 good morning. So my pleasure is to thank all of you, your attendants. This is the first integrative seminar series of this year. We have been, we got a gap in the last semester but we have decided to start with Luca. So thank you Luca for accepting the invitation from the department. But first of all I would like to say a couple of things. There have been small changes in the organization of the integrative research seminar. So after the seminar we will have several snacks outside just to try to keep the research discussion ongoing. This is part of one of the changes. Another change is that there will be a small evaluation of the organization itself after the seminar. We will send an email to all the people that has been registered to the seminar. Today's speaker is Luca. Luca has been with us for six years I think. He is a NICREA senior researcher at the DETIC specifically at the Center for Brain and Cognition. That by the way they have moved recently to the other campus. I see this movement as a part of an invasion of the DETIC to the Ciutadella campus. But today talks is about ontogenesis of rational thinking. Luca has been moving all over the world. He has been in several parts of the state. He got a PhD at Rockers University some time ago. We won't say the year. But he has been also in France, in Italy and now in Barcelona doing nice things about rational thinking. So you have 45 minutes and we will have some questions I hope. Thank you very much. So what I am presenting here is sort of an overview and trying to make a small argument. So part of the work is from my lab and part of the work is as you will see the most interesting one from other labs. This is collaborative work and these are the persons who mostly contribute to what I am presenting today and there are other collaborators a bit all over the world. This is I mean the nice part some of you who teach at the Masters might recognize some of the names that are here because although the integrative part of the seminar doesn't work maybe too well from the top, certainly the bottom is starting getting integrated. So there are several students who took our master at CBC who also were past or would become future students in the Masters that the DETIC is offering. That's a good sign. A little word about the title I have to say I am a philosopher as a training and I always like to do something which is philosophically interesting but never I was never able to do anything interesting really so but I always like to do something really deep and food for thought. So this time I got this occasion to come and thank you very much for giving it to me and I thought okay these are engineers what do they stuck a title that would satisfy my philosophical ambitions but that's all what I have to say about ontogenesis of rational thinking indeed nobody knows anything about it and certainly I don't so the only reason for the title was to invite people to come here. Okay let me tell you really what I am after and I'm going back to it I really after this kind of thing the fact that as early as you look at the archaeological records you find productions in humans which are extremely sophisticated paintings like this one but also computers for example or for that matter and the sink and the siphon in the sink something that we should be grateful every time we go to the bathroom every day right what I mean is there is an incredible amount of material culture that we tend to underestimate in everything in every single behavior that characterizes human species and everything we see here is a set of invention is a set of solved problems from clothes to computers to everything and I'm fascinated by it and I would like to somewhat give my little contribution to not to explain this phenomenon we never be able to explain but to somewhat to somewhat know where to move in this search okay now the to explain these my talk is to study young infants before they start being able to teach so to interact with us but you as a current and future and prospective engineers or computer scientist might ask yourself why should I be interested in young infants and so first before going on to the actual research I wanted to make a couple of points which can be summarized by these this title you know you're very much into deep learning here but you know my feeling is that there are other things that maybe deep learning might not necessarily solve completely and here is one is one of the most active areas of research in these years self-driving cars great I mean this is something on which deep learning are applied however there are also and excuse me for the pan there are also some ethical issues that really deep learning is not going to help you to solve so if at a certain point you have a little child who is in front of you and the tree you've got to figure out whether your car says to you know crush against the tree or kill the little child right that's the problem is not going to be solvable if not by studying the philosophical issues what happens if instead of one person there are very many persons in the car right how does this change these are decisions that have to be taken no way to find them out of the data that's one of the point I wanted to make and this is the second a little closer to what I want to talk about and it has to do with fact that there are psychological issues that one has to keep present when when asking oneself how can I make a machine that thinks or how can I make a machine that interact with humans and to exemplify this let me just give you an example that come from the from these you look at on the web and you find that now we have these perfect companions who which are the mottics robots our new best friends who by the way is very good because it uses deep learning protocols to interact with humans okay very well but you know how do you actually interact with humans and here is an example that come from the tomasello laboratory of our human interpreter situation and so this is a little child who sees a set of actions and and what the child does is a spontaneous helping behavior right and there is not much data that is knows what the what the what the situation is about and he knows how to interpret the intentions of the speaker so my point is that if you really want to solve problems in robotics you need to figure out how to interact in the right way with these humans and for example with infants and you need to ask questions about what does a human or a child want what kind of psychology psychology does he have a little child doesn't have the same psychology as an adult what kind of conclusion and this is my my main focus what kind of conclusion can a little child draw from a given situation so that there is an inference being made expectations being created and this is the problem that I believe put together people who work in my domain and people who might work in your domain these are not issues that you saw by simply extracting information from the data you have to figure out something about the early structures of reasoning and thinking from which our ability to create these things come from on to business now if you want to sort of see the kind of words that you might use in order to explain this incredible productivity that characterizes our species you you find this word creativity problem solving