cover stefano ferraina

Stefano Ferraina on transitive inference and prefrontal cortex

  • cover play_arrow

    PLAY EPISODE


cover stefano ferraina
Season 2014
Season 2014
Description arrow_drop_down

Description

Can monkeys reason logically , and if so, what does that look like at the level of single neurons? Neurophysiologist Stefano Ferraina presents evidence that prefrontal cortex neurons encode both symbolic distance and serial position during transitive inference, suggesting a neural substrate for logical reasoning in non-human primates. Subscribe for more from the Convergent Science Network podcast series. Stefano Ferraina joins Paul Verschure and Tony Prescott at the BCBT summer school to discuss his research on transitive inference in macaque monkeys. The task requires animals to learn an ordered sequence of abstract visual symbols through pairwise comparisons, then infer the correct ranking of novel, never-before-matched pairs. Surprisingly, monkeys master this within weeks and show a robust symbolic distance effect: comparing symbols far apart in the sequence is easier and faster than comparing adjacent ones, mirroring findings in human numerical cognition. The discussion carefully examines whether this performance reflects genuine logical reasoning or simpler reward-association mechanisms. Ferraina describes a critical control experiment using two separate chains that are subsequently linked, demonstrating that monkeys maintain the transitive ordering even when reward history alone cannot explain their choices. Recording from prefrontal cortex, he finds that roughly half of task-related neurons encode the symbolic distance effect, about 40 percent encode serial position, and a subset of around 20 percent encodes both , suggesting that the same neural population supports multiple aspects of the relational structure. Key topics include how transitive inference is defined and tested in non-human primates, why the symbolic distance effect challenges pure reward-association explanations, what the serial position effect reveals about how symbols are organized along a mental continuum, how the two-chain linking experiment strengthens the case for reasoning over association, the limitations of single-neuron electrophysiology for establishing causality, and what the overlap between symbolic distance and serial position coding in prefrontal neurons implies about the neural architecture of logical inference. Part of the Convergent Science Network podcast series from the BCBT Summer School.

Tagged as:

About the author call_made

CSN Podcasts

Both the triumphs of humanity and its most evil deeds have resulted from collaboration. In a time where humanity is required to aspire to the former and minimize the latter, the question arises of how collaboration arises and why it fails. Surprisingly, this phenomenon, so central to who we are, is not well understood. Hence, a collaborative effort is required to understand collaboration in its full biological, psychological, sociological, cultural, and economic complexity and to translate this understanding into operational impact. This series of podcasts is one step toward achieving these complementary goals. The Collaboration Podcast presents interviews with people who are central orchestrators of collaboration in various domains including business, government, science, art, health, sustainability, and the military. The discussions were conducted by Prof. Dr. Paul F.M.J. Verschure and members of the Program Advisory Committee of the Ernst Strungmann Forum on Collaboration (https://www.esforum.de/forums/ESF32_Collaboration.html) during 2021 and had the goal to sketch a map of opportunities, challenges, and obstacles in human collaboration. The forum took place in May 2022, and now we would like to share this series of interviews with a broader audience. The full report of the Forum will be published in 2023 by MIT Press. The podcast was produced by the Convergent Science Network (https://www.convergentsciencenetwork.org/). Context: The stability of social systems depends critically on realizing sustainable methods of “collaboration,” yet how and by which means collaboration is achieved is not clearly understood; neither are the conditions or processes that lead to its breakdown or failure. Collaboration can be understood as cooperation between agents toward mutually constructed goals. Part of the reason for our lack of understanding is that the phenomenon of collaboration is, by nature, a highly multidisciplinary problem, and effective research into its complexities has been difficult to achieve across the broad range of scientific and technical disciplines involved. The need for a fundamental understanding of collaboration, however, has become increasingly important. Not only does humankind demand answers as it attempts to address critical challenges at multiple scales (e.g., climate change, migration, enhanced automation, social and economic inequality), but ever-increasing technological and economic means of interconnecting people and societies are disrupting long-established, familiar patterns of how we interact. Radical technological changes that are ongoing have the potential to reshape collaboration in ways that are currently hard to predict or influence (e.g., by altering configurations in interaction, information creation, and modes of communication). On one hand, such changes could disrupt hitherto stable forms of collaboration by affecting critical communication channels and traditional roles, as can be observed in the rapidly changing patterns in governance, commerce, and social interaction. Conversely, technology could lead to the emergence of novel, successful forms of collaboration that deviate from traditional “hierarchical” architectures. Evidence of this can be seen in areas as diverse as highly automated manufacturing plants, the open science movement, collaborative software repositories, user-centered services, and the sharing of economy-based modes of organization. Without a fundamental understanding of the mechanisms, processes, and boundary conditions of collaboration, it is not possible to evaluate or predict which of these possible scenarios are sustainable or even plausible. The Forum “How Collaboration Arises and Why it Fails” (May 8–13, 2022, Location: Frankfurt am Main, Germany) Chairs: Andreas Roepstorff and Paul Verschure Program Advisory Committee: Jenna Bednar, Julia R. Lupp, Bhavani R. Rao , Andreas Roepstorff, Ferdinand von Siemens, and Paul Verschure

