Francesco Casetti - Why our conversations with AI matter

[00:00:10]we can get started. And so it's my pleasure to do that uh to do that first gesture um and to welcome back uh Franchesco Casseti. I say welcome back because he joined us in Bergkamo um last year for a a very nice talk and in our launching effort there in Bagamo. So it was it was really it was a great pleasure to meet him and so we're we are really happy to have him back for this talk. Um for those who weren't there not familiar with him, let me just uh recognize his very uh honorific title.

[00:00:49]Um, if I may, that's he's the Sterling Professor of Humanities and Film and Media Studies at Yale University.

[00:00:57]Um, he's published quite a lot. He's been working for as a prominent scholar in in media studies, film, television, visual arts. Um, he's he's a I think [clears throat] very very accomplished um theorist in this area. I just um going to note a couple um a few texts that he's written. There's one called Inside the Gaze, which is quite prominent, goes back to 1999, published by Indiana. It's one called Communicative Negotiation in Cinema. Uh Va Vita Aensier in Milan published that.

[00:01:35]Um it's actually negotiation in cinema and television. I'm sorry. Um then there's a book entitled The Eye of the Century, Film, Experience, and Modernity published at Colombia. Another Columbia, the Lumiere Galaxy, seven key words for the cinema to come. And then a last one, and I forgot to note the title uh that the publisher, but it's entitled uh [clears throat] screening fears uh on protective media. Uh oh, no, actually that was that was published at zone. All of these, you know, very very significant contributions. And um and you'll hear this background work in his presentation. I believe there's a certain uh uh fullness to his his approach to these matters. So um I I'm really happy to welcome him and uh to pass the screen to him. Franchesco, thank you for joining us again.

[00:02:28]>> Thank you Chris. Thank you also Nemania for arranging that. It's an honor to me to go back to the graduate school, European graduate school. And today I will I will in I will present um something I'm working on. So it's is still a slippery territory to me in which I'm trying to understand the way in which uh uh artificial intelligence is speaking with us is com making conversation with us and on on the horizon there are a number of questions that uh Chris, it's it's it's everything is okay. You can hear my voice. Okay.

[00:03:25]>> Yes, it's it's coming through very nicely, Franchesco. Everything's good.

[00:03:28]>> Yeah. So, on on the horizon there are looming a number of question that I list them. We are not fixing all this question today not answering all the question but in the work in progress I'm trying to work now they are very very strong uh poll stars first there is a media problem how artificial intelligence is interacting with humans and then there is a semiodic problem.

[00:04:05]How we can describe this interaction?

[00:04:10]There is an epistemic problem. Is this interaction similar or different from the interaction we have in our everyday life? And there is a peripheral problem. Why and how we trust a machine during a conversation as much as we trust or we distrust somebody in our facetoface conversations.

[00:04:41]And finally, there is a political problem that to me is super important.

[00:04:47]What kind of power artificial intelligence has to us? This is the horizon of the problems. But I start immediately with one uh one uh um one image.

[00:05:06]This is uh can you see every everybody can see on on on the screen the image.

[00:05:13]This is the landing page the original landing page of CHBT.

[00:05:21]And still you can find it. But now there are several var variant variants of the same landing page.

[00:05:30]And uh what is super interesting is the question and what is doing this question basically is doing two things. One is establishing an axis IU.

[00:05:51]Through a question, there is an annunciator and annunciate.

[00:05:59]There is an addressor and address IU as communicative axis.

[00:06:07]But second uh while defining let's say the subject of the discourse this sentence also creates what the good semiodics by gass uh was saying is to establishing two thematic roles.

[00:06:33]The I is associated to somebody ready to give care to somebody else and the U is associated with the idea that the U is somebody in need.

[00:06:51]So there is absolutely strong definition of the parties in this conversation.

[00:07:01]And let me say another problem which is super important.

[00:07:06]When in the following step the user is taping his as as we do is is typing his question his own question addressing the artificial intelligence. Can you answer me? Can you copy copy edit my my text for example as I do normally?

[00:07:30]What happens?

[00:07:33]The eye of the uh um of the artificial intelligence uh um of the artificial intelligence becomes you and you addressed by the artificial intelligence become I. There is the reversal of the axis.

[00:07:57]But this means but there are the stable of the thematic theatic roles.

[00:08:04]Artificial intelligence is still the provider of care.

[00:08:10]But this reversal gave the artificial intelligence two pronouns I and you. And accepting that I give the artificial intelligence the status of subject of persona as Bvenanist was used to say the role of persona that assimilate artificial intelligence to the persona of our everyday conversations.