tool invention use material culture cause explanation and science among all these kind of things but all of this starts with the fact that we we we are born as infants and and therefore we somewhat have some a priori about what we want what kind of conclusion we can do what kind of psychology we have and what is the model of early psychology that has been very successful in the last 40 years well this is the model the model that the infants can do a lot of things much more than what people thought before because they have modules which are essentially pre-compiled ways to solve problems which have been selected by evolution and these are some of the areas on which researchers have found that that infants have priors very well structured in terms of ability to solve problems categorization number geometry theory of mind objects and so on there is no denying that these exists recently there is another picture that starts being proposed which has more to do with rational thinking and the picture shows tries to show that there is much more than just pre-compiled ways to solve simple problems and I want to mention three areas in the literature that I think are particularly interesting one is the work that has been done in Budapest by the group of Yuri Gagel and Gergo Cibra about how we interpret actually infants interpret actions in a rational way one is the work done at MIT by Laura Schultz and her group about how infants quickly attributes causality and test hypothesis about failures of situations and one is the work by the group at Johns Hopkins by Lisa Feigens and her collaborator about how infants use rationally failures of understanding as occasions to learn. These are all three kinds of evidence sets of evidence that suggest that really infants are early rational thinkers and I invite you who are more into artificial intelligence to think of how you would as it were incorporate into a machine the kind of phenomena that I am presenting now so let me talk about the very famous work by Yuri Gagel and Gergo Cibra in the early late 90s and the idea was the following you take an infant and we're talking about infants from now we know from three months on so quite early and you habituate infants you show them this ball that jumps this fence several times and then gets close to this ball and then makes like kisses to this ball okay so this is the physical action that infants are habituated to and then you present them with two novel situations that they've never seen one is the one in which the goal is achieved in the most optimal way and one is the one in which the actual action that infants have been habituated to is repeated okay you would imagine that if infants as it were stay to the surface of the situation they should find this action surprising because it's new these they have seen it many many times before okay turns out that that's not the case infants are surprised when they see the old action because now the old action doesn't make sense with respect to the goal that the yellow ball had before okay that suggests that infant don't look at the situation as a physical set of happenings but they look at them in by asking themselves what is the goal of something okay now goals can be very sophisticated and indeed as i was mentioning you before the fact that people agree on the fact that infants have a very good understanding of physical interaction between objects but they are also able to override this understanding if there is a goal so this is early i mean closer to us work by always the the group the buddha best group and what south gate and colleagues showed is that when infants see something like that okay they're just not surprised at the physical violation provided that the goal is achieved they have very clear what what the goal should be about but the goals can be something very educated so consider this situation again coming from the buddha best group this is previous work in which and the mouth of showed that when infants see a very funny action they repeat it they imitate it as many other animals do so if you show that this light lights when you hit it with the head infants will tend to repeat these actions and this is what the gargoyle group showed in this situation here is a situation in which this woman shows that she is called as she might be in barcelona in these days and and that's what this woman does she puts a particular you know a blanket on her and then she performs an action okay the action is very funny but you see the action is made in such a way that to the infant it should appear clear that she wants to do it that way okay and that's what an infant exposed to this will do he will try to repeat the actions in the following way however however here is a very similar situation where there is only a little change now experimenter is always called she always puts a blanket on her uses her hands to keep the blanket okay therefore now the action is being repeated equally but as it were she does it that way because she couldn't have done it in another way because she has hands occupied sophisticated interpretation of an action right and what happens in this case is that infants do not repeat the action before they tend to actually you know hit light the ball in this way the the light in this way so that's really these are the data the the probability of repeating the action physically only occurs very frequently when the experimenter is showing that she could have done it otherwise as she wanted to therefore the relevant repetition there is the physical repetition of the action but if the experimenter has her hands occupied okay then infants don't repeat the physical action they repeat the goal of the action that's to tell you that when infants see a situation they understand not just the physical sequence but the deep goal of what it is in the mind of the actor okay this is something that comes from the this first or second point something comes from the laboratory of MIT when the idea is very simple we are always talking about infants about 14 months old of age you have a little toy that when you press it makes music and the experimenter and then the infant has two other toys one is close to her and the other is far from her and here is what happens the experimenter takes this toy and every time he presses the toy the toy makes this like nice noise always succeeding and then the nice I mean the the the the manipulation here is that in some condition the child gets the same toy or gets the toy that is close to her okay and the trick is when the child activates the toy it never works what does the child do okay and that's what what the question here is what does she do does she get another toy because she thinks that the toy is broken so it's not she who doesn't know how to activate it but it's the toy that is broken or does she ask help to the experimenter asking to see it again because she thinks that she didn't understand how the the toy works okay so the evidence available should give information to the child to get the right solution and this is what happens in the when the child gets the green toy okay the one that was activated by the experimenter successfully when the child sees that this experimenter was able to to succeed she gives back to the experimenter but when this doesn't work the child goes a long way to