Héctor Pérez Urbina

I will say to learn and to become comfortable and familiar with discomfort, right? That cognitive dissonance that like, don't get it. Like you're reading a paper, you're reading something new and you're like, what is this? Like that feeling, right? You need to become

more comfortable with it. You need to almost, almost appreciate it. Right. It's almost like it's almost when again, the gym, when you're going and you're failing that terrible, wonderful feeling or failing because that's precisely when the gaining is the most prominent.

Welcome to Taste Bench. I'm Bruno Marnett. My co-host is Philipp Zan, and this is a podcast about taste and the creative side of AI. Our guest today is Hector Perez Urbina, an AI practitioner and researcher who spent over a decade at Google working on responsible AI. He holds a PhD in AI from Oxford and a master's in AI ethics from Cambridge. And he recently founded the Tlanextli Group, a public benefit LLC, with a mission to promote the judicious use of technology.

and help individuals remain the architects of their own thinking in the age of AI.

Welcome Hector. thank you so much for talking to us today. We're very very excited to have you on because you're you're someone who comes from quite a technical background, both in AI and in in tech in general. You you work for many years at Google. And now it has been a couple of years that you've you've really specialized in this area that is known as cognitive sovereignty. And yeah, so we're very curious to dig into what else

kept you busy recently. and maybe if you could start with a if you could maybe just give us a very short introduction of what basically what is cognitive sovereignty. and then maybe we can, you know, discuss potential connections to to the topic of taste that we are covering in this podcast.

Sure, Cognitive sovereignty is something that is very, or at least in my circles, is very trendy right now because it is discussed in the context of AI. how cognitive sovereignty could be understood without AI, without technology. Cognitive science has a long tradition, et cetera. But cognitive sovereignty in the context of AI is a response to the threat of, or the harm.

potential harm of cognitive surrender. This is a type of cognitive offloading. So cognitive offloading is basically making your life easier. For example, if you take a note, for example, if you use a calculator, if you do something like that, you're offloading part of your cognitive load to something else so that you can be more effective. Cognitive overloading can be very beneficial. Like using GPS is cognitive overloading at its best. That is great.

Cognitive surrender happens when you jump in into a driverless car. Like you forget about the whole thing. You just sit down and enjoy the ride. Like it's basically that is some sort of like difference between offloading and surrender, right? And it can be bad. It could be good. You know, it's typically very fast because, you know, surrendering something to AI, it might be beneficial for some.

some use cases. The problem is, of course, when it starts affecting your autonomy, your agency, your liberty, and your own judgment. When you offload too much, and then cognitive sovereignty is basically the response to this. It's basically saying, how can we use AI, but technology in general, so that we remain the ones driving?

we remain the architects of what we're trying to do. And yeah, so that is in general terms cognitive sovereignty.

that makes a lot of sense. So so you said in so in in in some ways this is not a new science, cognitive sovereignty, because it predates Chat GPT. but I'm curious what of in the l in the last few years there must have been a lot of research specifically about LLMs and the the way they interact with the brain is arguably different from, you know, Google search, right? It's more interactive and and there's all this aspects. I'm curious from what you've seen so far. What what are the kind of

I don't know, the first lessons that are emerging from the research specifically on L L Ms.

Yeah. So one thing that I think is quite worth noting, like it's that there is this very famous model of cognition by Kahneman, the system one system two like model of cognition, right? And system one is kind of, you know, immediate. It is very, there's stereotypical automatic, right? Reflective intuition. System two is more deliberation.

reflection, reasoning, et cetera. There is a new model that has been introduced very recently this year. And the model now has another system, the artificial part. it's system one, system two, the Kahneman model, and now there's system three, which is artificial, right? And there's different ways of these things interacting with one another. Cognitive surrender is what happens when you start with one,

and you delegate to three by passing two. So basically, you start with some sort of intuition, delegate too much again to the artificial part. It does the thinking, it does the reasoning, it does the solving for you. And basically, you respond. You adopt its response as if it were yours. You basically say, I'm so clever. And basically, I just copy paste. I become like a rubber stamp. Cognitive offloading.

or cognitive, that is the surrender. When you have the offloading, which is beneficial offloading, is basically you start with one, you go to two, two remains the main thing, and then you use three selectively judiciously in order to do certain things, but two is the one that is still in control and basically you're.