More posts

Timestamp

  • fast_forward00:00:03 - This is the Convergent Science Network podcast.
  • fast_forward00:00:08 - Leading researchers in the domain of neuroscience, brain theory and technology
  • fast_forward00:00:12 - are interviewed by Paul Verschure and Tony Prescott. Okay.
  • fast_forward00:00:18 - This is Paul Verschure with the Convergent Science Network podcast.
  • fast_forward00:00:22 - And I'm here with Stefano Ferreira, who was one of our speakers in our summer school today.
  • fast_forward00:00:28 - And Stefano you talked about logical neurons for logical reasoning what what does it exactly mean,
  • fast_forward00:00:36 - yeah to me it's a
  • fast_forward00:00:39 - first evidence I mean and at least in the neurophysiological literature on non-human
  • fast_forward00:00:47 - primates that neurons in the brain are able to code for a logical property that is the
  • fast_forward00:00:57 - one that is related to the transitive inference task.
  • fast_forward00:01:02 - Okay, so we're talking monkeys here, primates, right?
  • fast_forward00:01:07 - And so what you're trying to assess is, okay, what's the logical capability
  • fast_forward00:01:13 - of the monkey as an organism and how much that can we recover in a neural response?
  • fast_forward00:01:18 - So what's exactly the task that you were studying?
  • fast_forward00:01:22 - Yeah, the task is named as a transitive inference task has been used in different
  • fast_forward00:01:28 - forms in many types of animals.
  • fast_forward00:01:32 - In a few words, it's the ability to conclude that A is higher than C after learning
  • fast_forward00:01:41 - that A is higher than B and B higher than C.
  • fast_forward00:01:46 - So to conclude that the first one is necessary to have a kind of relationship
  • fast_forward00:01:53 - between all the values presented And then to be able to argument,
  • fast_forward00:01:59 - to obtain more information than those provided at the beginning.
  • fast_forward00:02:04 - Right, so it's like you have a sequence of magnitudes, let's say,
  • fast_forward00:02:08 - and you learn to make inferences about these relationships.
  • fast_forward00:02:12 - Correct. And you also drew a parallel to the social life of certain kinds of primates.
  • fast_forward00:02:19 - Was it just to introduce the topic, or do you really see this as reflecting
  • fast_forward00:02:23 - a need that they have in the wild?
  • fast_forward00:02:28 - No, it was a way to introduce, but it's also true that in the other animals
  • fast_forward00:02:32 - where the property of transit inference has been shown, this is always stronger,
  • fast_forward00:02:39 - I mean, more evident in animals that are used to live in such organized groups.
  • fast_forward00:02:45 - Okay, so there might be a correlation. There's one study that compared two types of birds.
  • fast_forward00:02:54 - Very similar in the species, but since one of the two lives normally in the
  • fast_forward00:03:02 - large group while the second one is as a group very reduced,
  • fast_forward00:03:08 - the evidence in this study was that the transitive inference properties was
  • fast_forward00:03:15 - more evident in the all-in-one living in large organized groups. Okay.
  • fast_forward00:03:23 - Okay. So now the specific task that the monkey was exposed to was essentially
  • fast_forward00:03:29 - using different kinds of shapes, fairly abstract shapes.
  • fast_forward00:03:34 - And you would present pairs of shapes. And then one of them would lead to reward
  • fast_forward00:03:39 - in the training phase, right? It would give reward. The second one would not give reward.
  • fast_forward00:03:43 - And then in the next pair, you would take, let's say, the second pattern that
  • fast_forward00:03:49 - didn't get reward, you would present it with a third pattern.
  • fast_forward00:03:51 - And then the second pattern would get reward and the third one wouldn't get
  • fast_forward00:03:55 - reward etc and that's how you go down the chain so we have a certain specific sequence,
  • fast_forward00:04:00 - and then in the test phase you present the animals now with randomly selected
  • fast_forward00:04:06 - pairs of the whole sequence that they were trained on including those trained
  • fast_forward00:04:11 - exactly and then they have to choose the one that would give them more reward than the other,
  • fast_forward00:04:16 - that's the whole task so.
  • fast_forward00:04:20 - So what do you really observe when monkeys perform this task?
  • fast_forward00:04:25 - How good are monkeys at this task?
  • fast_forward00:04:28 - Actually, surprisingly, the monkey were very good on performing,
  • fast_forward00:04:32 - actually on learning the general idea of the task and acquiring the strategy.
  • fast_forward00:04:38 - After less than three weeks of initial training, meaning both monkeys we used
  • fast_forward00:04:45 - were able to play with figures.
  • fast_forward00:04:50 - To recognize by trial and error that there is an imposed series that the experiment,
  • fast_forward00:04:59 - presented to them, and then to use this information also at the moment of the test when,
  • fast_forward00:05:08 - novel figures, never matched figures during training, are presented,
  • fast_forward00:05:13 - like concluding in a sequence from A to F who is the,
  • fast_forward00:05:19 - rewarded one between B and E.
  • fast_forward00:05:23 - But now, in the training, what kind of, if we talk about a difference in reward magnitude,
  • fast_forward00:05:30 - the amount of juice you get, if you go to the stimulus, the pattern with the
  • fast_forward00:05:35 - highest reward and lowest reward, what's the difference in reward that they
  • fast_forward00:05:39 - got, what kind of magnitude?
  • fast_forward00:05:40 - We never introduced this variable to the task that is actually interesting and
  • fast_forward00:05:46 - could be explored in the future, at least this is our hope.
  • fast_forward00:05:52 - But to say differently, the amount of reward was always the same.
  • fast_forward00:05:58 - For each pair, there was a rewarded symbol and a rewarded one.
  • fast_forward00:06:04 - So there was never a symbol with a strong amount of reward and the other one with less.
  • fast_forward00:06:14 - So the comparison was always relative within each pair.
  • fast_forward00:06:20 - And the series was created because the different pairs were presented consequently.
  • fast_forward00:06:27 - Right, exactly. Exactly. So that would mean in your test phase you cannot combine
  • fast_forward00:06:30 - all possible patterns because then you would have two patterns that in that
  • fast_forward00:06:36 - sequence would never have been rewarded or would always have been rewarded.
  • fast_forward00:06:39 - No, in the test all the comparisons are compared. I mean all the possible pairs are compared.
  • fast_forward00:06:46 - Okay. So that would mean you might present patterns that in the combination
  • fast_forward00:06:51 - would never have been rewarded in any way.
  • fast_forward00:06:54 - Well, that's not true.
  • fast_forward00:06:58 - No, it's not true because the only one figure on a sequence of six using letter is E.
  • fast_forward00:07:09 - F is F. The F is never rewarded during the learning. That's right. Exactly.
  • fast_forward00:07:16 - While A is always rewarded. So, one easy conclusion is that it's very simple
  • fast_forward00:07:25 - for the monkey to conclude when A and E are present in the test,
  • fast_forward00:07:34 - but our data show that it's not true, it's not just a simple relationship between
  • fast_forward00:07:40 - a real world history and logical assumptions.
  • fast_forward00:07:46 - And what's the question again? No, my question is just about the test,
  • fast_forward00:07:51 - whether there's not a bias in the test.
  • fast_forward00:07:52 - Because like you say now, stimulus F is never rewarded.
  • fast_forward00:07:56 - So in some sense, there are probe trials where you have pairs that include stimulus
  • fast_forward00:08:01 - F, which would be much easier to evaluate
  • fast_forward00:08:05 - than if you would have any other combination of stimuli from A to E.
  • fast_forward00:08:10 - Yeah, but still having E and F that are adjacent in the pair presentation during learning is sometimes,
  • fast_forward00:08:21 - because the performances are probabilistic, sometimes more difficult than comparing B and F.
  • fast_forward00:08:28 - Even if F is always there and F was never there.
  • fast_forward00:08:33 - That's a surprising effect that you found, right? If you now do these test trials
  • fast_forward00:08:38 - and we present combinations of stimuli, if you would have A and B...
  • fast_forward00:08:46 - In terms of reaction time, the monkey might take longer to make a decision than
  • fast_forward00:08:51 - if you present A or A and E.
  • fast_forward00:08:53 - So how do you explain that? This is right.
  • fast_forward00:08:57 - One was suspecting that those figures more presented during learning are the ones more easily coded.
  • fast_forward00:09:09 - And this is not what emerged from the behavioral data.
  • fast_forward00:09:12 - The behavioral data shows that comparing figures that are more far located in
  • fast_forward00:09:20 - the series is absolutely easier for the animal.
