It makes sense when you read the whole thing together. It makes sense when you already perform the experiments and do the analyses and stuff like that. But it had to make sense before all of that. And I think that it's not like in the sense of think of our eyes. Right. What does it mean to. [00:29:56] Speaker D: Sorry, sometimes I'm going to jump in. Because regions of interest. And when people say iit, it's always integrated information theory. And GNW is global neuronal workspace. [00:30:07] Speaker A: When a theory says, I expect this and this activity in the front or in the back, posterior areas of the brain, that's nice. That's even backed by literature. But then if you go again, if you go to literature, people can vary far and wide in what they consider to be actually posterior, what they consider to actually be frontal. And then when you come down to the nitty gritty of that, you find that there is no Literature detailing what are the specific regions that this theory predicts that will be related to the maintenance of conscious percept and stuff like that. So I think that saying it in advance had meaning because we had to go there, we had to do that. And that was the first time it was actually done is this project, this is the first time that the theories actually defined those regions of interest that are related to what we tested in experiment one, the maintenance and consciousness, visual stimuli. It wasn't done before. So, so it's easy to say, oh, it makes sense, right? When you read it. It makes sense. Of course it makes sense. But someone had to do it. [00:31:15] Speaker D: But even. Okay, so then, so then you have to define the regions, but then you put an EEG cap on which EEG has high temporal resolution, really low spatial resolution because you're inferring where the neural signals are coming from because it has to pass through the cranium. Then you have to decide, even if you have like a really high density EEG net, which just means there's a lot of electrodes on a lot of different places on your skull, then you have to decide which electrodes are we going to pay attention to and which frequencies of which electrodes with which question and which range, how to define that frequency range. Just at every step there are decisions to be made that I think degrees of freedom, Alex, that you said earlier, where these degrees of freedom just open and then that leaves everything open to poke holes in as well. So it's just so hard. [00:32:08] Speaker C: Yeah, it's hard. And also it's not that common. So nowadays there are more studies that are pre registered, but like here we have to pre register everything before starting analyzing the data. So I can say something about how we approach the data analysis, because that's also very peculiar of this project. [00:32:26] Speaker D: But let's do that. But then I also want to. I'm burying the lead because we. I want to talk about the actual predictions and why those predictions were what they were and then kind of how it came out. But yeah, please go ahead. [00:32:38] Speaker C: Yeah, so what we did is to collect a lot of data. We collect the meg, micro denture phagalography, electroencephalography intercanial electroencephalography and FMRI in six different labs all around the world, actually seven, because we had an extra lab at the. And then what we did is to take all the participants, all the data we collected, and divide them in two chunks. One small chunk of participants was used to develop the analysis. So we developed analysis pipelines for all the modalities from pre processing to all to the analysis required to test all the different predictions. And we only have access to this limited amount of participants. And then after that we pre registered the methods so that all the. We publicly shared this also in a conference in Amsterdam some years ago. And then only then when we had everything pre registered, everything was then approved by the theory leaders. And only that time we had access to the big data set, which is what we use in the paper, in the Nature paper. And so the results that we show are results that were conducted on an independent, on an orthogonal sample compared to the one that we use to develop the analysis. [00:33:56] Speaker D: But when you say develop, so the goal is to pre register. So you have to, you use a small cohort of the large population sample. And then are you, what are you, what are you trying to figure out there, like which classifier to use and. [00:34:11] Speaker C: Yeah, for instance. Yeah, for instance, there is a prediction about where in the brain is the content of our conscious experience. [00:34:21] Speaker D: But you're not getting the results from that cohort and analyzing the results. You're actually just go ahead. [00:34:28] Speaker C: Yeah, we are not using those results to test the theories, but we are using those data to see whether the analysis method that we are using to decode, for instance, using different classifiers or in terms of interior connectivity, there are thousands of different measures that one can use to estimate connectivity between different brain regions. And these are all things that were kind of on us and the lab, different modalities, PIs. So these are not things that the theory can propose because they propose how the consciousness works, but not how to analyze MEG or FMRI or intercarrial EEG data. So there was a lot of work on that side, and then only when everything was good from a technical point of view and approved from also external consultant, only at that point we could test the actual theories, the actual predictions. [00:35:23] Speaker D: So then how do you adjudicate between. Well, let's take an example, right? You said that there's so many analyses for inter aerial communication. How do you, how do you figure out which one is the right one to pre register? [00:35:38] Speaker C: Yeah, I think I can take this because. Yeah, you choose the. Actually that's my. [00:35:45] Speaker D: Okay, let's just go to decoding classifiers. That'll be more straightforward. Maybe. See, already we can't even. It's not straightforward, right. [00:35:54] Speaker C: It's a long story. For decoding, for instance, what we try to do is to have control analysis and then develop the pipeline using the actual data that we're gonna use, but not the final sample and then also find control analysis things that should be there anyway independently of what the theory predicts. And if we can find those things and then the data can be used to also test the predictions, then that means that this is how we can analyze the final sample. [00:36:33] Speaker D: Okay, I think I understand that. So you're not. So you have to decide which analysis to use for any given question based on things that have been in the literature that have nothing to do with the theories. But it's like just a reality check of whether you can use this machine learning classifier to decode whether. Whether you can decode whether someone's looking at a ball or a house or something like that. [00:37:01] Speaker B: Exactly, right. Yeah, exactly. So I think essentially we use some sort of benchmarks, if that's what you want to call them. And in the case of the decoding analysis, we are lucky that. So one of the prediction is, you know, we should decode faces from objects in different brain regions that are relevant for the theories. We were considering using benchmarks such as responses and so on. But the thing is, even in both of the regions, we did find decoding of faces versus objects. So we knew our classifier is sensitive to picking up something. Had that not been the case, then we would have optimized on something else to try to avoid any biases. So we had decoding in both of the regions for this one particular feature. And then we tried to optimize to get the highest decoding accuracy possible by trying a couple different things, like using pseudoscharles and so on. [00:37:52] Speaker D: But you had to do it with a task that was unrelated to the task that you eventually use to test the theories. Right. [00:38:00] Speaker B: It can be on the same task, especially for in the case of the MEG and fmri because we have this extra data set that's untouched. [00:38:06] Speaker D: Right. [00:38:06] Speaker B: So the question is, does all parameters generalize to another data set? Because if not, it's just overfitting. Right, but that was baked in. We had the same task running hundreds of participants, and we use 20 of them to optimize our pipelines, and then just check if it generalizes to the other participants. [00:38:22] Speaker D: And in that case, don't you. Like if let's say you run a classifier, right, and you see it in prefrontal cortex, and you see it in. And not in posterior cortex, then that already says something about which of the theory which. About which theory that evidence leans towards. Right? You already know something about that. So in that case, you're like confounding the control with the eventual I'm sorry I'm harping on this. I'm just trying to understand it. [00:38:51] Speaker A: It's totally legitimate, so. Not exactly. Because first of all, even if you look at it like that, again, like Oscar said, we didn't use it for that. We didn't use the patterns that we saw when we use the optimization data, even once we optimized our methods, you get something. We don't use that to inform the orthogonal data set that we have. That's one. The other thing is that when you find or not find things, like Alex said, if you find the decoding that you want and you know that your methods are good enough, if you don't find, then you're right that it might be because you didn't use a sensitive enough method, you didn't use a good enough method. And then you need to go back to the drawing board and prove to yourself and also to all of your 49 other colleagues on your paper that it is that your methods are good enough to identify something and that something is really not there. And not that we could go on and try a different method or that there is some problem with the sensitivity of what we were doing and so on. So I think that kind of controls for that. But you're right that, like, it's kind of. It's kind of a risk to look at this tiny, like, sample out of our bigger sample and start thinking, oh, what the results might be. But I actually think that this is the advantage of working in such a large multimodal team. If you're tempted to do that, you kind of always prevent it from doing that because you always need to be neutral. And again, like, demonstrates this neutrality. Linear analysis to everyone else. [00:40:33] Speaker C: Basically, yeah, there's replication. So we try to replicate everything that we did within modalities, like in this case with different samples, but also between modalities. Because, for instance, what you mentioned before, deciding which frequency band to analyze, when we look at the time series data, we have to standardize this between MEG EEG and intercannial EEG and which classifier to use that was also standardized amongst the different teams. So there was also an internal, like different levels of internal replication that we aimed for. [00:41:10] Speaker D: Okay. Okay. So hopefully the listeners will have a grand sense now of how challenging something like this is. And I'm sure we'll come back to some of those challenges as we kind of move through the meat of, of what you guys were testing. Right. So maybe we can kind of go through. And we don't need to go into the nitty gritty details. But sort of the big picture of each of the. There are three predictions essentially that you guys tested and pre registered and then tested as we were just talking about. So would someone like to just describe. Let's start with the decoding. Oh, Ronnie disappeared. Oh, you're back. But let's start with decoding since it's the first one in the paper and it's also the example that we were just using because there's a challenge already with decoding. But who would like to describe what the prediction was and the predicted results for both of the theories and why and then how you went about testing it. [00:42:09] Speaker A: Should I start? Yeah, I can. I can start with maybe. I wonder if we first need to say something about this theories for it to be. [00:42:19] Speaker D: Yeah, oh yeah, yeah, please. I mean I will do that a little bit in the introduction as well, since we didn't. I meant to do that beforehand but because we could spend hours just talking about the theories themselves as well. So how much weight to put on each of these topics is a challenge, right? [00:42:36] Speaker B: Yes, but. [00:42:36] Speaker D: But yeah, please, please. That'd be great if you say something overview about that. [00:42:40] Speaker A: Just like in a nutshell. So the decoding, for example, makes sense, right? Because. Okay, so the decoding question is basically can be simplified to where in the brain can we find this specific information about the content that is being experienced right now? Right, because this experiment is all about super threshold stimuli. So it's not some sort of manipulations of consciousness. We assume that you are aware of the stimuli and like you said before, you have faces, you have objects and they are presented for different durations. So where can we find in the brain information that is related specifically to that experience content? [00:43:19] Speaker D: And why are we looking at where? Why is where? The important question. [00:43:24] Speaker A: Where is the important question with respect to the theories? Because, for example, if you consider global neuronal workspace theory, they expect they view consciousness as this sort of message that being broadcasted, selected to be broadcasted in the brain. So that workspace is located according to gnw, in more frontoparietal regions of the brain. So that's what you can derive and predict from gnw. But for example, with integrated information theory iit, because of the way they define consciousness, they actually start. It's not a cognitive neuroscience theory like gnw. It starts from first person perspective and experience and its structure. That whole thing of the structured experience is expected to be actually correlated with more posterior areas of brain. So this is why where even matters if we Think about the theories and their neurobiological implementations. They expect to have different answers to the question of where, which is why decoding is an interesting prediction in the context of this experiment. [00:44:38] Speaker D: I just want to. Let me just pause and say, Oscar and Alex do jump in, like, feel very, very free just to jump in. Right. Anytime. [00:44:45] Speaker A: So, yeah, anything I. Anything I missed or misrepresented? Any. Yeah, no, I think it's the simplest one. Right. In terms of. [00:44:54] Speaker D: To understand, at least not in terms of interpretation, but yeah, to understand. [00:44:59] Speaker A: Yes, to understand the motivation. Because one can be very, very crudely simplified to front and one can be very, very crudely simplified to back also, like in terms of how it was presented in the previous picture. [00:45:14] Speaker D: And you use the word correlation there. And is that why decoding, was this the proper analysis here? Because decoding, if you can decode something from a data set, it means that there is some correlation of that information. Right. [00:45:32] Speaker A: So we would go even further in the project. We go to like maximally decodable, right? We go to what. What information is not only correlates with that, but do more areas add, like, improve that decodability in the brain or not? That was, for example, a main question from Integrated Information Theory. Do frontal areas, if we add them, does it add to the decodability or do the posterior areas, are they sufficient? Right. So it's both this correlation and also this correlation specifically to the areas that they defined, if I understand the question correctly. [00:46:14] Speaker B: Yeah, yeah. [00:46:14] Speaker D: I just wanted to make the distinction between correlation and causation. And decoding is about correlation and not causation. [00:46:22] Speaker A: Yes, right. [00:46:23] Speaker B: But I guess maybe one different way of putting it, if I may. Wrong is basically, you know, if it makes a difference in terms of your conscious experience, whether you see a face or an object. Therefore, whatever brain region is involved in consciousness also has to have a similar distinction. Now, you can look at this in terms of a simple univariate analysis. So you compare activation from one electrode or whatever between two different stimulus categories. That would be one way of looking at, does this region or electrodes or whatever represent information that's consistent with what experience is? Or you could look at it in a multivariate sense. And that's what a decoding analysis does. Right. So it's like instead of looking at just one electrode, you look at the pattern of activation across many electrodes. And that tells you. Oh, yeah, indeed, these brain regions seems to have a similar distinction in terms of activation that is also present in experience. So that's minimum requirement for that brain Region to be involved in consciousness. And that is correlation. [00:47:22] Speaker A: Yeah. But even that is not enough in the sense of like because we don't manipulate consciousness in this experiment. That even if we find, I think the more meaningful result would be to not find an expected pattern. Because if we find the expected pattern which goes back to the correlation, it doesn't really mean that it's unique to conscious experience. Right. It can be representation that can also be there when you're not conscious of the stimulus. So that that's a point to be careful on. With respect to the posit more positive results of the project. [00:47:52] Speaker D: We should describe the task, I think can we do that fairly succinctly in a way that like listeners could. But everything's easier visually. Of course. But let's describe the task and then we'll carry on with that first prediction and experiment. [00:48:08] Speaker C: So we use a very simple task like we didn't manipulate consciousness because that is something that we didn't want to do for the first experiment. And what we did is just to present centrally in front of the participant different images so they were very clearly presented so that they were always experiencing these images. And we presented images of different categories like faces, letters, objects, fold fonts so that we could use this as information about the content of the experience. And we also manipulated the duration which each of this image was presented so that we could also test the duration of the experience. Some images were presented for half a second, other images for one second and other for one second and a half. And then we also manipulated the relevance of these images. So before presenting a series of images, we show the participant 2 Target pictures and then they were presented with a series of pictures and they have to find the target. So they only had to respond when the target was in front of them. And then sometimes the target could be a specific face and the participant was presented with a face which was not a target. So in that case we will consider that picture as a task relevant non target picture because it's same category as the target but it's not the target itself. While other images are images that are from different categories like objects, if you're looking for faces, for instance. And in that case we have simuli that are non relevant for the task. So these are the main manipulation, the three main manipulations. [00:49:53] Speaker D: Yeah. So very, very simple task in ways that you can kind of get rid of some of the confounds of responding, whether it's a higher cognitive issue. Right. During the task. Whereas a lot of consciousness based tasks manipulate the stimuli where it's like ambiguous and whether you're attending to a house or a face, if it's house, half house, half face and things like that. So there are lots of confounds that you guys wanted to get rid of. So these are very simple images. It's very easy task. I didn't look, but I'm sure people performed at ceiling like what, 99%? Something like that. [00:50:36] Speaker B: Yeah. [00:50:36] Speaker A: In the 90s. Very, very good. [00:50:38] Speaker B: Yeah, yeah. [00:50:39] Speaker D: So very, very clean. Everyone's happy with the task, right? [00:50:42] Speaker C: Yeah. And it's so that we have this task relevant and task irrelevant images also because we want to exclude the effect of responding because there's also a big topic in the search of consciousness about report tasks. And there was another manipulation that is the orientation of these pictures. So some of the images were presented in the front view and some in a lateral view. So. And we had all these kind of different manipulations in terms of the content of the visual stimulus. Because the experience should be full, should contain all the different components of what we're seeing, not just the category. Because decoding category is one thing, but then the experience should also contain information regarding the orientation of the face that we're seeing or the relevance of it. [00:51:31] Speaker D: You all use the Mona Lisa as an example in the paper where your conscious experience is. Oh, hey, that's the Mona Lisa. But also you notice the skin tone, the slight facial expressiveness, the background. Right. So there's the phenomenal. The phenomenal experience of consciousness is I think multi dimensional is the way that it's phrased in the paper. Yeah, okay. Okay. I think. Did we miss anything or should we go back now to the, to the predictions? Right. So we have the task and now we're predicting there's a difference between decodability of certain features in the front and back. I don't know if you want to continue, Ronnie. [00:52:10] Speaker B: I guess. Sorry, if I may jump in. I guess one thing also I think which is the very overarching theme regarding all the predictions. Exactly what Ronnie said before is because we don't have this confound like when you do consciousness experiment where you have an unconscious and conscious condition, you can not really always know is the unconscious condition always truly unconscious