Right. So that is, that is the difference. And of course you could go to automatic pilot. You could go to three directly in this is agentic AI in many, in many cases, right? Basically you just let it do what it does, take notes and write me like a summary, right? I don't want to, I don't want to be involved. I just want to, you know, take, you know, have the result. and of course you could bypass three already. And that is the traditional way of cognition. But so that, that is the first thing that I think is really important because it is like a new.

theoretical kind of model as to how to understand what's going on. And then, you know, different paths, different things, how can we study them and how can we actually tell if they are happening? And it turns out in this study, this is this is by Sean Nave, by the way, the paper. It turns out that this cognitive surrender thing is a thing is like a phenomenon that they're seeing. And basically what what

it means in practice is that people rely on AI and they adopt AI's responses, whether they're right or wrong. Right. And that is kind of like the signature of cognitive surrender. If AI does well, we do well.

Would you?

it does poorly, it does poorly. Yeah, we do poorly. Was that, Yeah, please.

I have a follow up question. Yes, sorry. I have a question on this.

So would you agree that I think the interactions just talk about chat interfaces, the interactions with them have changed in a sense that there is much more follow up prompting from their side to the human users, which would directly lock into what you said, right? It's like, you know, you ask a question and then, Hey, should

Yeah.

I give you this la la la on top? Which seems to me like it's exactly this feeling into the chump from one to three. Do you agree? Do you see this more widely in?

and other capabilities of AI.

You mean the follow up questions that chat bots are usually asking now? Yeah,

Yes. Yes.

Yes. I well, I think it depends on the type of question or follow up, right? Because sometimes that could be the invitation

or the start of an interaction, which is very system two oriented, right? If it is challenging.

If it is asking you questions, if it is asking you for your opinion or your thought or whatever.

Mm.

But that could also be the start of surrender. And the problem is that the difference between of loading and surrender is very blurry. And it's a slippery slope. And we don't even notice when it's happening. And on top of everything, a lot of us operate under a lot of pressure, environments with lots of pressure, students, workers.

in our day-to-day lives. And so, of course, it is very tempting to go for the surrender because it is easier, right? Less friction. But yeah, I really like it when there is some sort of interaction, right? Interaction in general, would say, leads to my system two kind of turning on. However, it doesn't necessarily have to be that

When that's a very good point actually. The one critique people make very often of ChatGPT is that it's trying to ask questions to just do more for you. In a way it's like even more control over your work, right? Which is let's say you're writing an essay and you only ask for I don't know, let me help me a bit for the introduction, and then it's gonna try to jump into like can I just finish the entire essay for you now? Right? So it's almost kind of

incentivized in a way to kind of keep you hooked on the screen in the same and and have all this sort of in all this all this bias we've seen in in other kind of tech products in the past, right? is this yeah, th that's the the tension you were hinting at, right?

kind of why I was saying that it depends on the type of questions, right? Because of course, it could be very kind of, you know, it could lead you to a certain to a certain outcome, right? Because it could prevent you from figure out yourself. What is next? Right. Because someone is suggesting that.

And of course, when you have we know this from brainstorms that usually the idea that gets the most attention is the first idea that makes sense, not necessarily the best idea. Right. So we latch on to that thing because by definition is going to make some sense. And we are going to we are trusting it more and more progressively, right. Increasingly,

as we use it more and more, we trust it more. And so if it suggests something

you're going to be inclined to trust its judgment and say, well, yeah, maybe that's probably what I should be thinking about next. Right. I suppose to you be thinking about next. Right. So it's a it's a dangerous kind of situation because it can lead you to a specific, you know, to specific solution, of course. Right. It could prevent you from developing your own thing. And this is related to taste. Right. Of course, because it can.