  • fast_forward00:09:26 - This is quite a surprise. According to some people that presented the model
  • fast_forward00:09:32 - in the literature, this is probably due to the noisy representation of each symbol in the series.
  • fast_forward00:09:39 - And then comparing two very close symbols that are noisy is much more complicated
  • fast_forward00:09:45 - because of the noise when that comparing two symbols that are far located.
  • fast_forward00:09:50 - But could it not be an interference effect of multiple memory systems?
  • fast_forward00:09:54 - It's a certain way. It's an interference.
  • fast_forward00:09:55 - I think interference and noise is the same because noisy means that the symbol
  • fast_forward00:10:02 - is not very well coded when you compare the code to the very close to them.
  • fast_forward00:10:09 - To the other symbol very close to them in the series. So if there is no noise,
  • fast_forward00:10:13 - it means that they are very well classified.
  • fast_forward00:10:17 - So if they are not very well classified, there is interference.
  • fast_forward00:10:21 - Sure, but don't you agree that noise is a bit of a dissatisfactory explanation
  • fast_forward00:10:25 - of variability in nature?
  • fast_forward00:10:27 - Because basically we're not actually explaining anything. We just say,
  • fast_forward00:10:30 - well, there's some uncontrollable variability here.
  • fast_forward00:10:32 - Well, you could also argue that maybe the strange thing is If I'm being trained on a certain pair AB,
  • fast_forward00:10:40 - right, and I'm being tested on ANC or ANB, and I compare my reaction time or
  • fast_forward00:10:47 - my accuracy, it's worse than ANB.
  • fast_forward00:10:49 - So I could argue, well, maybe since I'm instantaneously confronted with ANB
  • fast_forward00:10:54 - in the training set, it's also a pattern association mechanism that comes into
  • fast_forward00:10:58 - play. There's a perceptual mechanism that comes into play.
  • fast_forward00:11:00 - It says, well, maybe ANB is, let's see, one category.
  • fast_forward00:11:04 - Okay. And now I get two competing players.
  • fast_forward00:11:08 - Memory-dependent processes, one telling you something about,
  • fast_forward00:11:12 - let's say, the magnitude order, the order of presentation, and the interference
  • fast_forward00:11:17 - comes from a process that not all these things belong together, let's say.
  • fast_forward00:11:20 - So that you really have interference between specific cognitive processes as
  • fast_forward00:11:24 - opposed to a nonspecific noise with a decision-making stage.
  • fast_forward00:11:28 - I agree. I actually believe that there are different levels of representation.
  • fast_forward00:11:33 - One is that the single symbols are represented in the brain.
  • fast_forward00:11:38 - The other one is that the pair starts to be represented.
  • fast_forward00:11:42 - Since the figure B is present in both A, B, and B, C,
  • fast_forward00:11:47 - and in the two cases the probability of reward is 50%, then the representation
  • fast_forward00:11:54 - in the pair need to be, because it's much more convenient, more noisy than the
  • fast_forward00:11:59 - representation limitation on the single item.
  • fast_forward00:12:01 - So at the time when the monkey is required to answer which one is linked to the reward.
  • fast_forward00:12:11 - Of the never presented, the never experienced pair, for the monkey it's very
  • fast_forward00:12:15 - easy because there was no code, no previous code assigned to them.
  • fast_forward00:12:19 - So there's no ambiguity because it's not part of the A, B, B, C conflict, it's a B, D.
  • fast_forward00:12:27 - So, for the monkey the job is easier, but this is my point to try to figure out what's,
  • fast_forward00:12:37 - the category of the pair based on the, even in the conflicting experience,
  • fast_forward00:12:46 - but based on previous experience.
  • fast_forward00:12:48 - Because one interpretation could also be you might
  • fast_forward00:12:51 - interpret a single stimulus presentation is let's
  • fast_forward00:12:54 - say an episode a memory episode right in which everything comes
  • fast_forward00:12:57 - together it's okay i saw these two patterns together and now
  • fast_forward00:13:00 - i see some other patterns together so then those episodes
  • fast_forward00:13:04 - might be then chained together in the sequence right so and that comparisons
  • fast_forward00:13:09 - between episodes and within episodes are actually very different processes that
  • fast_forward00:13:14 - that's essentially what you're what you're seeing yeah okay but then you made
  • fast_forward00:13:19 - you made also That's an important point in terms of,
  • fast_forward00:13:22 - because for you, this data is pointing to an ability for logical operations.
  • fast_forward00:13:28 - Now, for that, you must exclude interpretation that would say,
  • fast_forward00:13:31 - well, this is just reward.
  • fast_forward00:13:33 - Dependent on the reward that you receive, you form certain associations between
  • fast_forward00:13:37 - events, and it's just a very direct recall of these associations.
  • fast_forward00:13:41 - So how can you exclude this interpretation?
  • fast_forward00:13:46 - The approach that we use that is referred to other similar approaches in the
  • fast_forward00:13:54 - literature is that we ask the monkey to learn two different chains.
  • fast_forward00:14:03 - And in this case, the history or reward is distributed differently in the two chains.
  • fast_forward00:14:11 - And then at the end of the learning of the two separate chains,
  • fast_forward00:14:15 - we asked the monkey to link the two chains by providing them the relative value of the extremes.
  • fast_forward00:14:23 - That means C in the ABC chain and D in the DEF chain.
  • fast_forward00:14:30 - And the evidence shows that the monkey is still expressing behaviorally the same,
  • fast_forward00:14:42 - performance than when the chain is unique.
  • fast_forward00:14:46 - Even if the reward is not any more uniformly distributed to explain the relative
  • fast_forward00:14:55 - value only based on these attributes.
  • fast_forward00:15:00 - So we conclude that probably the monkey is not only basing their conclusion
  • fast_forward00:15:07 - on the reward, but it's also using the information provided previously differently.
  • fast_forward00:15:13 - And in this kind of chaining test, so let's say we have two sequences,
  • fast_forward00:15:19 - like you say, ABC and with DEF, And now suddenly I present the monkey with B
  • fast_forward00:15:25 - and E as a pair, right, to evaluate.
  • fast_forward00:15:29 - Does the monkey immediately respond to that correctly or they have to learn
  • fast_forward00:15:35 - that this is now a relevant question?
  • fast_forward00:15:38 - Unfortunately, I have no answer to this question because we haven't looked to
  • fast_forward00:15:42 - this data, this level of detail.
  • fast_forward00:15:45 - Detail, but I know from all of our data that during the task,
  • fast_forward00:15:50 - the monkey continue to learn, but this is normal. Of course.
  • fast_forward00:15:55 - So I don't know if in the chain experiment, there is already the evidence of
  • fast_forward00:16:02 - the link between the two.
  • fast_forward00:16:05 - Right, exactly. So your point is, look, that they can do this,
  • fast_forward00:16:08 - that they perform well in the chain experiment experiment would
  • fast_forward00:16:12 - require some form of reasoning let's say deliberation which
  • fast_forward00:16:15 - said well no i had to have this sequence abc and ah
  • fast_forward00:16:18 - this was the other sequence df now e was later than b in in my in the learning
  • fast_forward00:16:25 - in the learning phase so now i know i have to choose for b right but i you could
  • fast_forward00:16:31 - also argue that maybe again this is a perceptual learning process because now
  • fast_forward00:16:35 - uh the first time i I see B and E,
  • fast_forward00:16:38 - so the two exemplars from the two different sequences,
  • fast_forward00:16:42 - I don't know what to do, let's say, but they're presented together,
  • fast_forward00:16:46 - they're now conjunctive, and dependent on my choice, I get reward or no reward.
  • fast_forward00:16:52 - So I could learn a new contingency.
  • fast_forward00:16:54 - So how could you exclude that interpretation? Yeah.
  • fast_forward00:16:58 - Again, I'm not able to exclude without looking at the first trials of the test.
  • fast_forward00:17:06 - But I bet also that you do agree that this is no simple sensory motor transformation
  • fast_forward00:17:13 - of the symbol in the response.
  • fast_forward00:17:16 - There is something that is more than associating symbols to response.
  • fast_forward00:17:23 - So when the overall effect of the symbolic distance effect,
  • fast_forward00:17:29 - that is the performance that changes with the distance between the symbols that
  • fast_forward00:17:34 - are supposed to be located in the mental space differently,
  • fast_forward00:17:38 - and this behavioral effect is still there after linking two different chains,
  • fast_forward00:17:45 - I think it's a good argument in favor of. It's not a conclusion.
  • fast_forward00:17:49 - No, that's indeed interesting, right? So we have these alternative interpretations,