or is there sometimes conscious trials mixed in there and the other way around and so on. In that case, it' simpler. There's no doubt that potspain always experience what's right in front of their eyes because it's so big. No attentional competition, no nothing. And so that means that if a prediction of a theory fails, they can hardly pull the card of saying, oh, maybe it's because the experimental contrast wasn't as clean and maybe they didn't experience it. It's a really hard argument to make to say, oh, well, maybe they didn't notice any difference between seeing a face and seeing an airplane. Like that wouldn't. That wouldn't make sense. [00:53:03] Speaker D: Right. However, you could still zone out right during the task and maybe maybe not be consciously aware of one of the stimuli as it passes through, maybe catnap, you know, for example, etc. [00:53:18] Speaker B: That is an option. But then you would expect performances to be actually lower. Right. Like, not in the 95% because they still have to respond to some of the stimuli and there's no reason why they would zoom out less in the target stimuli that they have to respond magically. They don't know. So then, yeah, that's kind of the control for that critique. And that's kind of the other basically, that runs through all the predictions. Like, well, if we find it, that doesn't really tell us much because we don't know if there is unconscious processes that are mixed in with the conscious processes. But if we don't find it, then that is a clear challenge for the theory. [00:53:52] Speaker C: And there's the control experiment that you ran, Alex. [00:53:54] Speaker B: Yeah, yeah, exactly. Because we also tested do they remember after the fact the stimuli that were presented? And especially, is there any differences in memory performances between different task relevance conditions? And the simple answer is basically, no, there is no such differences. They can remember, not perfectly, because they are not told. It was a surprise memory test. They weren't told beforehand. You're going to have to remember them because you can have a test afterwards. It was a surprise, but they still perform fairly well, which means they experience it, most of them, and can remember them after the fact. And no significant differences between the different types of relevance conditions that combined with. [00:54:33] Speaker A: The behavior and the eye tracking. Like, we see that they gaze directly at the stimuli. We see that they respond to them when they need to, when they're very accurate. So all of that together, it's just like, yeah, even like, they were on task, the task was very easy. They were looking directly at the stimuli. And then if we bring it back to the decoding, then the question is, okay, is all that information that we have in the multidimensional conscious experience that Oscar mentioned, so can we decode the category? If that's a face or an object, can we decode the orientation? Can we decode those bits of information in the areas that the theories a priori defined as relevant. Right. That's the whole meaning of doing these things beforehand. Because it's not like, okay, let's try and then post hoc say, oh, you know what, I found something in the front that's interesting. So that was the hard part of it, of being specific and say, okay, but where do you expect to find it? And by it we mean the specific, like the ability to decode face versus object, the ability to decode this lateral view versus the frontal view and the ability to do that both in the task irrelevant and the task relevant stimuli. Because there's no reason, if we talk about experience, that there will be a difference there. If there is, that's a problem because it means we found something related to the task and not visual perception, to. [00:55:58] Speaker D: Responding, as Oscar was saying. Yeah, yeah, exactly. [00:56:03] Speaker A: So that is the first prediction. That is basically the question of decoding. And then maybe, well, let's, let's go. [00:56:11] Speaker D: Ahead and maybe talk about just how that came out, right? So then we can kind of wrap up the predictions and we don't have to go through like every prediction because you know, people will read the paper and there's way too much to discuss, but just kind of the over overview. [00:56:25] Speaker B: Right Of. [00:56:27] Speaker D: So I was going to jump in when Alex earlier said, and the simple answer is that no, it wasn't a problem. And I was going to say, dear listeners, that is the last simple answer you will hear for the rest of this episode. Which is not really true, but just kind of overall, what were the results of that first experiment prediction? [00:56:48] Speaker A: So overall, like for category, it is true that we did find the decoding of category from task irrelevant to task relevant in the other way around. We found it both, but it was not only tested in the FMRI modality, it was also, like we talked before, tested in the other modality. So that's also a very important point for the, for the project because what we do find, even though it's not when we both. When we find something and when we don't find something, if it's consistent across the modality, that makes it more robust. So even if we cannot say something about conscious perception and it can be visual perception without consciousness, it's still something that is consistent, for example, between the FMRI and the intracranial EEG recordings. That is very nice that it was consistent. The orientation was. Correct me if I'm wrong here, but the orientation wasn't that simple, especially for gnw. Decoding of orientation was not found in the areas that GNW predicted. Even though, like, it's. It's simpler than the others, it's still not. Not perfect in terms of brushing it off, because it's still something that needs to be explained. If we have all the information about experience, why. Why isn't orientation there? [00:58:11] Speaker D: Right. Yeah. And this, this would maybe I'll just jump in and say this. The reason why we've been talking about how complicated all this is is because a lot of the results are. Are a mixed bag. Right. Are they lend some support for one theory, less support for the other, and then we move on. And then another one lends a little bit of support for both, we move on, and then another experiment lends a lot of support in one modality, very little in the other modalities. For example, I'm making these things up, but that's just sort of the way that science works. And then we have. We're. We're left to interpret what all that means. Right. [00:58:49] Speaker A: But I would flip it. I would flip it. Sorry. From support to challenge. So I would. I would easily. I would brush up the support and say, okay, but here I found this challenge because then also challenge is something for the theory to explain. Support is something that to me is not interesting as a reader. Right. Because it's sort of. [00:59:09] Speaker D: Okay, well, also in the paparian falsification sense, challenge is more in line with that as well. So thanks for correcting me. [00:59:17] Speaker C: Yeah. Because if we support all the theories, we're not going anywhere because that means that all of the theories are right. So nobody's right. In that case, we have to challenge them. [00:59:29] Speaker D: Okay, well then. All right, I'll restate what I said. So there are lots of different challenges and lots of different modalities given the different. [00:59:36] Speaker A: Yeah, yes, that's a good summary. [00:59:38] Speaker D: Sorry, sorry. Okay, so, but so then the end result, then, Ronnie, like in terms of the different theories, like the, the decoding predictions. What, without interpreting, without giving a gloss on the interpretation, what. What are the challenges? Right. Which one is more challenged? Right. [00:59:59] Speaker A: So in terms of content, I would say both of them were like, not really challenged to the same degree with respect to decoding of orientation. I would say that GNW was more challenged on that prediction overall. [01:00:15] Speaker D: Neither. Neither theory was fully falsified by prediction one. [01:00:20] Speaker A: Yes, correct. [01:00:23] Speaker D: Okay, so maybe. I'm not sure if Alex or Oscar, maybe Oscar, since you mentioned during the task about the different durations, because one of the things that the next prediction. Oh, sure, sure. Yeah, yeah. Okay. Alex, maybe you can talk about the second prediction. [01:00:42] Speaker B: I might be a bit biased there, but that's my favorite prediction because that's the one I've worked the most on, probably. Okay, I think. [01:00:48] Speaker D: Yeah. [01:00:48] Speaker B: So that's. This one is actually back to what you were saying. That's the only simple answer you'll get. I think this one might also be a simple answer and also very simple stories. At the end of the day, these are not very complicated predictions to understand, despite the complexity of the theories involved. But so it's present three different stimuli for three different durations, right? And so you look at the screen, you see a face for 500 milliseconds, and you see a face for 1.5 seconds. You would expect that to make a difference in terms of your contrast experience. So you'd expect, however you want to phrase it, if it's about time perception, which I don't think you necessarily need to go there, but just generally in terms of time, by time, what's the content of your experience? The fact that you have a face after 1.5 seconds still on the screen versus just a blank screen does make a difference in your experience. And that's something the theories must be able to account for to some extent. Integrated Information theory made a very simple prediction. So if we have a phase that's presented for a given duration, we have a stable experience of that stimulus for that duration. Therefore, we expect sustained activation and also sustained content representation, or sustained decoding, if you will, in the region that they believe is relevant to consciousness, which is the posterior cortex. On the other hand, you have gnw, which says something a little bit more convoluted, which is, well, they expect at the onset of the stimulus, you have an increase of activation, but then it goes back to baseline. And then at the end of the stimulus, you have a reactivation that basically marks the end. So you have something that marks the beginning, something that marks the end. And importantly, both these events are somewhat content specific. So they represent distinctions in what was just on the screen. So what is appearing and what is disappearing. And then you basically bridge in between the two. [01:02:37] Speaker D: So that first bump, that's the ignition of global network neural workspace. So maybe just say a word about that. And you mentioned that that ignition event, it's not just an ignition, it's somewhat content specific. [01:02:52] Speaker B: Yeah, exactly. So basically, according to gnw, what makes something being contrast is that it is. So you have a content that basically travels from lower sensory regions all the way to frontal cortex or parietal cortex. And if it's, you know, there's