It can influence. It can influence what is it that you value? What is it that you think it's important? And what is it that you're going to actually do? And then this, course, can be quite useful in case you don't know what you're doing. Right. I mean, it can give you some guidance. It can give you like, well, what is a business plan? Well, OK, you know, a business plan is this. And you should know. But of course, as the stakes get higher, right, then

then your own authentic thought and your lived experiences and your own, what you bring to the table is the thing that is going to make a difference. Is the thing that is going to you distinct, that is valuable, I think. Because by the end of the day, a lot of your competitors or a lot of your colleagues have access to the same tools.

If you ask these tools a generic question in the usual path, it's going to give you precisely that because these things are statistical echoes. There is a threat or a potential risk of a homogenization of expression, homogenization of thought.

Hundred percent.

homogenization of taste, right? Because like, look at LinkedIn, right? Like, we're starting

Yes.

to notice now there's like some sort of formulas and they're sort of like a little and you, I, don't know exactly where it is, but we kind of can feel it that this is like, is, you know, this is AI and, and, and that is sad, you know, that is just sad and tragic because of expression, but it's not only that it's also

It's also, I think, dangerous, right? Because diversity makes us stronger for many, many reasons, right?

And it's and like you said, it's not just the style, it's it's actually the content, right? That I think that's where people are starting to realize the real danger is that it's not just short sentences with dashes. You know, the problem AI is a thought partner that influences your thoughts. Go on Philippe, yeah. Yeah.

Yeah, you I was using AI when I was doing

my master's for AI ethics, right? And that was an interesting experience because, even I, as a mature student, you know, doing a course on AI ethics, I felt really tempted to do the, you know, to delegate it too much because I had stuff to do, right? Like life was going on. I tried my best not to do that, but of course I did see how certain ideas

kept coming up. Like it kept suggesting.

the same sort of thing. And I was like, I've

Yeah. A hundred percent. Yeah, yeah.

Yeah.

already told you that, you know, I don't think so. Right. Like, you know,

Yeah, yeah.

and I had to fight it. And of course, when you fight it, of course, this is brilliant. And you're like the most, you know, intelligent person ever because they're sycophantic, which is another problem. But

Yeah. Yeah.

someone who might not feel ready or in a position that they can fight back like a novice learner,

Mm-hmm. Yeah, yeah.

a student or a junior engineer, say,

Right? They might be inclined to just take it, to just accept it and to basically do not even challenge it. Basically just accept it and

Yeah, yeah.

also adopt it as their own. And that's on top of everything. You believe that this is your idea and you believe that this is actually your authentic thought when in fact this is precisely what millions of other engineers are suggesting to their boss.

I have a, I mean, actually several follow up questions. Let's see how far we get. Let me start with the first. So you raised already this point, questions are, can go either way. If you are, if you were free to design the user interface flow, what would be the elements that you would bring in that could avoid the stepping over from one to three?

That's a good question. know, when I was at Google, there was this project called the neutral point of view, right? And basically it had to do with sensitive topics and how and in which ways models should address and should tackle, should talk about sensitive topics. Let's say homosexuality in Uganda, right? And it's a very...

complicated question because on the one hand, if you impose your own values onto some other society, right, that could be construed as some sort of colonialism, data colonialism, right? Very, very true. On the other hand, do you really want to

you know, exacerbate an issue that you feel strongly like human rights or something like that. Right. So it's a complicated thing. Now, this is this is, you know, related to what you asked, because I do think that that LLMs should not only give you some sort of like, should I be working on this next kind of interaction, but like a taste of a taste, a variety.

of different possibilities,

Mm-hmm.

right? Like, or even so, but also maybe asking your opinion next is just, what do you think we should do next? We could go this way, we could go that way, we should cover all of this stuff, but what do you think is most important? Right? Like that sort of thing immediately by design activates my system two, because all of a sudden I'm asked something that I need to reflect on and that forces me.