  • fast_forward00:17:54 - and we can come back to that later.
  • fast_forward00:17:56 - But indeed, what you subsequently showed is that the two main phenomena you
  • fast_forward00:18:03 - then looked at was what you called symbolic distance and serial distance,
  • fast_forward00:18:07 - and you compared the two.
  • fast_forward00:18:08 - So what's the difference? What's a symbolic distance and what is serial distance?
  • fast_forward00:18:12 - How are they different? Yeah, the symbolic distance is defined as the effect
  • fast_forward00:18:16 - in the behavioral measure, that is the measures that are the reaction time and
  • fast_forward00:18:22 - the performance, that is essentially better.
  • fast_forward00:18:26 - I mean, the performance increase and the reaction time is reduced when the symbols
  • fast_forward00:18:31 - to be compared are more distant each other.
  • fast_forward00:18:34 - So, comparing B and E is easier than comparing the learned one B and C pair.
  • fast_forward00:18:43 - So this is the symbolic distance effect.
  • fast_forward00:18:47 - It's a very strong effect that has been previously described in humans too,
  • fast_forward00:18:51 - and it's common to other quantity comparison like in the numerical literature.
  • fast_forward00:18:58 - So again, in the numerical literature, when a subject is asked to compare between
  • fast_forward00:19:05 - a number that is five as a reference in the first 10 and six or five and eight,
  • fast_forward00:19:14 - normal subjects have more difficulty on concluding the six is higher than five
  • fast_forward00:19:21 - than eight is higher than five.
  • fast_forward00:19:25 - The second effect is the serial position effect. That is the evidence that in the learned series,
  • fast_forward00:19:35 - the different symbols are not represented equally, but they are some way organized around a center.
  • fast_forward00:19:47 - This is in our case the letter C that stays in the middle of the series.
  • fast_forward00:19:52 - But before we go to that point,
  • fast_forward00:19:56 - if we look at symbolic distance, and we now take the case that we're linking
  • fast_forward00:20:07 - two sequences together, right? Right.
  • fast_forward00:20:10 - Now, the only sticking point I have is that if I have a three-element sequence, ABC,
  • fast_forward00:20:20 - that the combination C with anything from the second sequence is now an inconsistent
  • fast_forward00:20:25 - pairing compared to everything else.
  • fast_forward00:20:27 - Because C was never rewarded. Correct.
  • fast_forward00:20:30 - So how is that exception handled in terms of the performance of the monkey?
  • fast_forward00:20:40 - Because it's like an exception that they must handle, right? Yeah.
  • fast_forward00:20:45 - It's true that you are referring to local phenomena, while unfortunately we
  • fast_forward00:20:52 - analyze our behavioral data at the global level. Okay, okay, all clear.
  • fast_forward00:20:56 - But I think it's important to remember that what is one of the strategies that
  • fast_forward00:21:06 - I suppose our animals were able to learn,
  • fast_forward00:21:08 - that there is no absolute value of every symbol presented.
  • fast_forward00:21:13 - It's the same at the time when B is presented for the first time with A,
  • fast_forward00:21:21 - B is never reinforced in the block design of the learning, but then right after,
  • fast_forward00:21:28 - B is presented together to C, and now it's B reinforced.
  • fast_forward00:21:31 - So, for the message that is provided to the animal is that there's a warning
  • fast_forward00:21:38 - that the value of the symbol in terms of probability of getting a reward could be changed shortly.
  • fast_forward00:21:48 - So this is probably the same that happens with C because the monkey already
  • fast_forward00:21:52 - know that this is possible and the two chain experiment,
  • fast_forward00:21:56 - even if the consolidation and the training was very long, longer than the A-B
  • fast_forward00:22:02 - training of a single series.
  • fast_forward00:22:06 - The monkey knows that the C is possible for the C to be categorized differently.
  • fast_forward00:22:13 - And this is what happens with the linking pair. At a certain point,
  • fast_forward00:22:18 - I'm telling to the monkeys, C is also a winner only when presented with D. Right, exactly.
  • fast_forward00:22:26 - Okay. Okay, so then you're saying in the test phase itself, this might then
  • fast_forward00:22:31 - mitigate this idea that it's an exception, that the monkey is just processing
  • fast_forward00:22:35 - it as any other sequence because it's, well, look, I might not get reward now, but maybe in the future.
  • fast_forward00:22:40 - This is what I believe, yeah. Okay, all right. So then, for symbolic distance,
  • fast_forward00:22:46 - it's a term used also in the psychological literature, but in your task,
  • fast_forward00:22:52 - is it reasonable to really speak of a symbol as you're using these patterns?
  • fast_forward00:22:57 - And the monkey must also really react to the patterns as they're localized in space, right?
  • fast_forward00:23:02 - They must really touch the screen at a certain X, Y position.
  • fast_forward00:23:05 - And you could argue, well, a symbol might be a difficult concept to define.
  • fast_forward00:23:10 - But the one thing is that it is an internal representation that is independent
  • fast_forward00:23:14 - of the actual sensory state.
  • fast_forward00:23:19 - I think that the symbols are very large.
  • fast_forward00:23:24 - They don't need to be precisely recognized, I mean, and not even the position
  • fast_forward00:23:31 - on the touch screen need to be precisely coded.
  • fast_forward00:23:34 - For the monkey, the response is left and right, it's the same of a key press.
  • fast_forward00:23:40 - So, what I mean more is that, is it fair to call the pattern that you use? You project a pattern.
  • fast_forward00:23:49 - Yeah, but the monkey, if you remember the video, the monkey is looking for the figure.
  • fast_forward00:23:53 - It's not stay still on the pattern.
  • fast_forward00:23:58 - I mean, it doesn't look at the pair. It looks for this figure.
  • fast_forward00:24:03 - Right. And then there is another behavioral data that we haven't discussed. It is the eye movement.
  • fast_forward00:24:10 - We know that
  • fast_forward00:24:13 - if the first figure that is
  • fast_forward00:24:16 - 4V8 is the wrong one then he moves the
  • fast_forward00:24:19 - eye to the next one and then decide
  • fast_forward00:24:22 - to move the arm so there are other aspects of
  • fast_forward00:24:25 - the behavior that need to be further explored okay what
  • fast_forward00:24:29 - I'm after here maybe it's a completely irrelevant semantic concern I'm expressing
  • fast_forward00:24:35 - but you want to measure something called symbolic distance but what you're using
  • fast_forward00:24:41 - is also a spatially organized task with visual patterns that in themselves cannot be considered symbols.
  • fast_forward00:24:47 - So what makes it symbolic?
  • fast_forward00:24:52 - Okay, I could agree that the term symbolic is not proper but for me it's the
  • fast_forward00:24:59 - way to refer to something that is not in the literature.
  • fast_forward00:25:03 - Okay. And I agree with you that probably a symbol is never created in this kind
  • fast_forward00:25:09 - of strategy that we are imposing to the animal.
  • fast_forward00:25:12 - Because just to be complete, often in the animal study and even in human study,
  • fast_forward00:25:20 - the symbol used before testing are always the same in a way that the subject
  • fast_forward00:25:26 - creates a symbolic representation of the symbol. Exactly.
  • fast_forward00:25:29 - And actually for some of the people that discuss a lot on the behavioral relevance
  • fast_forward00:25:34 - of transitive inference, this is a necessary step.
  • fast_forward00:25:39 - Right, okay. But in our case, we want to avoid that the monkey was just using a well-coded algorithm,
  • fast_forward00:25:48 - symbol to conclude on the comparison between two symbols.
  • fast_forward00:25:57 - I think this is just maybe not the right one solution, but it's a way to reduce
  • fast_forward00:26:03 - our... No, no, look, I completely get it, sure.
  • fast_forward00:26:06 - The number of factors that are there, because remember that then we try to correlate
  • fast_forward00:26:13 - this behavioral data with a neural code.
  • fast_forward00:26:17 - And if we were very too many factors at the same time.
  • fast_forward00:26:22 - Of course. No moreover it might so as an operational definition of a symbol
  • fast_forward00:26:30 - I think it's completely defendable but we just have to be clear about what we're looking at.
  • fast_forward00:26:34 - So I agree with you that the symbolic distance in this case is not properly.
  • fast_forward00:26:39 - But now what you found is that if you look at the performance of the animals,
  • fast_forward00:26:46 - so you look at the errors they make and their reaction times,
  • fast_forward00:26:49 - then you see that they perform better on patterns or symbols,
  • fast_forward00:26:55 - as we now can call them, that are in the middle of the sequence as opposed to
  • fast_forward00:27:00 - those that are at the beginning or the end of a sequence, right?