get the second one right assuming that because the evidence is consistent with the fact that she knows how to use the toy but it's the toy that is broken okay it's not the proper thing to do to ask the experimenter to repeat the action okay that's suggested that infants at the age are okay these are the data so you can see this is the amount of time that the children give the toy back to the experimenter or get the other far toy from them and it's very clear what happens infants use the available evidence to discard the hypothesis about why something is working why something is not working and how they should you know achieve their own goal in this situation okay so another example in my opinion of a very rational behavior and let me just give you the last one and this is really a nice one everybody who works with infants work with this method which is called the violation of expectations which i'm going to show you data about later which essentially consists in showing infants things that work in a funny way and look at how long or how surprised they are when they see their expectations being violated right nobody asked the question all right but when infants see something that is funny what do they do with it and and this is what Lisa Fagenson did recently and the way they did it is they presented infants with physical violations that they know infants are surprised about okay for example violation of support or as a violation of solidity okay we know that infants are surprised they look longer this situation than the normal situation but that's not the question that Fagenson asked the question is all right now i give infants the toys that underwent these violations what do they do okay when they see a violation of support and they give you i give you infant the toy what does he do does he bang it or does he drop it when he sees a violation of solidity what does he do does he bang it or does he drop it right assuming that if infants are testing hypotheses about the violations they should bang the toy that you know didn't respect solidity more than drop it and they should drop the toy that didn't respect support more than bang it and that is exactly what Fagenson showed this is an index of how much dropping actions or banging action there are and you can see that when there is a violation of support infants essentially bang drop the object and when there is a violation of solidity they bang it that is there are really hypotheses about what the violation says okay okay so these were three examples of far away and recent results about infants that seem to suggest that infants understand goal and intentions these goal and intentions are very subtle and they can understand they can understand them by analyzing the situation at a level which shows an extremely sophisticated psychology they have expectation about that they make a hypothesis what should happen they understand you know that if something in an expected that is a learning occasion they can test cases of failures and they can explore rationally alternatives all of these seems to suggest that infants are very good rational thinkers and therefore the question that we are trying to work on with our little tools of the what we have available is all right if infants are not just sets of modules if infants have this ability to rationally solve problems perhaps an ability that explains a little why we have all these fantastic problem solving activities that make our life of everyday well how do they start from what mechanics make rational thinking possible okay well certainly if you have a hypothesis you need to conceive thoughts and you need to combine them okay but on the basis of what primitives what are the very early thoughts that you can have on the basis of what combination principles how what inferences can you draw okay and the way we do it we try to figure out what whether infants have very very basic intuitions about what could happen in the future what might have happened what didn't happen why didn't happen okay and to do that we test the ability of infants to have expectation about probable events probable future events or expectation about some logical consequence of a very simple situation let me just show you some of the results that we had here is a so our game is always to present situations in a non-linguistic way i'll come back on it in which there is information infants should have they had a rational way to analyze the situation so in this situation for example infancy a billiard device in which a ball hits they hear the noise i took it away here and at the very moment in which the ball you know disappears i cut the movie so what infancy is a little more extended than that of course the ball is at the center so you cannot see you know you cannot predict in this moment where it would come out okay and the question here is okay do they expect that because there are three exits here it's more likely that the ball comes out from this side than from the other side okay this really would require Monte Carlo simulations if you really needed to to get to this situation and our game is presenting least possible so that we thought and now the story is yeah they are surprised when this ball comes out from this side okay however they could be surprised for many reasons so we compare this situation with a situation in which the device is very similar perceptually but now exiting from one particular side becomes impossible so if they respond by using very surface as it were analysis of the situation they should be maybe surprised if the ball comes out here but they should be surprised equally here but if they analyze the situation as probable improbable possible impossible by using these modal terms then they should reverse their behavior when they see an exit from this side or from the other side is the logic clear and that's what happens if infants see an exit for one hole here they are more surprised but if they see the exit from the exit where the three patches here they are more surprised in this situation okay showing that they're really analyzing the probability and the possibilities involved in this situation we also can show that in certain condition they even anticipate the exit so in the very same situation there is an occlusion in our situation and then the ball is behind and the question here is before they see where the ball comes out do they expect that it will come out when there are the three holes so we test the anticipation to the exit in the one two seconds in which they don't see the ball and what you're seeing here is well for simplicity suppose the ball always comes out from this side and we flip the we flip this frame so that sometimes it has to be exit on one side sometimes it has one exit on one side okay and what we found here is that before the exit occurs their eyes tend to move more towards the three sides than towards the one side okay showing that probable to occur now let me mention this in the sense of probability that we propose here they don't get this information out of experience it's not just frequency checking it's really feelings that things should occur in this way if you analyze the situation in a particular way that i try to make clear in a second okay now interestingly okay so that's that's the data is not important interestingly they don't always