as to think, to consider, to respond, as opposed to basically saying, yes, yes, yes, yes, accept, copy, paste, right? So something like that is more like it. Now, I'm not convinced that the chatbot interface is the best interface for this sort of interactions, right? Like, we know a lot about human computer interaction, and we know a lot about how certain interfaces are, you know,

conducive to certain behaviors and to certain things, right? And so just to have like a little prompts or maybe like a conversation even or something like open-ended conversation like that, not sure is the best kind of interface, especially if you want to, for example, if you're interested in learning or cognition or something like

I was perfect. Next question here. How should interfaces be different for people who are already experts and are using basically, as you said, like an AI as a thought companion or basically as a support system versus someone who's starting out and wants to avoid and following the trap of believing these thoughts are their own.

Yeah, think that is a very good difference, right? Because it is not the same when you use an AI and you happen to be an expert in the subject or you happen to know what you're talking about. That puts you in very good position to discern whether the AI's outputs are good, right? Basically, can tell

Yes.

and you can push back, right? Sometimes the AI was telling me something that I knew was completely false and I was like, you know, what?

are you saying,

Yes.

Right. But, know, typically when you are in that kind of, use case, you don't have time or you don't you don't care for the sort of, you know, coachy questions, socratic questions, things,

Yeah.

right? You just want to get it done because I mean, I know. So just just, you know, get it done fast. Right. It's almost like

almost like surrender, but with a verification at the end. So basically, get it done. I set up a few agents, and I tell them to do work overnight on this problem and interact with each other, and you have different roles, and get it done. This is the solution. These are the constraints. And I expect to see two different implementations of this thing. Go. And then.

Then I don't want, do you want to do this? I don't want interaction. I told you the instructions and go for it, right? And then the morning I take a look and with my expertise, I'm like, great, no, what the hell? Yes, right? That's very different from a junior engineer that doesn't actually know what's happening. That is very different, right? And right now, as I keep pointing out to the interface, right now, the entry point seems to be the same, right? Seems to be the same.

And that is just wrong. And of course, you could do something with prompt engineering. You could have some sort of gem-like treatment beforehand so that you could give, or novice learners, could give some version, Socratic version, like Claude-like thing, and you give the others a more like a co-pilot thingy or open AI thing. You could do that. But I truly believe that you should actually be doing more.

in the interface space.

Interesting. Related to that, so there's there's a question of interface, but I'm also kind of curious about the I guess the game that people are playing. So maybe to link what we discussed also with the topic of taste, as we as we've discussed it before with Philip. I think there's some some professions. Let's take for instance the legal sector.

Where there was kind of there used to be kind of a natural trajectory for people, right? You start as an intern, a very junior person, and for the first many years of your career, you would actually struggle with a lot of kind of low level work that can be quite painful, like reading a lot of documents, doing kind of laborious work. And over the years, you know, having done this work would contribute to making to teaching you like really good instincts about you know what a good document looks like, what you know, like a dangerous contract looks like and so on.

And then eventually you would make it to to partner if you're lucky. And then you could delegate. Then you could have junior people that you just give all all the the easy work to, right? And now what one one issue we keep hearing is happening in some law firm, at least, is that they don't know what to do if they they don't know how this process is gonna look like now because the junior, the kind of typical Gen Z hire today will actually take so much shortcuts, they would just delegate so much to to Harvey or you know Chat GPT equivalent.

That they're they're not going through this I guess struggle, you know, the

to the I guess the pain of, you know, like going through all this experience, reading all these cases and studying all all of these. so so you you said already like sometimes people take a shortcut because they have to, but I think one of the questions that I guess we have to ask ourselves either in public education or in in companies is how do we actually create a system where people still have to struggle, right? They have to go through the exercise.

I I wonder if that's something that that that that that you've also looked at. Like w what would be some ways to make sure people still make the efforts basically?