  • fast_forward00:27:05 - No, the performance is better for the anchor points, even for the one that includes
  • fast_forward00:27:15 - the never enforce it. Okay.
  • fast_forward00:27:18 - So the beginning and the end. The beginning and the end. So how big is that modulation?
  • fast_forward00:27:22 - In terms of difference in percentage of performance, it's I think twice.
  • fast_forward00:27:29 - So it's a big difference. It's a big difference in the two animals that we test. Sure.
  • fast_forward00:27:35 - But how do you explain that or what does that mean?
  • fast_forward00:27:38 - I think that this is a partial evidence that the different symbols are organized,
  • fast_forward00:27:46 - relatively to each other around the two anchor points that are the A and F of the sequence,
  • fast_forward00:27:54 - and maybe also the central one that's the letter C.
  • fast_forward00:27:59 - And at least in the human literature, This is a partial evidence that a line,
  • fast_forward00:28:05 - a mental line is used because sometimes people refer that to solve the task,
  • fast_forward00:28:10 - they do explore mentally the line by moving from the extremes toward the center
  • fast_forward00:28:16 - to find the item that is required to be compared.
  • fast_forward00:28:21 - So if you interpret this as arranging now these symbols on the line,
  • fast_forward00:28:27 - that would basically mean that okay i'm exposed
  • fast_forward00:28:30 - to the task i know now what the beginning and
  • fast_forward00:28:33 - the end of my line is and then now
  • fast_forward00:28:36 - it's like i build a chain right now i can stick the elements in
  • fast_forward00:28:39 - that on that line is that correct that would be the
  • fast_forward00:28:42 - idea yes but that's what it was interesting of course that the two anchor points
  • fast_forward00:28:46 - are in their quality radically different because that this is a surprise it's
  • fast_forward00:28:52 - still we have a very similar behavior that means that the the behavior is It's
  • fast_forward00:28:57 - not just explained by the reward. That's exactly right.
  • fast_forward00:29:01 - So do you see this as a form of latent learning for the endpoint?
  • fast_forward00:29:04 - So this is learning without an explicit instruction?
  • fast_forward00:29:08 - Could be. Yeah. It could be a possibility. Yeah.
  • fast_forward00:29:12 - Could it also be a violation? I named this a strategy because I think that the
  • fast_forward00:29:17 - monkey needs to have a solution because every day he's introduced in the lab.
  • fast_forward00:29:23 - He doesn't know which figure will be used. use it but the
  • fast_forward00:29:26 - monkey knows that you have to use this new figure to
  • fast_forward00:29:30 - figure out uh how they
  • fast_forward00:29:33 - are located relatively each other right so the it could be that just the monkey
  • fast_forward00:29:40 - normally I mean spontaneously adapt uh to this uh allocation in this in the
  • fast_forward00:29:49 - mental space of of the symbols presented. Right, exactly.
  • fast_forward00:29:54 - Without any control. But now, could you argue that this is like a segmentation
  • fast_forward00:30:01 - of the task? Would you accept that?
  • fast_forward00:30:06 - Or the task domain?
  • fast_forward00:30:10 - Yeah, let's say differently. I have no argument to conclude that it's not segmented.
  • fast_forward00:30:16 - So I think that it's in favor of a segmentation. Right.
  • fast_forward00:30:20 - Because what is interesting then is that sometimes the monkey is saying,
  • fast_forward00:30:25 - okay, I have a problem to solve. Yeah.
  • fast_forward00:30:28 - These experimenters only show me always a subset of the task I must solve.
  • fast_forward00:30:33 - I have to figure it out myself.
  • fast_forward00:30:36 - So I have to learn the boundaries. I have to figure out what are the boundaries
  • fast_forward00:30:38 - of my task, right? Where do the relationships start and where do they stop?
  • fast_forward00:30:43 - And in the performance, you see that actually a monkey is able to do that and
  • fast_forward00:30:46 - also to draw very strict boundaries around that.
  • fast_forward00:30:51 - But the strange thing is that this almost looks like a form of perceptual learning
  • fast_forward00:30:57 - in the sense that it is an acquisition of knowledge about the task without explicit
  • fast_forward00:31:03 - instruction because the beginning gets reward and the end gets nothing, right?
  • fast_forward00:31:08 - So do you see that as a separate kind of learning process that imposes boundaries
  • fast_forward00:31:15 - on this local structuring of the task or do you see it as an integrated process?
  • fast_forward00:31:21 - I believe that both are present because the assumption that A is always reinforced And F is never,
  • fast_forward00:31:34 - it's not easy to explain why A and B is so difficult and A and F is so difficult.
  • fast_forward00:31:41 - So I think that there are two interfering process that works together.
  • fast_forward00:31:49 - One is the tentative of the monkey to generalize, to organize the item according
  • fast_forward00:31:55 - to the general rule that has been introduced by the experiment.
  • fast_forward00:31:58 - The second one is that there is a conflict on the representation that is created. Okay.
  • fast_forward00:32:04 - And then, so now for this symbolic distance, which is the magnitude relationships
  • fast_forward00:32:09 - that we are being trained on, like our binary relationship, then we have the serial order effect.
  • fast_forward00:32:16 - Yeah. Right? How is the serial order effect or the serial distance effect different
  • fast_forward00:32:20 - from the symbolic distance effect?
  • fast_forward00:32:21 - The serial order is the phenomenon that we just described, that is the fact that the.
  • fast_forward00:32:31 - Learned pair are not coded similarly with the U shape described,
  • fast_forward00:32:40 - while the symbolic Symbolic distance is the relationship with the distance of the symbol compared,
  • fast_forward00:32:50 - in particular for never-matched symbols during the learning,
  • fast_forward00:32:54 - so that could be named novel pair.
  • fast_forward00:33:03 - So, with that in mind, you started to look at the neural substrate. Yeah.
  • fast_forward00:33:08 - And you record it in premotor cortex? It's prefrontal cortex this time.
  • fast_forward00:33:15 - Prefrontal? I present today the data from prefrontal cortex.
  • fast_forward00:33:20 - We do also have data from premotor.
  • fast_forward00:33:22 - The reason is that in the literature, that is mostly in the human literature,
  • fast_forward00:33:29 - using imaging data, the evidence is that some of the area of the cortex,
  • fast_forward00:33:37 - essentially the prefrontal cortex, the premotor cortex and the posterior parietal
  • fast_forward00:33:43 - cortex are involved to the transitive inference task.
  • fast_forward00:33:49 - You know that the functional MRI doesn't provide the details to conclude what is the mechanism.
  • fast_forward00:33:54 - So we only know that there are areas that are statistically activated while
  • fast_forward00:34:00 - the subjects are tested in a similar task to the one that I use in the animal.
  • fast_forward00:34:11 - So what did you observe recording in the frontal cortex?
  • fast_forward00:34:15 - The observation that I think is very straightforward is that we have been able to find neurons.
  • fast_forward00:34:25 - That are modulated in a way similar, another way is to say they are correlated
  • fast_forward00:34:34 - to the behavioral effect that our animals are able to show.
  • fast_forward00:34:40 - So there are neurons modulated for the symbolic distance, other neurons are
  • fast_forward00:34:47 - modulated for the serial position,
  • fast_forward00:34:49 - and a subpopulation of them, when analyzed at a single neuron, are modulated by both.
  • fast_forward00:35:00 - I specified single neuron level because when the average activity of all the neurons task-related,
  • fast_forward00:35:11 - are further investigated, then what emerges is that the population of neurons,
  • fast_forward00:35:19 - which is about half of the total population we recorded with our movable electrode
  • fast_forward00:35:24 - in the prefrontal cortex, is able to code both for the symbolic distance and
  • fast_forward00:35:31 - the serial position effect.
  • fast_forward00:35:33 - That is quite important because the two phenomena could be separated in the
  • fast_forward00:35:38 - brain as used differently,
  • fast_forward00:35:40 - by whatever code the brain is using to solve the task,
  • fast_forward00:35:47 - while being able to demonstrate that the same population is able to use both
  • fast_forward00:35:54 - signals means that probably the one signal is dependent on the other. Okay.
  • fast_forward00:36:00 - But now, so the numbers you listed were about 50% responsive for serial position.
  • fast_forward00:36:08 - 40% for symbolic distance, and then there were a few neurons across these two
  • fast_forward00:36:13 - pools that were responsive to both.
  • fast_forward00:36:15 - Yeah. So like 30% or 20% was something? It's 20% of both.
  • fast_forward00:36:20 - So, but now how do we know that this actually scales up? because what you really