[00:08:42]This question is not innocent is load of consequences.

[00:08:50]An axis an axis i.e. you, the assignment of a tmatic role, the possibility to create this uh this I and you as both as humans.

[00:09:12]And finally, the content of this question, what the agent says, I'm ready to give you care. there is a dedication rather let's say an imposition or an act of force etc. So this is one landing page, one simple sentence, not innocent.

[00:09:38]Let's see another landing page.

[00:09:42]This is a landing page is no longer active but has been active for two year for three for two years and a half by a company.

[00:09:55]Who are you? There is not a question mark but nevertheless is what in semiotic is called an interpolation an address.

[00:10:07]Once again there is a sort of axis I you.

[00:10:16]But this time this axis is quite more interesting because the you who are you is both the user of this h but also the name of the app.

[00:10:33]So there is an ambiguity.

[00:10:36]Who are you means who are you user and who are you up this means that the you in this sentence is in a way exploding on one hand it's connected with the source of the message the I and you are connected I do not know you stop me if I'm not clear but this You you come is I it becomes also you and there is a we which emerges. But on the other way on the other side this you who are you?

[00:11:20]Whoever can be you. Even me I can be you.

[00:11:26]Whoever can take the position of you.

[00:11:31]And uh what is the effect? Is the effect that uh this sentence make clear that it is open to everybody independently from where he is or she is.

[00:11:52]What is important is everybody who wants to fill the position of you.

[00:12:01]There is a multiplication of this you there is a multiplication of users and that means that what is addressed is a population not a territory. I'm not speaking to Italians. I'm not speaking to Americans.

[00:12:19]I'm speaking to the you who want to jump in and recognize themsel as you.

[00:12:29]Let's see another.

[00:12:33]This is Perplexity.

[00:12:36]Perplexity is still uh in use. It's still a popular uh uh app.

[00:12:46]And uh what uh this landing page also um also perplexity but is you know studying studying the way in which artificial intelligence is is is trying to com make conversation with us. It's funny because they change every six month they change the landing pages. So I have a an archive of landing pages. This was until last year. Now there is a different landing pages and I can speak about also that why they change etc. There is a logic in the way in which they change.

[00:13:29]But the kind of conversation is still the same.

[00:13:34]And also now perplexity is not no longer asking what is perplexity but is presenting itself and is addressing the user. And what is perplexity?

[00:13:49]And uh uh it's uh it's it's very interesting because the definition of perplexity you can read I I do not know if it's big enough but you can read perplexity is your artificial intelligence power the Swiss army knife.

[00:14:12]It's quite surprising. I do not know if everybody knows what a Swiss Army knife is. is the kind of nice with dozens of little utilities attached.

[00:14:26]So what is here?

[00:14:30]What is the sense of this sentence? Why define if I define myself in speaking with Will or Lexi or Setare or whatever you want.

[00:14:47]If I speak and I introduce myself as the Swiss Army knife, you do not have the idea that I'm ready to hit you. You have an idea that I'm helping you in different ways that I'm ready to help you whatever is the task you have to accomplish.

[00:15:14]And there is tons of task and but each one has his own task to accomplish.

[00:15:26]So like the soldier who has in his pocket the Swiss army knife and then he used that for every purpose and the entire company can use the Swiss army knife for different purposes.

[00:15:49]They are associated but they are also dis they distinguish themsel each one from the fact that they use one of the utility not the other I do not know if I'm clear you stop me if I'm not clear so thinking to that presenting itself as a a Swiss army knife means that I'm taking care in multi-way in a multi- manner but I take care of each of you according its own need. I don't know if it's clear which means here we are that the specificity and the multiplicity of the Swiss Army knife means that artificial intelligence in this landing pages is addressing not the user in general But each single user, each single user deserves the care and attention that he needs.

[00:17:18]There is extraordinary extraordinary work of what uh um what uh um what is called a process of individualization.

[00:17:34]And Fukall called that exactly a process of not speaking about artificial intelligence a process of individualization.

[00:17:44]I define each of my user according the specific needs she or he has.

[00:17:57]This means a little change in the logic of the conversation.

[00:18:05]And I put this way if I am dressing in this moment there is a number of uh person I mean online listen to me.

[00:18:18]But if I address the audience, I'm addressing a plural you.

[00:18:28]But if I start addressing to setare, for example, or will or lexi singularly, this U becomes a singular U.