anticipate and that is very funny right remember i gave you this situation this two situation the situation i'm oops i'm sorry something that happened here probably doesn't there is there is a situation in which there is a physical impossibility this situation here and there is a physical possibility probable or improbable and what we found is that infants are always surprised at the improbable or the impossible this is after the fact but before the fact so when there is this expectation potential expectations they tend to look more at the probable okay but they don't care to look more and the only possible way out in other words they anticipate the probable situation they don't anticipate a possible the only possible situation that is somewhat funny so at a certain at the beginning we thought okay they do that because our movies are really too subtle for them to pick up what is probable what is improbable what is possible what is impossible that makes things clearer and they make sure that the infants understand that the ball cannot really come out from here you know just we put every possible information they should not come out will they anticipate that the ball will come out from the only possible hold now and the answer is no they do not more specifically after the fact so when they are when we test their violation of surprise they're always surprised at improbable exit or impossible exit okay but before so when they have to expect something they're only surprised when there is a probability stake not when when not when there is something essentially deterministic now if you think about it that's also very rational because what infants are doing here is they move their eyes only if there is a possibility of reducing the uncertainty of a situation why should i anticipate something that i know is going to happen that way when there is a probability distributions that's the moment in which i have knowledge to gain this can be made formally precise and that's what we find that infants anticipate only when there is probability a stake okay unless i mean there is nothing else of course if you beat them up they will probably anticipate also the possibility but that's not what happens okay this situation this kind of behavior does not occur because of any particular feature of our stimuli here are completely different situations which maintain physically different situations which are maintained though the same kind of logic here is not the device the car is the probability but the number of objects and you can see here that here it would be an improbable exit here would be a probable exit and what we found is that even in this case if infants are surprised when they see an improbable situation and again this is not carried by the fact that these stimuli are different physically for example they could look longer here because there is some sort of gestalt you know principle here so that they look longer because there are three things here which are sticking together okay not the case because if you transform these situations in possible versus impossible situation by just simply putting a bar that forbids the yellow object to come out we find again a complete reversal of infants looking behavior so they look longer at an improbable situation but they look longer at the impossible situation even if physically the situations are very comparable okay very well they don't we can do something more we can actually ask questions about okay on the basis of what mental representation do they do these kind of conclusions and do they represent possible outcomes possible future states or do they simply represent ratio between quantities and to do that we did the situations in which we presented the same distribution I'm only presented the final part of the movies I hope that the situation is clear here there should be an object coming out and it doesn't so in one case we have a three versus one in the other case we have a 12 versus four and what we found is that infants can anticipate the next future event in this situation but they have no expectation in this situation suggesting that they don't represent the ratio but they represent really the amount of possibilities so in so far as they can imagine them few possibilities of what might happen in the next future moment of their presentation we can also show that this is really not depending on advice and difficulty at understanding these devices because if we put another kind of information for example density making these objects bigger these are the same objects but bigger so it's really hard here physically for these objects to come out say in one second but it's easy for these objects to come out in one second then they they they do the right kind of reasoning okay okay all of this suggests that they have a pretty sophisticated sense of probability doesn't depend on the surface feature of the situation but it does depend on how many probabilities they can envision okay so what we are trying to do in in our current work is to ask the question all right if they really have this sense of probability where does it come from and the idea that you are trying to push is that it really comes from their sense of possibility from the modal interpretation of the situations this is an old idea that you can still that you can find already in in Wittgenstein and this idea that oops that facts in logical space so what we see is not what we see mama mia sorry i don't know why what we see is not what we see but what we see is what we could possibly see had we present the possible distribution of possibilities for the future events so if you assume this idea if you have a logic so if you have a sense of modality probability comes for free essentially just count the possibilities and that's what we are trying to figure out now whether we can we can show that infants have a sense of logic that would allow them to reason about the probabilities of the next future events if they can count them there is primary work that we did well but does it make sense to ask questions about okay do you have your a logic in your in your brain maybe you don't know so the first thing we did we work with adults and we asked okay if i show adults very simple reasoning like this one for example um suppose i give you a premise that means nothing all roughs are blanket okay then you see a second premise which is john is a knock what can you do nothing these are unconnected that's the only thing you can do but if sentence would have been all knocks are blanket then you can integrate these two and you can make a logical consequence so what we do here is we we compare when adults see this sentence whether the previous sentence was integrable by by a logical inference or not and we are asking the following question if you give me the the the activity of your brain can i predict what you are doing okay whether you will integrate this information logically or not and what we found is that okay that's not important what happens later is that in this very moment there is a there is a lot of activity in the brain in these situations some of this activity can depend on the logical forms for example you i can tell you all a's are b's all c's are b's and you can conclude all a's are c's and that could be wrong right but every time you see this reasoning you do