a couple reason why it might or might not trigger conscious experience. But if it becomes conscious, what happens is you have a nonlinear ignition in the frontal parietal cortex that's basically a really strong increase of activation all of a sudden. And that's basically transmitting this information or making this information available to wide variety of processes that fulfill various cognitive functions. So this ignition is really necessary to make the information so what you experience available to many processes. And it's this, the fact that it's this specific information is available to the processes that make it conscious. According to Gnome. [01:03:46] Speaker D: So but, but it should have a slightly different quote unquote representation that ignition event based on the different content so that it can signal what the content is when it's ignited. [01:03:58] Speaker B: Yeah, exactly. So the ignition, you can rephrase that if I'm not mistaken, as amplification of the content for it to become conscious. Okay, so then yeah, it has to be content specific because if it were not, you know, there's no information there about what's in front of your eyes. So then you would experience something a specific which is not what experience is like. [01:04:16] Speaker D: Okay, so then just. I just want to like make sure I state it as clearly as possible. So IIT says that for like a long duration, the posterior cortex should come on and stay on for the duration. Whereas global neural network space theory says that in frontal cortex there should be this ignition event that amplifies the signal and then it goes back to baseline and then at the offset there's another ignition event which signals the offset and onto another conscious experience. [01:04:46] Speaker B: But I think like the slightly confusing part about that prediction I would say is the offset. So it's like this ignition, they're always kind of markers of two things is what just finished and what is starting now. So you have the ignition at the beginning of the stimulus is marking that there is a new stimulus following a blank screen. Pretty much that signals. Now we have a face following a blank screen. And after that, after the offset of the stimulus predictions, now we have a blank screen, but following the face that was just there. So there is always. So it's a temporal anchor. And basically because if you it's following the idea of Daniel Dennett, I think of how time and the observer paper, which is. So you don't need to always represent things for the duration that they are there and any temporal information about them. You just need to send information with a temporal tag so that then you can reconstruct based on them temporal tags of the different events that occurred. [01:05:40] Speaker D: Okay. And so it doesn't. Because I was going to say this doesn't feel like my phenomenal experience, that I'm always jumping from one conscious, content experience. And then that one ends, and then a new one begins. Then that one ends and a new one begins. Right. But it sounds like the activity, the dynamics could be not from the ignition, but from the ongoing activity elsewhere. [01:06:03] Speaker B: Yeah, I guess there. That's a bit debated, I think, because whenever I talk about this to different people, I realize everybody have slightly different interpretations of that. But my personal interpretation, which is. I think what Stan would always say is it's kind of reconstructed backwards. Like, you have the ignition at the offset, and if you attend to what you've just experienced, then you'd have the impression that you had a sustained experience of something, when in fact there was no sustained experience at the moment that it was there. That's one interpretation. The other one is the idea of activity silent, which is. I think the truth is probably in the middle, which is in the between. In the baseline period, information is still represented in an activity silent state, which is basically encoded through synaptic potentiation, I think you would say. So. Like, it's kind of saved in the. [01:06:54] Speaker D: Synaptic weights that facilitate silent synapses. [01:06:58] Speaker B: Yeah, exactly. [01:07:00] Speaker D: Okay. Yeah. [01:07:00] Speaker B: Okay. So that's kind of like the two interpretations, but I think it's just me misunderstanding because Stein has a very precise interpretation. [01:07:10] Speaker D: Oh, okay. But. But so that the ignition, let's say, is not the. Itself, the contents of your consciousness. It's a marker of the fact that there is a certain kind of content that will be in your consciousness. [01:07:26] Speaker B: No, it is. [01:07:27] Speaker D: It is. [01:07:27] Speaker B: It is. Yeah. It is the content of your consciousness. [01:07:31] Speaker D: Okay. They're. They're equated. [01:07:33] Speaker B: Yeah. [01:07:33] Speaker D: The ignition is the content. Yeah. Okay. All right. [01:07:36] Speaker B: That is then broadcasted. And that gets. So you have the information that's, you know, amplified through the ignition, and then this information that's accessed by the processors is conscious. [01:07:48] Speaker D: Okay. All right. [01:07:51] Speaker B: Or maybe to reiterate on the idea of reconstructed backward, it seems, like, really not intuitive because that's not what our experience feels like. But I think the argument from the global workspace theory, and generally speaking, the illusionist came. So we believe that consciousness is more of a. An illusion. That's kind of the point. Like, well, you know, you wouldn't know. But whenever you look at it, whenever you try to capture yourself in what is your experience right now while you reconstruct it, and therefore you have the impression that you had a sustained experience on Experience like this or like that. [01:08:26] Speaker A: But you don't even have to go to that extreme, like in the sense of expecting the neural activations that happen, like the patterns of information computation that's happening in the brain to match experience. I think that's a whole other discussion. So the fact that you have like this ignition and the onset and offset. That doesn't mean that this is what it's like to experience, to have an experience. [01:08:50] Speaker D: But doesn't that. Isn't that what content means in this case? If the ignition is the content, doesn't that mean it's equated with what it's like? [01:08:59] Speaker A: So the ignition is like what's being broadcasted, like this moment of broadcasting of this content throughout the brain. And then like Alex said, you can have many interpretations of how it's maintained if it's silent or if it's. You're actually. But like in. In the end of the day, it's on them to explain how that accounts for this feeling of continuity that you have. [01:09:26] Speaker D: So this is one of the differences between global normal workspace and integrated information is that integrated information theory starts with the. What it. It's what it's like, the phenomenal aspect. And they have axioms from which they derive the theory, whereas global normal workspaces. Just saying here's the brain activity that is related to the conscious stuff. And then they have that later step to explain the feeling of what it's like. [01:09:51] Speaker A: Yes. Like maybe not in a. In a. Like in a nutshell, yes. But in any case, for any, for any theory of consciousness that claims to have a neurobiological implementation, it's on them to explain conscious experience, the neurobiological implementation and the correspondence between them, which doesn't have to be as like one to one as I expect. When my experience is maintained to see a sustained activation in the brain, that's one way to go about it. Another way might be more computationally saving resources. Right. If I'm silent in between these two, these two bursts, that might also have some justification in terms of if their presentation is still there in silent synapses and so on. That doesn't necessarily mean that I'm not consciously experiencing the stimulus at this time. I'm just saying. [01:10:41] Speaker B: Right. And I guess also maybe another way to make that prediction a little bit more intuitive. GNW doesn't say that there will never be sustained activity in the prefrontal cortex when we are seeing a stimulus for a given duration. They actually say there will be a sustained activation and content representation if we are effortfully maintaining this information in our conscious experience. So, for example, if you have to attend, if you are told, attend specifically to the stimulus really precisely for as long as it's there, or try to attend to its duration and so on, Basically anything that makes it such that you have to attend for the stimulus for its entire duration, then we would expect sustained activation of piece. But if we don't, then we just encode the beginning, the end, and reconstruct in between, and then we can kind of build the story after the fact of us having a sustained experience. If we look at it. I think that's what they would say. [01:11:32] Speaker A: Just for the completion. And then I'll shut up about it. I just want to say that both theories, both GNW and IIT and more other theories, they assume consciousness is discrete. Right. And not continuous. Like in terms of implementation wise. Right. So the feeling of continuity isn't of itself a type of content of conscious experience. But in the matter of fact, IIT too, assumes at the bottom, discreteness just at a different. At a different level. Just. Just so it's not like one of them assumes this continuity, then the other one doesn't. They are both actually assuming that consciousness is. It happens like in a discrete manner, just like a different way of going about it. [01:12:12] Speaker D: All right. [01:12:13] Speaker B: Okay. [01:12:13] Speaker D: All right. [01:12:13] Speaker C: So. [01:12:14] Speaker D: Sorry. No, no, that's okay. So, all right, so we have different durations of the stimuli, and you're going to test the durations based on the predictions from it and global neuronal workspace theory. [01:12:25] Speaker B: Right, Exactly. And so that's what we did in all three modalities. And I've worked mostly on the intracranial data, but I think in general, what we find is in the posterior cortex there is sustained activation and sustained content representation. So we use an RSA method for that, which is very similar to a decoding approach. But that's basically, we can essentially decode a phase that's presented for 1.5 second for 1.5 second, and we can decode a phase that's present for 0.5 second for 0.5 second. [01:12:55] Speaker D: I just want to say RSA is representational similarity analysis. And there are different ways of doing that. I just wanted to be using an acronym. [01:13:04] Speaker B: Yeah. And so, in other words, we found that the prediction from IIT in the posterior cortex just works out the way they were saying in the prefrontal cortex. What we find is that the prefrontal cortex does not care for the duration of the stimuli. And I think that's one of the predictions. I think the failure of that particular prediction is really significant for gm, but. [01:13:29] Speaker D: So