Yeah, so in the context of learning, learning science, there is this notion of desirable difficulty. Learning has to be intrinsically difficult because translating from short-term memory to your long-term memory and creating all the schemas that you actually need to recall and blah, blah, that is learning. And that requires some effort. And that is, why

it our brain prefers system one because learning is difficult system to the liberation, etc. is difficult. And so that is why it is it is very natural for us to prefer something that is not, you know, doesn't have a lot of friction. Right. So it is it is not surprising, right, that people are taking shortcuts. People's will people reasonable people will take shortcuts when they can. And so

Yes, like how do we decide? This is a major, major question for me right now in the company when I work with high schools and we do workshops with the faculty and the teachers and the students. Like how do we design assignments and homeworks and projects where learning is not by passable? Right. Like you have to, you know, it's it's a matter. It's a complicated question, right? But the idea is basically to basically.

to move away from the result, the essay, the test, et cetera, as the main thing, to the process as the main thing. So they have to show artifacts as to how they got to that result. And I recall when I was interviewing at Google, and I was interviewing myself, a bunch of people later on, if you had a brilliant person,

And they just stayed silent for like five minutes or two minutes. And then they give you the perfect answer. Just, you know, that is not as valuable or that is actually not valuable. What I need to see, I need to see what they're thinking. I need to see what they're saying, you know, where they're going, what they're asking themselves, what things they're trying, what things didn't work, why not like all of that. Right. The process as opposed to the 42 answer. Right. 42. Yeah. Great. And so

That is the challenge, right? And I think the challenge is there's different challenges in different scenarios. In a school, for example, you can say, as opposed to asking students to create essays, I'm going to generate an essay, an AI-generated essay with some sort of flaw, for example, right? And the students need to actually find the flaw and tell me why this is like a misunderstanding or something like that.

The issue with those sort of approaches is that teachers are quite overworked, right? And of course,

Yes.

if you suggest this to someone at Oxford, they're going to say, great idea. We'll just implement it shortly, right? That's kind of what we're doing anyways. But if you go to UCLA and you have hundreds of students per election and stuff like that, well, it's very difficult to implement something like this. So that is kind of why this is very challenging, right? Now, in the context of a company,

Companies are very focused on efficiency and AI is being sold as this key to efficiency to make things faster and better and just faster. so there is little time to... This is something that is not as valuable or valued. And I think it should be more valued. You should give, especially novice learners, junior people,

a bit more space, a bit more time for them to actually learn, right? And have some sort of interaction where senior people are transmitting knowledge, or because they're teaching as well, they're junior folks, but in a way that respects the human tempo right? Because if you ask them to deliver as much as companies are asking them to deliver, there's no room for learning.

There is no room for learning. And not only does this happen, not only do you not learn, but your current abilities and knowledge atrophies. So it's like a bad situation in a few years, right? Seniors are gonna be great, but juniors are going to be stuck, right? And when seniors go away, you have a big problem.

Mm. Yeah, I think that's that's a real concern for especially kind of aging firms or you know

No, let me just

share a little experience. I was invited to give this talk about cognitive surrender at a big bank. And I was a bit nervous because that big bank turns out is one of the main or the top employer for data scientists in Mexico. big bank. And I didn't know they employed so many data scientists. They had a huge AI division and they were like top notch, right?

So I was a bit nervous because my message was more of like, the times are urgent, let's slow down, right? And be careful and take care of this and, know, efficiency, yes, but the efficiency track careful and all of this message. I was a bit careful. was bit nervous because I didn't know how the reception was going to be, but they were so receptive. Not

only the junior people, the junior people was obviously, you know, concerned that this may be happening to them, the atrophy and the stucking.

But the executives were also very interested in this particular topic because they see that they will have to deal with this problem. So yeah, it was really refreshing to see that the attitude is changing, I think. In schools, I was expecting that because people, teachers care about students, development is normal for them to care.

But in

Yeah.

companies, in the industry, I'm also seeing this more and more, that people are starting to care about these things because it's going to bite them back.