the same mistake okay so you act on the logical form but you don't in other situation all a's are b's or b's are c's you could act on the logical form and get the right solution and what we showed is that there are people for whom we can predict that they are reasoning on the simple form of the statement and there are people for whom we can predict that they reason by applying a very elementary logical rule and and the area of the brain that are more active in this in these two situations are different one is essentially one left ventratera profontal more frontal activity connected to reasoning logically and more as it were linguistic like activity connected to simple detection of the form so in the adult brain it can be shown that when you reason with by very elementary logical reasoning there is a particular pattern of activity that is predictive of what you are going to do so our our aim now is to really take extremely elementary things something that i'm ashamed to present to people that do artificial intelligence like instantiation or elimination alternatives a or b notated for b or modus ponens and transform these possible inferences in non-linguistic situation and see whether infants make conclusions on the basis of it and today i'm going to present and the interesting thing about that is all these situations have the same structure logically you need to have logical forms and you need to have elementary inferences that for which you derive information from a logical form and what i'm going to present to you today is the way we are trying to attack this problem do infants do a particular inference do they represent the logical form of disjunction and they get to a conclusion when we present to them since they might contain this logical information and the way to do it is the following we create the situations in which there are always two objects that are different different categories and everything but at a certain point you cannot distinguish which object is here you can only distinguish when you see the second object okay so the flower you know that the flower is here when you see the dinos are coming out and you know that the the little baby is here or the other of the what was the other or the umbrella is here when you see the little baby out sometimes the end of these movies is compatible with you having drawn a logical inference sometimes is incompatible okay so there are two aspects here i'm going to go back to it but when we look at the ends is essentially when they're when we look at surprise we find that both at 12 months and at 90 months infants are surprised when they see situation that is incompatible with a very elementary logical inference okay that is not dependent again on the way you present the inference here is another way in which you present the same kind of situation but when you have the epistemic evidence that about this object physically the scene is completely a completely different way and this would be a way in which this is the end of the scene this part here okay that you should say this is the ball but then maybe you see the ball here that would be the incompatible one physically this situation is completely different from the previous one okay nevertheless both at 12 months at 19 months we find that infants are surprised when they see a situation that is incompatible with a logical inference okay what is our concern oh yeah they also they also use the situation to predict so this is recorded directly so the stimuli are better when you present them but as you see here there is no violation of nothing where only ask me whether the fact that they can draw the position of the object and then at the end we show what was the what was the response okay and what we find here also that when infants see these situations because they're very interested in faces they use the inference to figure out where the face is even if there is no violation so suggesting and it's not like that they're surprised it's something that is not respecting a logical inference but they use physically inferences to make expectations about where objects should be okay now these scenes these scenes are simple but they have this this characteristic namely that they can be decomposing in very particular moments so here is a moment in which we could call this the event representations you present simply objects and the particular situation infants can represent them and that's it that's what they can put into memory right but then there is a little second moment in which say if there is anything like a logical form of thought they could make a disjunctive hypothesis i don't know which object it's either a or b okay then there's a third moment in which sorry that this moves it doesn't move in our stimuli when they could make deduction remember what i was saying to you before every elementary logic has this fact you need to distinguish logical form from inference and that's what you could do here right and then there is a moment in which you can verify if the hypothesis is respected or not okay so what we are trying to do in our laboratory is to go through these steps to figure out if there are behavioral markers that can tell us okay the infant is here the infant is here that's what he's doing here that's what he's doing there that's the aim and to do that we can modify situations in which in certain conditions we take away the disjunctive hypothesis how do we do it okay if i represent the same scene but now you see what is happening okay you need no hypothesis you have representation of the object you can check whether what you see here is respected or not but you do know inference so we compare situations in which at the very infant should draw an inference with situations in which at the very infant should look for behavioral markers what i'm going to present now are data that oh yeah yeah for example during deduction making how what what do they actually do so for example if they're really doing a thinking process in this particular moment that this object comes out therefore they know they could know what the object is here maybe they you know they switch back and forth with respect to these two objects maybe they tend to look more to this object here right which is not interesting i mean the most interesting object by far this one but this is the strategic one to figure out you know whether whether you're making an inference what is inside that object and so what we do is i tracking to compare these two situations in in the situation in which you require an inference in a situation in which you don't require any inference but because you know what object is inside the cup in this exactly equal physical situation in this particular 1.2 1.5 moment of of the setting up of our scenes what i'm going to show you to you now are i tracking data with adults we show that during the inference phase adults tend to look more at this object than at this object suggesting that oh these are adults that have no task so they don't so look at this point here of course they look more at the only object that you can see but when there is the possibility to draw an inference they tend to look also more at the cup the less interesting object this is the plot of the differences we suggest that they're doing something with respect to what is in the cup and more