why would PFC prefrontal cortex need to care about the duration if it. [01:13:34] Speaker B: Makes a difference in our conscious experience? So the assumption of the experiments, of the experiment that's in the published in Nature right now, is we present stimuli for three different durations and we experience them for three different duration, or experience is associated with the stimulus duration one way or another. I think that's like the minimal assumption. Duration does make some sort of a difference in our conscious experience. And it's not the same to experience a stimulus for 0.5 second than it is for 1.5 seconds. That's the main assumption. And so if that is true of conscious experience, then the fact that GNW doesn't reflect this property of conscious experience implies that something is missing either from the theory or something is wrong with our initial assumption. These are the two options, essentially. [01:14:23] Speaker D: So if the assumption is right, global neural workspace is significantly challenged in this. [01:14:29] Speaker B: Exactly. Yeah. And I think it's also important to emphasize why that's the case at a different level. And that's something, I think presenting the results at different conferences, I think that's something people often miss is all right, we see a stimulus, you might experience it for whatever, how long you want to experience it, or you're attending to it. But then when the stimulus disappears, there is all of a sudden massive change in the visual input. There is a stimulus disappearing, and all of a sudden there's just a blank screen there. And so according to gnw, for something to be conscious experience, you need this ignition. So the fact that we don't have an ignition at the disappearance of the stimulus, that means that participants don't experience it, according to Genevieve. So that means that they fail to notice something as drastic as the stimulus disappearing. That's the implication for the theory, if they want to stick to the core principle of the theory, which they do want to do. [01:15:27] Speaker D: Yeah. Okay, so that's. So we have two out of three. Anything to add before we move on to the third prediction? [01:15:36] Speaker B: I think, yeah, just one thing. And then also I'll shut up. So that's actually, if you read the discussion of the paper, that's one. And again, that doesn't seem. That seems completely counterintuitive in the reaction that most people have when you say that. It's like, well, no, that's hardly sensical, but maybe it is true that participants don't experience that the stimulus disappears. And that's essentially Akin to what you were saying in the beginning, sometimes you just zoom out. Right. And based on the structure of our task, where we repeat the stimuli for many, many times and we have the participant, they know at some point they've learned the structure of the task. They know they only need to care about it when it first appears. Who says that you can't just zoom out as soon as you have performed the task and then go direct your attention inwards and think about something else? [01:16:25] Speaker D: Right, right. And in that case, even though the visual displayed is different, that might not make it up to the region that needs to be ignited because you're not attending to it. [01:16:33] Speaker B: Exactly. [01:16:34] Speaker A: On the other hand, then it would mean that JW will have to come up with a different explanation of how. What accounts for you having experiences of different durations? Right. So what does differentiate seeing Alex for 20 seconds versus seeing them for half a second? [01:16:54] Speaker B: So we're actually now working on a follow up on that, actually. Precisely. But the answer is simple, is if you have to care about it, then that will make a difference in your PFC and frontal parietal complex. And if you don't care about it, then it's just irrelevant and therefore not part of your experience. [01:17:13] Speaker C: Yes. [01:17:14] Speaker A: And the point is that they still need to say that as a theory. They still need to account for that. [01:17:19] Speaker C: It's also tricky because one thing that we did. So I'm telling this to you, Paul. I don't know if it's something that we are going to, because it's not something that we are reporting in the paper, but we also had some. We asked some questions to the participant at the end of the experiment and we did like debriefing, scenario debriefing. And we asked them whether they noticed that the stimuli were present at different durations from the participants that I tested. I remember most of them said yes, but then also if they say yes, what does it mean? Were they noticing? Experiencing it at the moment is something retrospective. So there seems to be something there. But yeah, as Alex said, it's something that needs more research. [01:18:00] Speaker D: Wait, so what, what are the implications of whether they said yes or no, whether they experienced it? [01:18:08] Speaker C: Just to put it simply, if they would have said no, I didn't notice anything, then the idea that that is not in prefrontal cortex, the duration information cortex, there is no offset ignition. [01:18:21] Speaker D: That's fine. [01:18:22] Speaker C: It's not part of the experience. [01:18:24] Speaker D: Right. But most people didn't. I don't know how you couldn't notice. Right. I mean, it's they're pretty different. [01:18:29] Speaker C: One second is not a big difference. [01:18:32] Speaker D: Yeah, maybe. Especially if you're in the scanner for a long time and you're tired and. Yeah, whatever. All right, Oscar, Inter aerial communication. And you've sort of. The hair has stood up on the back of your neck a few times. Talking about the analyses with this prediction give us the overview of what the two different theories predict and why this is tested and what was tested and how it all turned out. [01:18:57] Speaker C: Yeah, I think this is. So if we just remember what Rooney said about where in the brain the neural correlates of consciousness are located based on the two theories, then it's another prediction that comes is how does the information reach these areas? Now, let's say that we are seeing a face. We know that in the brain we have specific areas that are specialized to processing faces, like the fusiform phase area. So we know that when we recognize faces, that area is where the activity is happening. Now, we know that we are seeing a face because the fusiform face area is telling you that that is a face. But how does this information reach the neural correlates of consciousness? How does it go to prefrontal cortex for gnwt, and how does it reach the posterior hot zone for iit? And this is the prediction. It's just simply that they do is information reach. And this is also combined to what Alex said about the duration. Because if the information goes from fusiform phase area to prefrontal cortex, and this is linked to an ignition, then the information should reach prefrontal cortex during that time window, which is a very small time window, between 200 and 500 milliseconds, more or less. And for IIT, instead, because the experience and the duration is associated with sustained activity, also the interior connectivity should be. [01:20:28] Speaker D: Sustained, but it should still get there quickly, but then sustain. Right? [01:20:33] Speaker C: Yeah, yeah, that should be. That should remain sustained because for iit, we have this posterior network of interconnected areas that affect each other neurons, affects each other activity, and that should stay on until the experience is there. Okay, then we shouldn't see this pattern of activity of connectivity anymore. [01:20:55] Speaker D: All right, so the predictions are fairly simple, fairly straightforward. So then how do you test this? [01:21:00] Speaker C: That was the problem in our case. And so I was in charge of the MEG part of this prediction. And we used different methods to try to find this connectivity. So there are different problems that we encountered. One problem is that there isn't a textbook example of connectivity. So as I said before, we usually try to have some control analysis, but we Couldn't find a way to control for connectivity between different areas. [01:21:33] Speaker D: Do you mean like white matter axonal connection connectivity or functional connectivity? [01:21:39] Speaker C: More functional. Yeah. Talking about functional connectivity in this case. Yeah. Because connectivity that is depending on the task, on what is presented in front of you, in this case a face or an object. Because if you see an object, then fusiform phase area shouldn't be connected with PFC with the prefrontal cortex. If GNWT is right and the theories provide some initial guidance about what they intend in terms of connectivity, which is this gamma band. So this activity that is closely linked to spiking and that is measured in terms of coherence, which is a measure that also considers the phase of when the spikes and the oscillations happens. And we started testing this, the predictions using this method, which we call a phase based connectivity. [01:22:31] Speaker D: Let me just jump in and just clear. Sorry. And just clearly state. So you said gamma band. So this is oscillatory activity. And oscillations are. We don't need to go down the road of talking about whether they're causal or epiphytes phenomenal, but, but it is, it is the oscillating activity, the reverberatory activity thought to underlie sort of conglomeration of lots of different neuronal activities. So and then the gamma band is in, there are different, there are different frequencies that, that the oscillations can be at. And the gamma band is a fairly wide range of high frequencies, but it is, it does correlate with spiking activity, as Oscar already said. So that's why you focused on the gamma activity and then you looked at the different phase lockings. And phase just means like what part of the wave of the oscillation is happening, whether it's at the peak or the trough, and whether two different areas are in phase or out of phase, if they're both at the peak at the same time, or if they're both, or if one is at the peak and the other is at the trough. Sorry. These are all very simplified ways of doing. I'm just trying to give a clear picture. [01:23:44] Speaker C: So thanks for doing that, Paul. And yeah, so it's. Yeah, we try to use this method because that's what predictions were made. They were thought from the beginning. And the problem is that we didn't manage to find any sign of connectivity not in MEG and also not in intracranial eeg. Then we start using alternative methods which are not that commonly used, but they are also very, very good. And then These methods that we use are methods based on power. So instead of looking at this oscillation in terms of phase, we just look at the spectral activity, the oscillatory activity in terms of power. So let's say if there is a lot of gamma activity in one area at the same time, there should be a lot of GAN activity in the connected area. And when this goes down, it should go down in both areas. And we use some mutual information measures that are methods