  • fast_forward00:36:26 - measured is you present a monkey now with a pair of patterns,
  • fast_forward00:36:29 - let's say B and C or A and B.
  • fast_forward00:36:34 - You record a neuron and now you basically sort all your trials.
  • fast_forward00:36:38 - And so, okay, let's put all the trials together where I presented C as my first pattern to the left.
  • fast_forward00:36:44 - And now I'm going to look at, okay, how do my neurons respond?
  • fast_forward00:36:47 - And you will find the sensitivity of neurons to, let's say, C in the first position,
  • fast_forward00:36:52 - right? And then you might find another neuron that says, okay,
  • fast_forward00:36:55 - I like B in the second position, etc.
  • fast_forward00:37:00 - But that basically at this point only tells us that we are representing in some
  • fast_forward00:37:06 - form the discrete element or the elements of these sequences.
  • fast_forward00:37:10 - So how do we know that this is the neural substrate of the inference we want to look at? Okay.
  • fast_forward00:37:19 - One of the limits of the approach is that neurophysiology in non-human primates
  • fast_forward00:37:26 - with the method of electrophysiology is unable to conclude about causality.
  • fast_forward00:37:33 - So, the only possibility for a study exploring whatever behavioral control in
  • fast_forward00:37:41 - whatever in whatever experimental setup is to find a relationship between,
  • fast_forward00:37:49 - controlled variable and the measured variable. So it's a relationship.
  • fast_forward00:37:54 - So in our case, we found a strong relationship between a behavioral phenomenon,
  • fast_forward00:37:59 - two behavioral phenomena, that is the symbolic distance and the serial position effect.
  • fast_forward00:38:09 - Being unable to find such a relation will allow us to conclude that the neurons code,
  • fast_forward00:38:17 - code or the firing rate of neurons in prefrontal cortex is unable to participate
  • fast_forward00:38:24 - properly to the behavioral counterpart of the transitive inference task.
  • fast_forward00:38:33 - But we are unable, again, for the limitation of the method, to conclude that
  • fast_forward00:38:39 - the prefrontal cortex is necessary and that the task is controlled selectively by that area.
  • fast_forward00:38:48 - That actually is something that I don't believe and other people don't believe
  • fast_forward00:38:51 - too in the area because we know that patients with the lesion at the level of
  • fast_forward00:38:57 - the hippocampus or the prefrontal cortex or the parietal cortex has a different deficit,
  • fast_forward00:39:03 - but always with the deficit that,
  • fast_forward00:39:09 - imply that this subject are unable to solve a task like the transitive inference one.
  • fast_forward00:39:17 - Okay. But now tell me if I look at the serial position versus symbolic distance,
  • fast_forward00:39:24 - how is the neural response to the symbolic distance different from the serial position response?
  • fast_forward00:39:32 - So if I would give you a neuron, a response profile of a neuron and I say well
  • fast_forward00:39:37 - we measure this neuron while the monkey is performing this task would you be
  • fast_forward00:39:40 - able just by looking at this response profile say whether it does the serial
  • fast_forward00:39:45 - position or the symbolic relation?
  • fast_forward00:39:48 - I think that what the analysis at the level of the population tell us is that
  • fast_forward00:39:53 - every neuron is able to provide a contribution to the two factors but maybe
  • fast_forward00:40:03 - be sometimes too noisy to reach statistical significance.
  • fast_forward00:40:06 - So the first numbers that I provided is that with a strong statistical support
  • fast_forward00:40:14 - there are neurons that are well coded for the,
  • fast_forward00:40:19 - symbolic distance or they are well coded for the serial position because this
  • fast_forward00:40:25 - doesn't need to have only one response different but also to follow the model.
  • fast_forward00:40:30 - So you need to follow the model of the behavior And because the requirements
  • fast_forward00:40:35 - are too many for a single neuron, it could be too much.
  • fast_forward00:40:39 - So the answer is that there are neurons that are.
  • fast_forward00:40:44 - More easy to it's easier to find neurons in
  • fast_forward00:40:48 - prefrontal cortex that remember is an area that often
  • fast_forward00:40:51 - doesn't provide in the discharge rate
  • fast_forward00:40:54 - code so much contribution to say
  • fast_forward00:40:57 - differently neurons in prefrontal cortex are not the same of neurons in motor
  • fast_forward00:41:02 - premotor or sensory areas the amount of discharge of this nuance is very low
  • fast_forward00:41:07 - often so this means that the signal is a It's a noisy signal by definition because
  • fast_forward00:41:13 - if you discharge rate is 10 hertz,
  • fast_forward00:41:16 - 10 spike per second, then the possibility to replicate trial by trial,
  • fast_forward00:41:20 - the same code is very difficult than when your neurons discharge 80, 100 hertz.
  • fast_forward00:41:27 - But it was surprising that some of the neurons you reported are at a relatively elevated firing rate.
  • fast_forward00:41:33 - Like you reported neurons up to 50 hertz, which seems rather than exceptional for that area.
  • fast_forward00:41:40 - It's true, but I still consider up to 50 Hz a low rate. Okay.
  • fast_forward00:41:47 - All right. So then, so okay, here we have- Anyway, in the delayed epoch is where
  • fast_forward00:41:55 - these neurons are more able to express themselves.
  • fast_forward00:41:58 - Generally speaking, the delay epoch is where the prefrontal cortex has been
  • fast_forward00:42:02 - often studied for one of the characteristics that is assigned to the prefrontal
  • fast_forward00:42:07 - cortex, which is the working memory.
  • fast_forward00:42:09 - Right. So the working memory is often very well able to activate this news and the delay activity too.
  • fast_forward00:42:17 - So now we have our transitive inference task, and you show monkeys can do it.
  • fast_forward00:42:25 - Now, you have also taken this as a method
  • fast_forward00:42:29 - if you want to look at certain neuropathologies, like schizophrenia.
  • fast_forward00:42:34 - So, why is this relevant if we try to understand schizophrenia?
  • fast_forward00:42:40 - Let's say that this is just the beginning of a story that we started recently.
  • fast_forward00:42:45 - The reason is our
  • fast_forward00:42:48 - goal is to link pharmacology and the electrophysiology with different methods
  • fast_forward00:42:55 - in particular the idea of this study that I present today is that it's well
  • fast_forward00:43:03 - known that schizophrenic patients have deficit,
  • fast_forward00:43:08 - in solving the transitive inference task.
  • fast_forward00:43:14 - Since there is a drug that is ketamine that has been used often in animal model to provide,
  • fast_forward00:43:23 - one of the possible model of
  • fast_forward00:43:29 - schizophrenia, that is the one related to the dopamine use in the brain.
  • fast_forward00:43:35 - We tested if ketamine as a receptor agonist of the NMDA, that is one of the
  • fast_forward00:43:44 - receptor of dopamine in the brain,
  • fast_forward00:43:47 - it's able to modulate the behavioral response of the animal that we trained
  • fast_forward00:43:56 - in the transitive inference.
  • fast_forward00:43:57 - And the figure that comes out is that the answer is yes, the ketamine is able
  • fast_forward00:44:04 - to reduce the efficiency of this animal to solve the task.
  • fast_forward00:44:08 - What is interesting is that the performance is particularly reduced more in
  • fast_forward00:44:16 - the figures that are considered transitive.
  • fast_forward00:44:20 - That means those that are about comparison between never experience and the Falshermore,
  • fast_forward00:44:30 - those that does not include the extreme spares that others refer as an anchor point.
  • fast_forward00:44:41 - So, but what you see in the monkey was a degradation of performance?
  • fast_forward00:44:46 - Yeah, generally it was quite similar for all the comparison,
  • fast_forward00:44:52 - but we partially explored the data and the result is that the statistical significance
  • fast_forward00:45:03 - is particular for the comparison that are more related to the transitive inference capability.
  • fast_forward00:45:11 - Okay but then that's also exactly what was observed in schizophrenic patients
  • fast_forward00:45:17 - so schizophrenic patients has no problem on concluding which one is has more value of A and B,
  • fast_forward00:45:24 - B and C on A and F including the anchor point or B and F because once again
  • fast_forward00:45:30 - it doesn't include the anchor point while when the question is which is higher between B and E E,
  • fast_forward00:45:40 - then the deficit emerges,
  • fast_forward00:45:43 - and the same happens in our monkey under sub-anesthetic doses of ketamine.
  • fast_forward00:45:51 - But now, how do you explain that in terms of the action of ketamine?
  • fast_forward00:45:56 - Okay, I think it's one of the possibilities, once again, to believe that the