precisely what we do here is we track the average position of the eye during this little moment okay according so this would be the moment in which you can draw the inference or in the other situation this would be the moment in which you can draw the inference what do you do with your eyes during this this moment and what we do here is we we look at two parameters one is so here i'm putting in the y axis the x position mean x position of the eyes in this very moment and what we what you can show here is that during this moment of inference adults tend to look more at the cup when there is an inference when is it possible inference to be drawn than when there is no inference to be drawn at the very same moment at the very same similar situation this is a 1.5 second is very very short their pupil dilates more when they can make an inference than when there is no inference to be made suggesting that there is work being done there even if adults have no logical task to perform they are really doing something that we suggest is a very elementary logical inference okay we are working on that with infants currently in the very same moment we look at what infants do these are 90 months old infants and we find surprisingly essentially that both in and in the size of their pupil a 90 month old response is exactly that of an adult that the the mechanisms that underline the resolution of a very small problem like this one seem to be a change from early infancy to adulthood okay this i don't want to show and yes and therefore we go back and see what what we want to do this was the model that we have in what is happening in adults there is a very clear marker of when you have presented a logical form of a problem linguistically and a very clear marker of when you are making an inference but what is the network organization of a brain when you present the very same things but without language we we know pretty well what adults do when we ask them to reason linguistically there is a long long literature on that but we don't reason only when we hear somebody speaking we reason when when we go around we reason when we dream we reason in any situation right and these situations are not linguistic are the brain and behavioral processes underlying our ability to reason logically the same regardless of whether there is language or not so what we are trying to investigate now is whether the very same structural organization of the brain the sponsor of the brain to very elementary logical problems occurs if we present the very same situations without giving adults any explicit logical task or any explicit linguistic information okay here's what i did i i tried to convince you that very elementary logical reasoning have this aspect of having a logical form which could be a logical form or language or a city and an inferential part i tried to show you that in infants and in adults you can find behavioral markers of all these processes i tried to argue if you have this you can also have a foundation of an of the intuitive sense of probability that we think we have shown that infants have this occurs in different ways but essentially in the same way at the pre-linguistic stage and i suggest i would suggest that this is might give some little insights into the origin of our rational thoughts and to go back to our story what i suggested you cannot forget these questions even if you work from the point of view of the abstract analysis of what a machine could do you really need to understand if you want to understand interaction is what is the kid doing right now how is he doing the inference that i think he should be doing now under what kind of evidence and what is the information that will lead the child to learn about a violation not a violation and so on how much of it is logic how much of it comes from his intuition of what is going to happen in the future and and so maybe there is something for which to run more than a simple integrative seminar when we think about our different ways to look at the very same problem which is essentially the origin of human rationality thank you very much 45 minutes and 22 seconds children so a couple of questions so first about so the status so anybody in psychology okay doesn't buy this idea that the baby is a rational little scientist on the crib I mean or something like that is well I would say 98.7% of the literature in fact the original work that I presented that were already started was this goal directness of gargely in the early 2000 essentially you couldn't and it's even now you cannot publish a paper if you put the word reasoning in inference because how could they possibly reason they don't even you know they cannot even speak and there is more precisely a literature that suggests that our reasoning ability are heavily dependent on language with some reasons so the answer is well maybe it may be surprising to you right like it may be surprising to me that there are people who think that we are very behind in understanding artificial intelligence right but yes that's not the normal as it were the standard view yeah okay essentially all the evidence that I gave you is all the evidence all the groups that work on this hypothesis three or four in the world yeah okay so it's second part so in terms of engineers and psychologists so your view actually is very close to engineers or computer scientists you're saying okay forget about psychology let's see observation let's use the tools of probability and logic okay to understand I mean how and these are not tools that are foreign to engineers at all actually so you talk about mechanisms but it's something probably talk about already but you're missing one ingredients you have probability you have logic that is about what it's computed but not how it is computed okay because actually it's a sort of intractable framework from a computational point of view and then you need to to do heuristics or whatever okay in order to use this framework so I have no objection to what you're saying what I am saying though is that you're not going to make progress if you want to understand the if you want to make a computer that interacts with a human if you really don't have a clear idea of under what condition can an infant draw this inference for example if there are three objects can he draw it if there are seven can he draw it and there are theories about that and there is evidence about that suggesting under what temporal frames can he draw it should I actually as a computer as a robot help the infant now or is it the moment in which the infant is learning something therefore I am interfering with his learning experience all these kind of questions require you to have a very clear idea of the detailed process of these very elementary kinds of reasoning so I don't dispute what they're saying absolutely what I'm suggesting though is that progress can be made if you really take into consideration in our case the what which is still something that needs to be and if the how once you try to if so if you set up heuristics for example is informed by the actual state of an infant mind that's what I'm suggesting thank you okay yes are you aware of here oh hello so are you aware of studies in the same vein regarding infants and ethics this is so that there are three traditional concepts of philosophy the true