to look at this type of communication between different nodes. In that case, we couldn't find strong evidence in the case of iit, so we couldn't find strong evidence of connectivity between these posterior areas. Specifically, we look at early visual cortex and areas that are specialized for specific category of stimuli like face or object area specific areas. We find some evidence in intracranial eeg and we also use some similar methods in FMRI to also look at internal connectivity based on the BOLD signal. And we find some evidence in favor of connectivity between the specialized visual areas and prefrontal cortex. So to take on message in terms of connectivity here for the project is that we couldn't find evidence in favor of iit, so we challenged the prediction of iit, but we found some evidence, but not the evidence that was originally proposed, which was on face based connectivity for gnwt. So in terms of connectivity, both theories were challenged, but also more in general, the way we study connectivity in functionality was kind of challenged by our project, which isn't. [01:26:05] Speaker D: That's maybe one of the great benefits of doing a project like this. Right. Because you actually have determined, I don't know if it's better. Would you say it's a better way? You've made one way of analyzing the data maybe more established and one less established. Is that. Would that be a. Yeah, put it that way. [01:26:24] Speaker C: Yeah, yeah. And with the most established way, we couldn't find any evidence of connectivity, which is something that we are now trying to expand to other domains in neuroscience because it's something that we would like to see in general if it's a problem, because we tested theories of consciousness. But this is something that relates to every theory. Cognition. [01:26:51] Speaker D: Right. It's like a new analysis where you can go back now over decades of literature and say, well, maybe you just weren't using the right metric. So it could. Yeah, it can challenge a bunch of research. [01:27:02] Speaker C: Yeah, yeah. And the good thing about our project is also that we are sharing everything that we, that we did. So we're sharing the codes, the data, everything. And last year we organized With Ole Jensen and other people from Cogitate, a data competition challenge to a biomech conference about biomagnetism, about meg. And we asked the community to, okay, take our data and prove us wrong, tell us that there is some sort of connectivity there. And we are still trying to figure out what we can do, what we can provide to the field in terms of also generally neuroscience for connectivity. Like in this example. That can be helpful. [01:27:51] Speaker D: How many people. How many people or teams have taken that challenge up? [01:27:55] Speaker C: So in that challenge we had at the end, we finished with only five teams, and we made things very, very hard for them. Like, we asked them to do what we did in five years, and we only gave them five months. So it was quite a lot. But hopefully now when the paper will be out, we will release the data and people will have a lot of time to try to find connectivity in meg, FMRI and then intracranial eeg. [01:28:29] Speaker D: All right, so we've gone through the predictions and the results and maybe before we move on to some of the lessons that you guys have learned, I would love to hear just in your own words, not an interpretation, because I know you can't do that, but how you would describe the results from a high level. Right. In terms of the challenges to the various theories and the balance of those challenges, etc. Like, how would you describe that? Each of you at a high level, and I don't know, Roni, you haven't spoken in a little while. Maybe we could start with you or whomever wants to begin could begin if. [01:29:11] Speaker A: You guys want to begin. I'm fine with it. [01:29:13] Speaker B: Oh, go for it. [01:29:14] Speaker A: Alex, you look like you're. I think, yes, I like that you said challenges now because I think it's important. And like Oscar mentioned in the beginning, it's a very massive work, but it's part of the project. We have another experiment. So in terms of the strength of this experiment is, in my opinion, it's the disconfirmatory results, the. Those challenges. Exactly. And I think. I don't know if maybe it's because I'm a pessimist and like not a positive person in general, that I focus on the negative. But to me, this is the most interesting part of the. Of the results in each of the predictions. Like you said in the beginning, it didn't come out perfectly. So there are things to account for with respect to the orientation, with respect to the maintenance and the synchrony between the areas. So I look at it as an overall win in terms of both theories. Have Some homework to do with respect to how they specify the neurobiological implementation of conscious visual experience. That. That would be, in my own words, the conclusion. [01:30:33] Speaker D: Yeah, I like that. I wish you weren't so negative, but I do like that. No, I'm just kidding. But. But this is a thing where it just dawned on me that the. Let's say the proponents. The worry is that the proponents can. Can accept these challenges, but then we'll start looking for reasons why it's okay that it has been disconfirmed and then kind of move the goal posts when the goal posts. Part of this thing, part of the. This endeavor is to establish the goal posts. Right? But in the. On the theoretical side, you can be a little bit slippery if you want to, because they, like you said in the past, you do an experiment and you look for, like, confirmatory evidence or something, and then you can adjust your theory. There's still room for that. Here is a worry, right? [01:31:21] Speaker A: There's nothing wrong with adjusting your theory in light of evidence. I think that's like, you know, that's a. It's a fine line between that and the goal post thingy. I think. I think accepting this as evidence, accepting this as something that you need to adjust your theory by, is a very positive thing for the field. The problem starts if you don't accept it and you start like, saying, well, maybe, you know, maybe finding alternative explanations and some. And things like that. I think if we take it in the positive way of taking negative, the negatives would be this need to adjust, this need to account for these things for real and not try to find some. Some way to avoid this interpretation of the results, saying, okay, we did not find that. How do you deal? [01:32:10] Speaker D: All right, Oscar, Alex, how would you guys. [01:32:12] Speaker B: I think I'd very much agree with the perspective of Ronnie, which is like. I think as much as the initial goal of this project was like, oh, I think for me at least when I started being very naive, like, oh, yeah, we do that and then, you know, we have eliminated one theory and then, you know, top down, and then we move and we repeat until there's only one left standing. Right? But I think now, looking back, I realize this is the completely wrong idea. And it's more through the process of pushing the theories to make predictions that they wouldn't have done otherwise if they had just stayed within the realm of confirmatory science. We've highlighted a lot of things that they need to improve. And I think you could be pessimistic and say, well, this theory, they Made a wrong prediction and therefore I'm a propyrent. I think the theory is completely wrong and we just can't. We need to invent a completely new theory. I don't think that's a very valid approach. I think it's more trial and error and theories are working process and we've just revealed stuff that we thought were really innocuous in the sense that. Oh yeah, sure, if you present a stimulus for 1.5 seconds centering on the screen, of course the participants are going to have an experience that match that duration. And of course when you present visual stimuli there is bound to be some communication between low level visual area and higher level visual area. That's just the way it is. And then you actually do that and you realize well, nothing is as easy as it seems. And that means there's a lot of room for improvement, but also therefore room for discovery. [01:33:41] Speaker D: Right. [01:33:41] Speaker B: And then that's kind of pushing the field forward in specifying and getting better theories. [01:33:48] Speaker D: Oh, I was about to jump in and say and ask is one take home from this that both of these theories and likely all theories about consciousness because it's, you know, the brain is so complex consciousness and linking them. It's age old problem that both of these theories in this case are under specified in terms of the theoretical predictions? Yeah, I think that's fair to say. [01:34:15] Speaker B: And I don't think the theories would, you know, ever have argued otherwise. I think for example, Stanislas Dian, the main advocate of Geneva is always Adam and saying like, you know, that's just, that's a work in progress. It's by no means a done and dusted theory. Like we are continuously expanding it and building upon it. And the same can be said for GNW. I think. When did the IT 4.0 paper came out? I think was it 20, 23? [01:34:40] Speaker C: So yeah, I think working on it. [01:34:45] Speaker B: So that's part of the process of iterative cumulative science. And I think as much as the goal of eliminating theory, at least in our particular case is premature, that doesn't mean that you shouldn't engage in such project with the aim of arbitrating between two theories. Because that kind of pushes progress in a way that's faster and more efficient than if you leave them each on their own and just looking at their own really restricted set of experimental conditions that they love and have been doing for a really long time. [01:35:24] Speaker D: One reason I brought that issue up is because integrated information theory has been controversial because a group of scientists has called it pseudoscience and etc. And we're not going to talk about that. I will have Dean Buonamano on who's part of that group on a later episode, and we'll touch on that. But. And so this episode's not even about that controversy. However, the fact that both of these theories are under specified makes me think about philosophy of science, like what counts as a theory. Right. And because you have this broad range of how specific a theory is with its predictions, and, you know, in this case, both are under specified, not specific enough to have passed all falsification challenges. Right. And so. And that's why I said probably all theories at this point are underspecified because they are all works in progress as well. So I don't know if you want to comment on that, but yeah, if I may. [01:36:26] Speaker C: Yeah, yeah, I agree. But also, I wouldn't be too harsh with them because there are always new questions. And the role of the theory is try to answer these new questions. And being part of the project, I think both Stan and Julio were very, very brave because they put themselves and the theories on the spot. And we asked them new questions. There were new things that they