  • fast_forward00:46:02 - way of the representation is on the brain at the level of the neurons, it's very noisy.
  • fast_forward00:46:10 - And that the ketamine reducing the efficiency of dopamine is increasing the
  • fast_forward00:46:17 - noise and then making more difficult to the comparison.
  • fast_forward00:46:20 - But now schizophrenia is also affecting, let's say, experience,
  • fast_forward00:46:26 - subjective experience, right?
  • fast_forward00:46:27 - So it has a very specific phenomenology as well.
  • fast_forward00:46:30 - So do you believe that the same holds for the monkey?
  • fast_forward00:46:35 - So in other words, is this task actually also informing us about the role of
  • fast_forward00:46:40 - consciousness in decision-making, or you see it really as disconnected from those phenomena?
  • fast_forward00:46:44 - I think it's very difficult to conclude such a...
  • fast_forward00:46:49 - No, but I'm just... I mean, we have no evidence that the monkey has...
  • fast_forward00:46:54 - We know that the ketamine is a dissociative drug.
  • fast_forward00:46:57 - Yeah, exactly. But I don't know how much our dosage was able to really produce
  • fast_forward00:47:06 - a monkey dissociated from the reality.
  • fast_forward00:47:08 - Because we want to have the monkey still able to perform the task.
  • fast_forward00:47:14 - Actually, the monkey was still performing the task with a good percentage of success.
  • fast_forward00:47:20 - But the deficit was selective for a portion of the task, not for the whole task.
  • fast_forward00:47:25 - So I don't know if this could be used as a very easy, I think, result,
  • fast_forward00:47:33 - might be too simple to argument that maybe the monkey was consciously unable
  • fast_forward00:47:41 - to participate to the transitive inference.
  • fast_forward00:47:45 - That is a conclusion. No, that would not be my conclusion.
  • fast_forward00:47:48 - My conclusion would be more like, if you use this paradigm as a way to also
  • fast_forward00:47:52 - investigate schizophrenia.
  • fast_forward00:47:54 - Is it then not by implication the case that we are implicitly also probing conscious
  • fast_forward00:48:00 - states and conscious operations in this monkey, and not only logical ones?
  • fast_forward00:48:05 - Yeah, it needs to be explored. Of course. No, it's clear. Yeah, yeah.
  • fast_forward00:48:10 - Since the dragon is able to modulate consciousness, it's by definition a possibility.
  • fast_forward00:48:16 - Yeah, exactly. I agree with you.
  • fast_forward00:48:19 - So So, basically the conclusion then is that, okay, you also on top of that showed,
  • fast_forward00:48:26 - which was very interesting, that if you now in this task go to a next level
  • fast_forward00:48:33 - of description of the physiology of the monkey brain, which is the local field potential,
  • fast_forward00:48:39 - that you also find correlates there of this transitive inference.
  • fast_forward00:48:46 - So, what did you observe there exactly? Okay, for us the analysis of local field
  • fast_forward00:48:53 - was a direct consequence of the observation in humans that fMRI is modulated by humans.
  • fast_forward00:49:02 - By the transitive comparison in a transitive inference task.
  • fast_forward00:49:08 - And since it's commonly accepted that the most close,
  • fast_forward00:49:15 - neurophysiological signal to the bold activity of fMRI is the local field,
  • fast_forward00:49:21 - for us it was very important to look also to the local field modulation during the task.
  • fast_forward00:49:27 - So it probably was not a surprise anymore more after finding that the single
  • fast_forward00:49:33 - neuron firing rate was modulated to find that also that the local field was modulated.
  • fast_forward00:49:39 - But it's also important to show evidence of this because some way put the, again,
  • fast_forward00:49:52 - the level of the multiscale, that means local computation,
  • fast_forward00:49:55 - remote computation that need need to be further studied in the future because
  • fast_forward00:50:01 - local field has access to the synaptic integration,
  • fast_forward00:50:04 - so we don't know from where the signals came, and to local reverberation from the level of the gamma.
  • fast_forward00:50:12 - Even if we haven't been able to show a strong effect at the level of the gamma in the local field,
  • fast_forward00:50:19 - but at the level of the low components that are those more related to the input
  • fast_forward00:50:24 - from From the thalamus or from the other cortical area like the hippocampus,
  • fast_forward00:50:29 - once again, or the posterior parietal region.
  • fast_forward00:50:31 - But you seemed to show that the deflections in this local field potential correlated
  • fast_forward00:50:38 - very specifically with properties of the task.
  • fast_forward00:50:41 - Absolutely. Actually, we have been able to find a very strong relationship more
  • fast_forward00:50:46 - with the local field than with a single unit. Mm-hmm.
  • fast_forward00:50:50 - And the reason could be different. One is that the single neuron expressed their
  • fast_forward00:50:59 - contribution with different tuning forms.
  • fast_forward00:51:02 - Sometimes with a shape very similar to the performance, like in the serial position, a U-shape.
  • fast_forward00:51:10 - Other times it was inverted.
  • fast_forward00:51:13 - And as I said before, Therefore, the noisy level of the single-neuron was sometimes
  • fast_forward00:51:20 - too much to reach significance.
  • fast_forward00:51:23 - And also the time in the task from the pair presentation to the decision time,
  • fast_forward00:51:32 - that is after the go signal during a delay,
  • fast_forward00:51:37 - for the single-neuron was never, I'll say, often not the same.
  • fast_forward00:51:43 - I mean that the single neurons are able to express their contribution at the
  • fast_forward00:51:47 - beginning of the delay, at the middle part of the delay, or the late portion of the delay.
  • fast_forward00:51:51 - While for the local field, it was very time-located to the pair presentation
  • fast_forward00:51:57 - at about 400 milliseconds.
  • fast_forward00:51:59 - And this allowed to extract probably a better signal to find a correlation as the one we found.
  • fast_forward00:52:07 - But it's interesting to see that you don't have such a tight correlation with
  • fast_forward00:52:11 - your single-cell recording. So would that suggest that there's another dynamic
  • fast_forward00:52:15 - process at work that shapes this local field potential?
  • fast_forward00:52:19 - Like other synaptic activities that you don't pick up with your single cell
  • fast_forward00:52:25 - measurements that shape the local
  • fast_forward00:52:27 - field potential? Yeah, the single unit activity is really local, okay?
  • fast_forward00:52:34 - While the local field, by definition, is able to sample at least a volume that
  • fast_forward00:52:41 - is huge compared to a single unit.
  • fast_forward00:52:45 - So, for sure, the conclusion is that we are not looking at the same computational level.
  • fast_forward00:52:52 - Right. and actually the better performance of the single unit coding is when
  • fast_forward00:52:59 - the neurons are considered together forming a population.
  • fast_forward00:53:04 - Local field is by definition a mesoscopic approach so this is probably expected
  • fast_forward00:53:12 - that if a signal is there it's much easier to be found and maybe correlated with behavior so So,
  • fast_forward00:53:20 - the experiments you described basically suggest that the monkey that we investigate,
  • fast_forward00:53:26 - and possibly also humans, organize the information that they need in this transitive
  • fast_forward00:53:32 - inference task on a line of magnitude, of magnitude relations in some sense.
  • fast_forward00:53:38 - And this line is very well demarcated at the beginning and the end. Mm-hmm. Okay?
  • fast_forward00:53:43 - Correct. So, how long can those sequences be, you think, for the monkey? Okay.
  • fast_forward00:53:50 - For our monkey, I think they need to vanish soon.
  • fast_forward00:53:56 - However, I haven't shown this data today, but we asked the monkey some time
  • fast_forward00:54:03 - to play with the series that was learned the day before, and the series is still there.
  • fast_forward00:54:10 - Even if the monkey knows that it's better
  • fast_forward00:54:13 - for him to have a very good performance to forget completely what was the assignment
  • fast_forward00:54:28 - of figures the day before because they are never used.
  • fast_forward00:54:31 - Right, exactly. But once again, there is something that's implicit probably.
  • fast_forward00:54:38 - There's something there. I'm not thinking to use them, but if I'm asked to use them, I'm ready to.
  • fast_forward00:54:45 - The performance was very rapid and perfect the day after.
  • fast_forward00:54:52 - We have no tested other days, I mean two days or three days,
  • fast_forward00:54:57 - so I don't know how long it remains. It's pretty astonishing, yeah.
  • fast_forward00:55:02 - But then the other thing is, so there's this famous, or the snark effect that
  • fast_forward00:55:05 - Stan De Haan has written a lot about and others,
  • fast_forward00:55:10 - which shows that also humans, when they have to deal with magnitude judgments,