[00:18:43]So in this uh uh in the perplexity uh um landing page and also in the current landing landing page perplexity is conver is trying to make a conversation with somebody which is not only a generic audience but a specific user.

[00:19:11]There is a process of singularization or individualization that is made here. Which means for example that the axis is no longer I, you singular you and plural you or in old English I, thou, you or in Italian and the other languages there is the second singular person and I can say for example in Italian to vo.

[00:19:58]So it's it's a way of articulating the subjects is a way of defining an attention not to a generic audience not a generic population but a specific single user a specific individual at stake.

[00:20:26]Let's uh go to another if things are not clear you stop me and I go back and etc. But there is we can ask more during the let's read not a landing page but is the page number four uh when you turn on uh another uh bot cla it's a bot that uh I love uh particularly it's a very useful is is a good uh is a good company or at least it looks like a good company.

[00:21:12]And in this uh when you turn on the landing pages there is uh immediately uh in in uh in the first in the in the landing page there is a frequently asked question. You uh click on on that and then you goes to this page. You click on frequently asked and that is is the third move. Is that what happens?

[00:21:50]And uh what he says which is uh absolutely he says I can help us think. I can uh do many things. If you want to dream, I can dream with you. Which is big big task.

[00:22:12]But at one point, if you can read the sentence, he uses a word which is quite uh quite impressive. He say I can coach you.

[00:22:26]What does mean to coach somebody?

[00:22:31]is no longer just to give somebody a thematic role. It's not just saying you're somebody or someone who needs my help.

[00:22:46]Is like I'm speaking with Setare and I say Set I need to coach you. Are you ready to coach you? And says, "I need that and I cannot fix the problem. I have to give setare the competence to fix her own problem."

[00:23:11]This is a important passage because in good semiotic when I assign aic role and I say you are going to somebody you're going to be somebody in need.

[00:23:29]I program you this is the term used used by semiotics.

[00:23:36]I program you to be somebody, someone.

[00:23:42]But if I coach you, I not programming you to be somebody.

[00:23:54]I'm manipulating you in good semiotic language. I'm manipulating you in order to make you do something.

[00:24:06]There is a passage from making someone be and making someone do.

[00:24:18]And that is crucial.

[00:24:21]This is a strong passage in which uh artificial intelligence is not just fixing our problems is coaching us to have the ability to solve or to fix our own problems.

[00:24:40]And uh once again in good semiotic language when you are ready to do something you are ready to perform a task means that you have the means to do that. The intellectual means to do that and intellectual means to do that is called in good language competence.

[00:25:07]You need a competence in order to perform something. That's clear.

[00:25:14]So coaching means that artificial intelligence is ready to give us not just to fix our problems but to give us the competence to fix our own problems.

[00:25:32]And I do not have time because it's uh uh is is less than an hour. But if you go and to go in depth in this kind of conversations and you can experiment with a free access AI supported app. It's easy. You can take lot of chat bots are for free immediately. Then you have to pay a subscription.

[00:26:04]If you go there, you can understand that most of the work is not only to do something in order to make you be something, to be happy or to be safe.

[00:26:21]But is transmission of competence in order to you fix your problems. This is what is called an educational tool.

[00:26:35]There is a pedagogical force. But the pedagogical force is transforming a tool.

[00:26:45]Is not something that is something that help you not to be something but help you to do something.

[00:26:56]And if I can do something, it depends on the competence that I get from artificial intelligence.

[00:27:03]My agency is depending, this is the message, my agency is depending on artificial intelligence.

[00:27:13]I'm not only a subject that artificial intelligence wants me to be.

[00:27:20]I'm on also a subject able to do what artificial intelligence enable me to do and that is a crucial passage.

[00:27:35]So coaching from a thematic role to a competence and this is the maximum of the idea of taking care of uh an user.

[00:27:53]Let's go to the last uh uh page.

[00:28:00]This is meta the artificial intelligence of meta terms of service.

[00:28:08]In my research I read a lot of terms of service and they are super interesting.

[00:28:15]Generally we sign them we we just put a a little sign on the box. Have you read do you agree with that? And we do not read that. But read them. They are super interesting.

[00:28:31]Once again, they are not innocent.

[00:28:35]They are quiet, smooth, friendly, but not innocent.

[00:28:41]And here, if you read uh the terms of service of Meta is free.

[00:28:49]You can have the entire four pages is not so long, but he's super interesting.

[00:28:56]First, he shows they show the the the terms of service. They show an incredible sense of responsibility.