the good and the beautiful right these are really the three pillars of philosophy we are making some progress on the true we are starting on the good and that we we don't know anything about the beautiful let's put it away so yes there is some little evidence that that infants infants children let's put it away have an ethical view for example they are spontaneous helpers as the tomas yellow group try to show they are willing to share a good even if these gives them disadvantage in so far as they have a sense of fairness that is respected in in the group this literature exists does it go down in our area namely pre-linguistic and when infants cannot even act I don't know there are many people here who work on this stuff I would say that we still don't know what happens in the pre-linguistic page but they start there is some work which is start being done on that thank you very much okay thank you very much for the talk there is extensive work by Kahneman and Tversky who argue that we humans are not very rational thinkers how do you reconcile that do you attribute that to the fact that they mainly used language to formulate their questions or yeah I would say take away the Nobel prize to Kahneman give it to me that's essentially the short answer no really there's no denying about all of this but from the so the the massive evidence for example when you present problems which are slightly more complicated than the ones they presented adults fail miserably in giving the correct responses now there is a long long literature why they do it I think the Tversky and Kahneman were extremely good at finding situations in which they would tend to do it in the wrong way but it's not very clear you know what happens in any of the single situation that Tversky and Kahneman presented from my point of view and there is work that we try to start doing we don't deny the existence of errors or reasoning what we are saying is that these become developmental problems so there is a moment in which you start making mistakes as an adult on the basis of the fact that you could potentially have the resources the logical resources not to make the very same mistakes what we try to do is to find out certain conditions under which the repeated response supposedly give you a very frequent piece of information that is countering your prior so this situation there is this object with three doors here and one door here your intuitive expectation is that the ball has to come out from the three door information that that goes on the opposite direction and then what how much experience do you need to give up and say well I don't know what happens but you know I would expect that it will come out from the improbable side right and I think it is possible to generate situations in which the biases that Tversky and Kahneman have give you they are mistakes they're reasonable mistakes if we want to execute frontal processes and as the situation I don't think they be saying something different to what I'm saying up to now the extra point in which Kahneman and Tversky would say something different is oh but this is all has to do with frontal processes they have to do with a a forceful you know cold reasoning we in the hot reasoning we really go by biases thank you um I think that's where we might disagree I think that there is evidence that this kind of early elementary logical reasoning is as hot as the mistakes of Tversky and Kahneman so to the short answer is I take these points very much and they're very crucial I think there is an answer and it is they become developmental issues not as it were initial state issues the question is to find out how mistakes arise not why they are there is it sufficient for you yes okay thank you should I thank you I'm absolutely not in that field but I recently read an article that was talking about empathy in children's so that children's they have early infants have an innate sense of empathy and you say they're natural helpers but then around two years old three years old then they start to have an acquired empathy based on experience which is extremely intricate to to interpret and might indeed sometimes conflict was a was a notion of empathy so the question is if you can understand so how a basic Christian thinking is working then can you further use it as clue premises to interpret intricate situations and exploit that knowledge for example in conflict management or something like this okay this is a really really tough question and it goes way beyond the what we could address I think experimentally now now consider what I'm studying here I'm studying a or b not either for b right so from there to figure out you know how intricate situations can be used to as it were to maintain the natural empathy as opposed to entering a conflict situation there is a lot lot of work to be done my my feeling is that if you understand how problems can be made simple by using the intuitive logical and probabilistic routines that we know that infants can handle then you have a handle also to this problem but that's as much as I can tell you right because they're you know we are talking about extremely complex situations and it's not very clear that what I'm saying can be done very nice talk so going back to the question of Philip is there any systematic bias or mistakes that infants make while reasoning is there like a like a gold standard for that or is they're perfectly rational in everything that has been tested so far the answer is they are perfectly rational in everything that has been tested so far because that's what has been tested so far right that's I mean I gave you this is a very as it were unique occasions and you should be grateful to me I gave you the full literature which is available now in this topic right so nobody has really worked on biases in infants and so I'm very sure that there will be many that you can find and there are certainly some which must exist because there are perceptual mechanisms that will lead you to see things that don't they don't they are not there and so on but if you want I am coming from the other point of view I'm coming from the point of view in which I'm trying to show the little part of rationality that exists on the basis of the mountain of rationality that has been claimed to exist at least in adults in infants it has been assumed that if adults really make these mistakes obviously infants will so nobody really had looked deeply and it's not so easy to for example take a problem a lot worse can come in and transform it into a problem that they can represent non-linguistically that's really really hard so the answer to it is these are all things that should be done what about the variability across children because you are providing like a I mean methodologically you have like average behavior across the population so so can you tell more about the individuals yes this would be our our hope the answer is no because we are so behind that we are really not at the point at least in this area to do individual differences studies right now because we are trying to find out for example and there are many many other let me give you one which strikes me as something that we never thought and those who work in infants and it is important we are working with eye behavior of infants nobody looked at it but there was a student