  • fast_forward00:55:16 - let's say something is bigger or smaller than something else and so on,
  • fast_forward00:55:19 - that there's a very specific spatial bias.
  • fast_forward00:55:21 - So if you have to respond with your left hand, you respond more quickly to small,
  • fast_forward00:55:27 - lower magnitudes, and with the right hand, you're faster. when the magnitude is higher.
  • fast_forward00:55:32 - Also suggesting that there is some spatial organization of magnitude judgments.
  • fast_forward00:55:37 - And what I found interesting about that is if we compare that to your results,
  • fast_forward00:55:42 - you interpret your results as also the title of your talk suggested as the ability
  • fast_forward00:55:48 - to perform logical operations.
  • fast_forward00:55:51 - But in some sense, I could also argue, well, look, maybe this is all just spatially organized.
  • fast_forward00:55:56 - And I use associative rules, I use analog reasoning, if you want,
  • fast_forward00:56:01 - to then make these magnitude judgments.
  • fast_forward00:56:03 - So, is it really necessary to interpret your results in terms of logical operations?
  • fast_forward00:56:09 - It's just the definition of logic. Maybe you as a psychologist,
  • fast_forward00:56:14 - if I'm not wrong, by training your definition of logic is much more stronger
  • fast_forward00:56:22 - than the one that I'm using at the moment.
  • fast_forward00:56:27 - And I agree that the monkey is using a space as a model to solve a task.
  • fast_forward00:56:39 - If the solution of this task is logic it's a term of definition and in my point
  • fast_forward00:56:46 - it's logic whatever is more than what was available by by,
  • fast_forward00:56:55 - the information provided at the beginning so it's not a level of logic that,
  • fast_forward00:57:03 - is probably in your mind but it's a first level of be able to,
  • fast_forward00:57:10 - to conclude if then right
  • fast_forward00:57:13 - it's very elementary but remember
  • fast_forward00:57:16 - that these are non-human primates yes and and as an animal we don't know if
  • fast_forward00:57:23 - they are able to we have no argument to conclude that they are in particular
  • fast_forward00:57:27 - for non-verbal conclusion able to to have a to use arguments properly
  • fast_forward00:57:37 - to obtain new information.
  • fast_forward00:57:39 - But it also means that maybe what you're showing us here is an alternative way
  • fast_forward00:57:43 - how we can think about cognitive operations that give rise to behavior regularities
  • fast_forward00:57:49 - that we as observers can interpret in logical terms,
  • fast_forward00:57:52 - but it actually follow very different principles.
  • fast_forward00:57:56 - And the same might hold for our own cognitive operations, right?
  • fast_forward00:58:00 - That we also then, let's say now, a posteriori, describe them as logical operations.
  • fast_forward00:58:08 - But actually, in terms of the internal operations, it's much more an analog,
  • fast_forward00:58:12 - spatially organized process.
  • fast_forward00:58:14 - I mean, would you consider that still as an open option? I agree with this.
  • fast_forward00:58:18 - Actually, it's a behavioral experiment that we perform in humans,
  • fast_forward00:58:24 - but I think it's important to describe.
  • fast_forward00:58:26 - At this point, at the beginning of our testing with animals,
  • fast_forward00:58:31 - we asked normal subjects at the student's level of age to perform a transitive
  • fast_forward00:58:39 - inference task while changing their gates.
  • fast_forward00:58:42 - So, looking at the left or to the right. So, for the visual attributes of the
  • fast_forward00:58:49 - symbols and the primary goal of the subject was not related at all with the gates.
  • fast_forward00:58:55 - But still the gauge is able to interfere as ketamine.
  • fast_forward00:59:01 - With the subject performance okay so
  • fast_forward00:59:04 - looking left and right in a whatever it's like changing a reference frame for
  • fast_forward00:59:12 - a mental operation right exactly it's not a logical operation maybe i agree
  • fast_forward00:59:15 - with you but it's a mental operation it determines that in in a space that is no physical,
  • fast_forward00:59:21 - these symbols need to be represented right and this space could be uh modulated by a physical.
  • fast_forward00:59:31 - Signal that is the angle of gaze that normally is well known to be able to modulate
  • fast_forward00:59:37 - visual response in the space. Exactly right.
  • fast_forward00:59:43 - We're saying the same thing. It's also on that account that I find that your
  • fast_forward00:59:48 - results I think are very exciting.
  • fast_forward00:59:49 - But now, so you're active in systems neuroscience, you look at these really
  • fast_forward00:59:55 - complex processes, logical operations in monkeys, whether They are intrinsically logical or not.
  • fast_forward01:00:02 - It's very interesting and very exciting.
  • fast_forward01:00:04 - You started as a neurologist, switching over to neuroscience.
  • fast_forward01:00:08 - So if we would like to follow the study of the brain in the tradition of Stefano
  • fast_forward01:00:15 - Ferreira, what would be the Stefano law of the study of the brain?
  • fast_forward01:00:19 - I think that is a little bit far from some rules that are currently available,
  • fast_forward01:00:28 - at the financing level in Europe. Okay.
  • fast_forward01:00:31 - I want first to know exactly how the brain works and then maybe to model it,
  • fast_forward01:00:36 - to understand properly and to translational apply to this knowledge to pathology. Okay.
  • fast_forward01:00:43 - So and then predictions. So soon I'm going to come to Rome four years from now.
  • fast_forward01:00:48 - I'm going to go to your lab.
  • fast_forward01:00:50 - I'm going to ask you, okay, four years ago you gave me this prediction.
  • fast_forward01:00:54 - Show me whether you tested it and what was the outcome. So what's the one prediction
  • fast_forward01:00:58 - you would make today that you feel most committed to?
  • fast_forward01:01:03 - For the four years? Next four years? Yeah, four years, yes. Okay,
  • fast_forward01:01:07 - let's consider first that monkey experiments, primate experiments in general are very slow.
  • fast_forward01:01:13 - So in four years I will be able probably to complete just one experiment.
  • fast_forward01:01:18 - Or let's say to have one finished at the second at the beginning of the trip.
  • fast_forward01:01:25 - But the next for me will be to be able to record from from many areas of the brain simultaneously.
  • fast_forward01:01:37 - In this task and other tasks that I'm using at the moment in the lab,
  • fast_forward01:01:43 - we are using it in the lab,
  • fast_forward01:01:46 - and at the same time to have assessed to different level of the neural signal,
  • fast_forward01:01:52 - to provide a better decode to whatever is going on in the population of neurons
  • fast_forward01:01:59 - controlling the behavior that I'm interested.
  • fast_forward01:02:04 - So in particular for the transitive inference task we are already studying the
  • fast_forward01:02:14 - premotor cortex and the goal is to go to the posterior parietal following what is the,
  • fast_forward01:02:22 - network described in humans and by having access to a high resolution signal
  • fast_forward01:02:29 - to provide details on what is still unknown,
  • fast_forward01:02:34 - on the possibility of the brain to provide support to such a behavior.
  • fast_forward01:02:42 - But that seems very methodological. Do you have a specific prediction you're testing in doing this?
  • fast_forward01:02:50 - The prediction is that it's not very singular, but I still believe that need to be confirmed.
  • fast_forward01:03:07 - Properly that the behavior like the one that I'm studying need to be coordinated by different areas.
  • fast_forward01:03:14 - I would like to find the key of how this coordination happens.
  • fast_forward01:03:22 - Okay, great. Stefano Ferreira, thank you very much for this conversation. Thank you, Paul.
  • fast_forward01:03:30 - The CSN podcast was produced by the Convergent Science Network of Biometrics
  • fast_forward01:03:36 - and Biohybrid Systems, a project funded by the European Sevens Research Framework Program.
  • fast_forward01:03:44 - For more interviews, recorded lectures, or upcoming conferences in the field
  • fast_forward01:03:49 - of biometrics and biohybrid systems, go to csnnetwork.com.
  • fast_forward01:03:56 - Music.

Be the first to leave a comment

Leave a comment

Your email address will not be published. Required fields are marked *

Convergent PozitivLinia Logo

Exploring the convergence of neuroscience, robotics, and AI through conversations with leading researchers since 2010.

A project of the Convergent Science Network Foundation.

© CSN Podcasts. Developed by IMCreativeWEBC

0%

Login to enjoy full advantages

Please login or subscribe to continue.

Go Premium!

Enjoy the full advantage of the premium access.

Stop following

Unfollow Cancel

Cancel subscription

Are you sure you want to cancel your subscription? You will lose your Premium access and stored playlists.

Go back Confirm cancellation