[00:29:09]Say I take care of you and I take the responsibility of caring you.

[00:29:17]But this is this is made through a process.

[00:29:26]The user is alerted about the bad uses of artificial intelligence. Don't do that is a negative competence. Don't do that. Don't desire that. don't make this etc etc etc. So the responsibility of artificial intelligence is at the expenses of the action of the user.

[00:29:58]How can I summarize this aspect?

[00:30:05]You can see here is the summary of this uh uh uh uh terms of uh uh use or terms of service responsibility rather than possession and the idea that uh there is a care for the user and a care for the service.

[00:30:25]But let's go this car this this this this conversation what they say to us what they tell us they tell us that artificial intelligence has a certain kind of power is a power based on carrying us but nevertheless is a power which in implies a strong definition of us as of users as users.

[00:31:02]And uh my hypothesis is that this power is very similar to the so-called pastoral power explored by Fukall in his seminar of 78 which was a fantastic form of power which was born 1,000 year before Christ in Palestine, Egypt etc. It was I mean summarized by the metaphor of the by white pastoral because the idea was this power is based on caring the same kind of care that the shepherd gives to his flock.

[00:31:59]So it was not a system of power based on force but on dedication not the care was not on the territory was a population.

[00:32:12]The attention was to each and all the members of this population.

[00:32:17]It was a mediation between needs and resources and it was a strong a strong power based on responsibility.

[00:32:30]Is this this does this mean that the artificial supported artificial intelligence supported media today?

[00:32:44]They have the same power as the pastoral power or they are are the ultimate uh uh reincarnation of the pastoral power.

[00:32:55]Yes and no. Yes. for many kind of reasons but no for one single reason what they do they extract systematically the data the bio data of the user all the question all the conversation we have seen they are I'm the capturing the data of the user and I show you something which is even more impressive This is replica.

[00:33:30]Replica is uh technically is forbidden in in Europe because because many ethical question but is in America is works well etc etc and uh replica is very clear from the landing pages it start to ask all the data of the user the name the gender the birthday etc etc etc and then it goes it goes start to ask and then replica is famous because what it wants to do it's it provides a companion and here the companion and choose your avatar you can see no on the page choose your avatar and Then replica ask the avatar the age the the gender the birthday the whatever you want the name etc etc etc. So he's digging not only the bodata of the user but is digging for also the bod data of the avatar which means is digging for theired data for a companion.

[00:35:05]How can I possess you more than I do not only know who who you are but also I know who you desire to be. It's a complete possession is a total possession.

[00:35:25]And ultimately after that there is a conversation in which if you go and you can experience yourself in which absolutely what happens is exactly the capture of the fine data regarding the imagination and desires of the user.

[00:35:54]So a counter hypothesis I supported media not only they are parallel to or they have a certain kind of pastoral power but also they are indeped with the algorithmic governamentality which is the logic of the algorithm is more data I have better I work and more that I have of you more I possess you better I can help you.

[00:36:31]So my conclusion so far in a conclusion the conclusion the conclusion conclusion comes in a second arrives to the idea that this conversation are a trap.

[00:36:44]They try ultimately to capture us.

[00:36:48]Not only they define who we are and define what we can do but also they define who we are in our data as let's say possession of artificial intelligence.

[00:37:09]It's a little bit uh uh let's say uh sad conclusion.

[00:37:16]If artificial intelligence is a trap, why use it? My answer is there is form of resistance and I end up with a little example of resistance.

[00:37:33]A few months ago uh a guy Harvard student invented the friend friend which is you can see here a brooch that was able to listen all the conversation of the owner and then to suggest what the owner was ready to do or was the best to do for the owner.

[00:38:01]and uh to launch this uh uh object.

[00:38:10]The subways of uh New York were flooded with this kind of announcement.

[00:38:18]What happened?

[00:38:21]It did happen that uh uh the New York citizen started to write on the posters condemning friend friend insulting friend insulting artificial intelligence and resisting the lure or the attraction of this object. subject and uh that that's strong. somebody is really strongly attacking to the point that the owner of friend say oh no it's good I'm I'm a liberal um you can insult me etc etc but the effect that he had to close the company it was a failure so is a little example of resistance irony cracking, hacking, you can go against the art, the algorithm.

[00:39:35]You cannot necessarily obey the algorithm. There are more and more analysis how ordinary people are cracking the algorithm in order to make room for what they are really without being nei without surrender themsself to the artificial intelligence. Thank you.