have to do because of the project. [01:36:52] Speaker D: And that's Stanislas de Haan and Giulio Tononi, like the two of the main proponents of these two theories. [01:36:58] Speaker C: Yeah, yeah, yeah. And then so it's also an act of courage to challenge yourself and your theory, the theory that you build your career on. And they are like big scientists. They could just continue doing their own stuff. They didn't have to be part of this project if they didn't want. So I think it's good to test theories, it's good to test their predictions, and it's also good to give trust to science, because there is something that Lucia Meloni, one of the three main PIs in the project, always says that a lesson that she learned from Kahneman is that even if you do this adversarial collaboration, people rarely, if never, change their mind. Oh, right, right. I don't think the aim of this project was to change Stan or Julio's mind, because that's one project is not enough. But we are providing questions and I'm sure there will be many, many PhD students that will start new projects just from what we did. I hope that would be the case for what we did with Cogitate. [01:38:09] Speaker A: Just to touch up on that. Sorry, Paul, because like Oscar said, we shouldn't be so harsh. I think we are not harsh enough with respect to the philosophy of Science question and the field, because it's true that this. I agree with you with the interpretation that all of the theories are currently underspecified. Underspecified in again, if we look at the neurobiological implementation of what do they expect to happen in the brain? Of course it's not just gnw, it's not just iit. There are many other theories out there and I think it's fair to say that all of them are in the same boat. But I think that our point of view from Cogitate and the iterative process that Alex has been talking about, we are privileged to say that now in light of Cogitate, that wasn't the case. And it wasn't the case in the way it was discussed, discussed and discussed in the literature and in conferences and presented to grants for years and years. Both of these series and also many others, despite being highly under specified, didn't really present it as such. Right. We kind of think of GNAW as a global workspace and you say, oh of course, the frontal parietal network without going into the nitty gritty and understanding that there is a problem here. So I think this project is kind of revealed this uncomfortable place for all theories that okay, we are under specified. There are things that need to be to be done. And I think in terms of really testing the theory so called, it's not just a problem of under specification. Because even if you were more specified in the way that some of your hypotheses operationalized two predictions about the brain, you could say that that distance is so far away from what you started with, which is the core of your theory, that it's a question, does falsifying these results, can it even falsify what your theory is about? And we have philosophers of science discussing this, we have Niccolo Negro who wrote exactly about that of these like belts of core of the theory and the actual predictions that we end up with in the brain in that sense, I think also again, IIT and GNW and all the rest of the theories are on the same boat. Because when you start, no matter where you start from, from phenomenology, from the cognitive neuroscience of things, what you end up with in that prediction about the inter aerial communication is kind of far fetched from where you start from, no matter which theory you are. [01:40:48] Speaker D: Well, that's right. Because I mean even when I try to keep all these things aligned in my head, I have sort of a very abstract symbolic vision with global neuronal workspace because we use Terms like ignition and amplification. And it's so odd just thinking, all right, it's going to ignite in the prefrontal. And then it's so unscientific. Right. And it's this abstract notion that we then have to get somehow to the metrics. Just as you were saying. [01:41:14] Speaker B: Yeah, I guess. Yeah. I might say something. I'm not sure if that's controversial, but I guess in a sense it is completely true to say that they're under specified. And the very proof of it is you wouldn't need to do adversarial collaborations otherwise, where you actually have the advocates of each of the theories involved. [01:41:31] Speaker D: Right. [01:41:32] Speaker B: You could just. If you have. Each of them are a fully specified generative model, and then we have also really good models of the dynamics of the brain and so on. You could just go about, have your experiment and have two competing generative models that are fully specified in existing papers. You just compare them and that's it. You don't need the person who wrote the particular generic model to be involved in that specific project. So the fact that we need to involve the advocates just speak to the fact that they are underspecified. But at the same time, the fact that the theories are underspecified has no bearing on whether they are on the right track or not. So they can be underspecified right now, but they could still turn out to be correct in the future once they are fully specified or not. So that's like the fact that right now we don't have the full specification doesn't disqualify a theory from being relevant. Interesting. Vertical or whatever. And therefore engaging in a comparison of two theories even though they are under specified, is still a very valuable enterprise. [01:42:39] Speaker D: Yeah. This is also very different than making a prediction that light will bend around a star and then figuring out how to measure it, which is all very hard. Right. In relativity, but it's just way more complicated. It's way more difficult to get from this amorphous idea of ignition and amplification or integrated information to like, I. You can think like intuitively. All right, that sounds. That sounds plausible. To then specify anything is difficult. Yeah, it seems like. [01:43:12] Speaker B: Right. Yeah, but that is true. But I guess that's kind of the work for the field. Right. Because if you think about it the other way around, what's the alternative? I think the alternative is just going bottom up, just look at data, whatever that may mean. [01:43:27] Speaker D: Well, the alternative is panpsychism, actually. [01:43:29] Speaker B: Or is it. Right. [01:43:32] Speaker D: But IIT is kind Of Pansuckist. But no, let's not talk about Panpsuckist. [01:43:36] Speaker B: Yeah, that's Abort. [01:43:39] Speaker D: Abort. Sorry, I think I cut you off, Alex. [01:43:44] Speaker B: No, I was just saying I think the alternative of the theories are never going to be specified unless you actually build them in relationship to experimental data. I think that's kind of part of the process of building a theory of you state something, then you compare it to data, then you kind of work your way up. And if we went to do that, I'm not quite sure what else we could do with the theory. Just wait and. Yeah, I just think that's just the best thing we can do right now. And indeed it turns out to be fruitful because as Ronnie and Oscar said, that led up to we definitely learn new things and there's definitely a lot of follow up projects that can spin out of this. So from a scientific perspective, that's just progress. [01:44:27] Speaker D: Okay, so speaking of now on, I. We don't have all day. I don't want to take your whole day and stuff, but I wanted. Oscar, you mentioned earlier that you're excited about what is happening moving forward. So after this herculean effort, which now probably feels like it's in the rearview mirror to you guys because that's how publication works. And I know you've been working on other things. So what is happening moving forward? [01:44:50] Speaker C: Yeah, we have different things going on. One thing is the second experiment of Cogitate, which is an experiment in which we use a video game that was developed by a video game developer and Ronnie. In this video game we create conditions which participant either will see stimulus or not see a stimulus. So we manipulate the awareness of the stimuli and now we are working on this data and then things are going well, There will be news. [01:45:23] Speaker D: Are you, are you in that first like small cohort phase where you're trying to figure out how to do the analysis to pre register? [01:45:30] Speaker C: Now we are in a more advanced phase. We are writing the actual manuscript. [01:45:34] Speaker D: Oh, okay. [01:45:35] Speaker C: Yeah, it's. It's quite advanced. [01:45:37] Speaker D: But so, all right, you're starting the manuscript, so we'll have you back on in what, six years? [01:45:42] Speaker C: Yeah, I would say that that's. Yeah, that's exactly. But also very, very interesting for me is the data release because when the paper will be out, we will release the data, all the data, the MEG data, the intracannial data, the FMRI data, also the behavioral data. So we have so much data to share and we will release data with a lot of information, details, a wiki, so as much information as we can. [01:46:11] Speaker D: I always wonder about this, like a data release like this and open science in general. Like I'm busy as fuck at my work and I would be interested in this data. I don't have time. Right. So like who will this intrigue enough? Like who's going to take up the task of analyzing the data? [01:46:31] Speaker C: Yeah, I think if you imagine the time that will take someone to collect this amount of data. [01:46:37] Speaker D: Yeah. [01:46:38] Speaker C: Then it will be more convenient to just use already collected data. And also the number of participants that we collected is a huge cohort. [01:46:46] Speaker D: But if I was going to like write a, Could I write a grant, you think? And my mission on the grant is to use. Use this data for a particular analysis. I suppose I could write a grant, why not? So that could create time. [01:46:59] Speaker A: Also, there's been massive effort within the project to actually make this data as simple to use as possible. Yeah. And that has been incredible separ effort. And as Oscar mentioned, it includes like detailed documentation. It includes storing it in very specific ways so it can be easily like, you know, generalizable. We also release all of our codes and pipelines so you can pre process it the way we did and stuff like that. So there has been a massive effort to actually make it as simple as possible to use our data. Not only just, you know, to dump it and share it and say there you go. [01:47:35] Speaker D: Yeah, yeah. So you've architectured it a lot. [01:47:39] Speaker C: Yeah. And we're sharing also the code that we use in the project so people can use our code and replicate with instruction how to run the codes. So we're trying to make this as easy as possible for other people. And also it's something that was there since the beginning of the project. So we spent a lot of time on it. And everything we did was keeping in mind that one day we're going to release everything. [01:48:00] Speaker D: That's great, right? [01:48:01] Speaker B: And I