who is finishing now his dissertation in in England and who might follow at the distance and at the beginning he had this very funny thing yeah do it but you know here this idea to find the psychological profiles of infants and see how they affect how infants got information we know that there is something that is close to truth there because when the infants come to the laboratory half of them want to escape just you know the and half of them stay there I mean maybe the one that we are looking at and the one that we introduce into in our data are the ones that are really the most stupid I mean if you're really smart you don't want to do our experiments okay so what you did is he looked at the scanning pattern of infants by coupling it with profiles that can be gathered of their shyness and he found the shy infants have a completely different scanning profile in looking at the same scenes right suggesting that there are lots of dimensions on which you can actually ask individual differences questions and we would very much like to go so we have the possibility for example in our data to look at these at the correlation that exists between for example I have this 1.5 seconds in which I give you the evidence that it would be allowed which is sufficient for you to draw an inference about this object but there are infants who don't do it right now if we were able to sort of profile a little better certain kind of covariates that we could allow us to determine the population then I think we could do it we are not at the point with the lab in the laboratory we started doing that with a severely impaired infants who have this is a work that is done by Nurese Bastian and the hospital clinic together we take our data our kind of experiments and we try to characterize whether some of them are predictive at least of some of the deficits that some of the infants might have if they have certain kind of neurological disorders that's as far as we go but of course the really interesting question for us would be studying the individual differences in the normal child I mean the normally developing child I don't think we are there and I would love if we were able to get there is it a good question or a very good question it's the last one you cannot so in you have shown that in adults for example at least some of the reasoning processing are language based or language mediated I'm referring to your study with Carlo yes well that's not exactly what we we try to show but we try to show and this is I mean at least mediated by language this is all work that uses language because there is no other way until now there's only one paper which has ever been published that they know of which studies non-linguistic reasoning in adults I mean in imaging okay because it's very difficult and it's in principle impossible really because if I give you a scene what is the reason that you think that the elementary representation that you are making is that representation and the advantage of language on which there are also issues is this one that I block for this factor right so these are all studies that have to do with the verbally presented material right and what we see is that in the verbally presented material this is by using predictive coding techniques right higher activation of me by a 45 predicts the behavioral responses of the adults who make mistakes but they make them consistently okay so so the language the language well this is not really language area yeah so the point is you get to hear and that is what we would think represents the logical form and there is evidence by the way presented by paulier estandiana and collaborator and talking about paulier because he's teaching a class at our master now we suggested this area is the activation of this area varies according to the syntactic complexity of the sentences that enter this area right but then this area is not is not necessarily a linguistic area so what is what is this is that really logic is not necessarily a linguistic process the fact that is mediated by via language is because we use language but the really interesting question for me would be supposedly give you my sins the sins that have no information about and you see them and do you represent the sins by using a part like we can say we say a particular logical form and does this logical form for example what happens in the brain of a 90 months old infant when he sees this machine in b.a.45 right maybe this this is not a language area but is a as it were language of thought area in which we're talking about it what about babies in preverbal well that's you know help us what can I say I mean that's really what you would like to do to just to figure out what is the network that is involved in seeing these scenes that contain potentially some logical inference in non in pre-linguistic infants and my obviously are going to have a lot of activation here but the point here is do you also activate b.a.45 can you show that what infants put into their memory is some abstract representation that has some sort of logical structure which then maps onto b.a.45 because that's the syntax area not of natural language syntax but of thought and yeah these are the questions that we are trying to find the way to work on okay I have a final question for you Luca which is I guess that there are no cultural or social differences among the infants right well there certainly are and we are unable to to pick on them because a change and this is a problem for a lot of our research we cannot select who comes to our laboratory and I mean these are people who come because they basically are either very nice or have a lot of time or our friends right that's it so it would be really important to figure out for example whether the ability to solve the problem in our sense is mediated by the situation at home but we don't have the power to do this kind of studies and the very last one why do you decided to make research about infants why do I decide to do it with infants because I am essentially masochistic that's the the short answer but the long answer is because I think that if you really ask yourself why did what pushed these guys who had a lot of trouble like for example surviving in the cold to paint it's because there is some internal pressure to create right and this creation is not random this creation is very organized it's very organized in the painting and it's very organized in the problem solving and what I I'm fascinated by is what is the pressure that leads you to actually create and in in the very elementary sense of creating like finding a solution to a problem in the Zildersende that's why and clearly this is the thing is universal what are the basis of this and one way to answer is it's a methodological way is to just go as early as possible and figure out what kind of mental processes infants can do before you can tell them anything that's a good basis not necessarily the conclusive basis but a good basis to suppose that we are talking about something which is the same kind of pressure that led people in the last case to just paint what they were painting thank you before closing the session I have an announcement which is the next seminar will be the 23rd of February and will be in the person in charge will be from the town okay thank you very much look up for your