guess, yeah, the question, I think it's. I would hope that it's a relevant data set for many people because it's unique, you know, like if there's so many different questions such as, oh, I don't know, how does, you know, FMR activity correlates with MEG signal and IEG signal. You could just do a very technical paper about that. And yeah, there's many different projects you can do and yeah, I'm sure a lot of people would be interested. And we've heard also from several people and the machine learning community to benchmark classifiers and all that. So there's I think quite A people. [01:48:34] Speaker A: Approach us and already come out like with questions, okay, when, when. Because they have plans like they're basically written the paper and just waiting for our data to be, to be available. [01:48:43] Speaker C: Yeah, yeah. And also something that it will be interesting also because some of the predictions that we tested are also prediction that could align with, could be, could be proposed by other theories of consciousness. [01:48:56] Speaker D: Yeah. [01:48:57] Speaker C: And one critique of our project is why you only tested these two theories of consciousness. But the time is limited. [01:49:04] Speaker D: And time is limited. But you also stated at the beginning other theories dropped out because it wasn't maybe relevant or appropriate. The experiments, et cetera. [01:49:13] Speaker A: Exactly. Because of the need to commit beforehand to these predictions is different than taking this data now and analyzing it in light of our results and come up with new predictions that would be of course useful, but it's much harder to commit for it beforehand. Right. [01:49:31] Speaker D: Briefly, and then I'll have one final question. What is it like working under, for lack of a better term, controversial issues like this Integrated Information theory is controversial on its own. But all theories of consciousness have a measure of controversy. What is it like, do you feel like pressure working under, under that controversy or are you sort of walled off from that pressure because. Because of the pre registration and you're just the researchers carrying out the research and you don't have a dog in that fight, et cetera. [01:50:10] Speaker B: But. [01:50:10] Speaker D: Or do you feel the pressure? [01:50:12] Speaker B: I don't mind going further. I guess for me, I think like for me there was definitely a lot of pressure, but it wasn't the controversy. It was more like, you know, being a cog in such a massive project and messing up something would have so much implication, not just for me, but for many other people and really qualified and really also people who really need this thing to work and so on. [01:50:31] Speaker D: It's responsibility. It's a lot of responsibility. [01:50:34] Speaker B: And for the controversy, I mean there is, I think that consciousness is a very important field with a lot of ethical implications. That is something that it's not related, really related to the project itself, but not my interest in consciousness. I think that's something I do spend time wondering about, you know, in my free time. Like, you know, is this, you know, what are the implications of whatever we are going to find and so on. But for the controversies related to the theories, that's something that often happens in conferences. Like, you know, as we discussed before, like why this specific prediction from the theory? I'm like, well, I can give you the boring answer which is I don't have to care, right. Like I Didn't come up with it. I don't have to defend it. I just have to test it and see if that works out. And you know, that's. I'm just going to try to do the best I can with that and therefore I have a clear conscious. [01:51:19] Speaker D: But you still have to deal deal with reviewers. Right? And the practical aspects of working under that are. I mean, I don't know how much you personally, the three of you deal with the reviewers because you're part of such a large team and maybe that takes some of the pressure off too. But yeah, I think it's not just the reviewers. [01:51:39] Speaker B: Oh, sorry, sure. I guess I'll just say I think like the. Because the RPIs, like especially the DPS took a lot of on their back to kind of like shoulder the brunt of the difficulty with the reviews. But I guess that's also something that was probably all on our mind, which is like what happens if we spend so much time and effort and then because of controversies, things don't come out as good as they could have. That would have been really very disappointing to put in mildly. But we were lucky that in the end it all turned out okay. [01:52:11] Speaker A: I think controversy is part of that though. It's part of the project even getting this much attention and even getting like to where it got even beginning. Right. So. So I think we cannot kind of ignore where it came from. It came from controversy. It existed in a field. Heated debates. People take things very personally. I think you can say it's because like Alex said, it has like real life implications. And of course we see that now not only with disorders of consciousness, but also like AI and stuff like that. But it's also been there like even before these discussions. And I think it's interesting how much people in the field care. And I think it's because consciousness, it's so trivial, it's so immediate. Why are we even continuing to fight about it and have 20 different theories at the same time? Like this is sort of kind of the mind boggling thing, right? How come we all know what we mean but we have 20 different theories that say different things, that have different predictions. [01:53:09] Speaker D: It's about legacy, but. [01:53:13] Speaker A: So I think it's true that the center PIs have done an incredible work of trying to separate the controversy from the work itself. But like in our, you know, when we go to conferences and also when we discuss with the adversaries themselves, like you feel the controversy in the sense of feeling that this is very high stakes project. It's not just A massive collaboration. It's a massive collaboration with massive implications because of those controversies. So it does put pressure, like Alex said. But I think that it's good to see that you can work, you can still work under those controversies, that it doesn't bend you over in any direction that you can do this meaningful work. And that's also. Yes. When you have incredible PIs, like one of these projects that sort of steer you in the middle of this very stormy ocean through that to the science within the controversy. [01:54:12] Speaker C: Yeah. Let me say something also about the incredible first authors. The PI did a great job, but I think also the fact that we have this big team, and I think it's also related kind of to the period which we started the project, which was during the lockdown. And like, we were. Each of us were in his own house and we were interacting with people mainly through computers. And so the fact that Romy is in Israel, Alex in Germany, and I'm in England, it didn't make any difference. So we were spending a lot of time together and we were only exposed to the controversy towards the end of the project. And we were already a good group of collaborators and friends at that point. [01:55:00] Speaker D: That's actually what was going to be my last question to you. So maybe you can speak to this as you finish that point is we discussed offline before that you guys have. I think you'd use the term like self organization of how you kind of came together and also have become friends through this. So, yeah, do speak to that before. Before we're done. [01:55:22] Speaker C: Yeah, yeah. So, yeah, as I was saying, so we spent a lot of time together online when we couldn't spend time with other people because it was lockdown. And I think we are very. I don't know, we're good people generally, and we liked each other and we like to work with each other. And when the controversy started, for instance, when we saw the letter. [01:55:51] Speaker D: Oh, that first pseudoscience letter thing. [01:55:53] Speaker C: Yeah. When that came out, our reaction was the reaction of a group who was supporting each other. And in that moment, we were also supporting the people who were mostly exposed to this controversy, to these critics, which was the main PIs of the project. And I think it would have been different if I was working on this project on my own with just another PI. So it would be the way which I would have coped with critics and the attention, the drama on the social media would be different if I didn't have this group tackling this with me. [01:56:33] Speaker D: Yeah. [01:56:33] Speaker B: And I guess, like, also to the point of like, related to the controversy as well. I remember very distinctively when I was like a bit, you know, rocky times because of that, that we had a few meetings and there was never a point where people were like, all right, I'm not interested in this project anymore. Like, there was always the attitude all throughout was just like, all right, so what do we do now? Like, everybody was always on board with continuing, just doing more and figuring it out altogether rather than just saying also, you know, we started a project and as Oscar said, Covid hit and nobody was. I think there was never a point where people get super demotivated and not willing to fight. I think we just always, we always stick to it and just continue through the end, no matter what kind of thing. To an extent that I don't think would also be the case in every big collaboration. [01:57:25] Speaker D: I would say to the end. We're at the end. Huh? [01:57:28] Speaker B: No, no, no. Well, we will continue collaborating until. I don't know what. [01:57:35] Speaker D: Till the end. End. Yeah, yeah. [01:57:37] Speaker C: So that I can advertise us as people who can test whatever theory of neuroscience you have. Bring it to us and we will destroy it. [01:57:49] Speaker D: Oh man, you don't know what you just asked for. Well, congratulations on this Herculean effort and the results of this Herculean effort. Thank you for coming on my podcast to share this and I'll link to it in the show notes because there's lots to for people to go over and hopefully we communicated a little bit about the results and people can take home what they will from it. But anyway, congratulations and thanks for coming on to talk about it. [01:58:20] Speaker C: Thank you. [01:58:20] Speaker A: Thank you so much, Paul. [01:58:27] Speaker D: Brain Inspired is powered by the Transmitter, an online publication that aims to deliver useful information, insights and tools to build bridges across neuroscience and advance research. Visit thetransmitter.org to explore the latest neuroscience news and perspectives written by journalists and scientists. If you value Brain Inspired, support it through Patreon to access full length episodes, join our Discord community and even influence who I invite to the podcast. Go to BrainInspired Co to learn more. The music you're hearing is Little Wing, performed by Kyle Donovan. Thank you for your support. See you next time. [01:59:23] Speaker A: Sa.