I'm I'm actually curious on this topic. The I

I guess one thing that would worry me a bit in the industry context is that usually the there's an obsession for numbers, right? In industry you try to have metrics and you try to, you know, measure everything. And I worry sometimes that you know cognitive sovereignty might not be the easiest thing to put a number on. but maybe I'm wrong.

Yeah.

Maybe there is actually progress in that in that space. Like given that you're looking quite closely at it, do you think there is

I I guess what what are the best metrics you see to to kind of convince a company that they have a problem of cognitive sovereignty or not?

Yes, that's a good question. I would try to get their attention. And we know, for example, that AI empowers people to create a bunch of different things. Like now, anyone can just create a demo of some sort of app or some sort of idea that

god, yeah.

they have. They can basically

you

make it more real very quickly.

But part of the issue with this is that everyone has access to this technology and everyone is doing this. And so this creates like a bottleneck. senior people need to integrate and they need to actually see through all this noise that for the most part is the same noise because a lot of people are just taking what the AI suggests, of course.

And they need to integrate it. So in the end, the productivity gains are not as big as you might initially think. You know, if you translate the amount, the increase in lines of code, and then you go from that, you know, from that percentage to the amount of pull requests and from that to the amount of releases, right? You will see a decrease in productivity because there is a huge bottleneck, as I was saying.

Right. So like I would start there. I will start like, is AI really right now in your company? Is AI really helping you with actual releases and with actual features that are bringing you value? Right. Or where is it that is what is it that is doing? Right. And so if, for example, if people were more critical and if they were if they were more critical and using the technology more

as offloading as opposed to surrender, I would expect the number of releases, the number of features that are actually helpful and better to increase. So that might be something that could be measured. But of course, as you said, it's very difficult because this does happen over time. And it is a very complicated thing. And the other problem with metrics, we love metrics, technical people, we love metrics.

But of course, when you have a metric, you optimize for that metric. then basically, you end up doing that only. So metrics are

Yeah.

complicated, and we need to treat them with a grain of salt. But yeah, I think there are ways of showing these people in which ways you could measure this sort of thing. There's a lot of also human perception. And I don't know if this is more like a

qualitative metric as opposed to quantitative. So maybe you don't like it as much, or we don't like it as technical people, right? But you know, a lot of teachers perceive cognitive surrender in their students. They notice, they're seeing that, you know, they're producing

Mm.

beautiful essays. They don't know what the hell they're talking about, right? And the same for senior people in a company, like at the bank, they were like,

we're worried because...

Yeah, managers

would would be able to tell. Yeah.

Yeah, it's just like, why did you use this algorithm over that one? What is another? Like, I didn't realize this algorithm was

Yeah.

being used. Yeah, and so you can tell. So I think there's a lot of metrics in that vein, right, from senior people, from people who

Mm, hundred percent.

actually know, who can tell, who can tell what is the temperature, right? And what is the usage and what kind of usage they're seeing, whether they're actually seeing learning or not, or adoption, blind adoption.

there is this popular culture thing specifically in law, right? If you think about the American US law firm, the senior partner basically handing over

work for the juniors and the juniors are basically buried in folders and they have to go through it. There is a lot of this idea that they have to figure it out themselves. Now, what I hear you saying basically is also it requires a different culture on the management side as well, in a sense of maybe opening up their own process, maybe communicating it well. It seems to be much more important than maybe it was before where results was the only thing that mattered if you are serious about onboarding the next generation.

Do you agree with that?

Yes. You know, that was the message that I was worried or nervous about giving at the bank because it requires some sort of institutional changes. Right. It requires to value the human tempo It requires not like to stop and to be critical about this cult to to, you know, efficiency. Right. It requires to have like

thinking Thursday, right? And today is like no AI day and we're all going to not use AI for a specific, you know, it requires a slowdown. It requires like a little break, right? And what I contend is that this is in the end going to be beneficial to the company, right? Like even if you don't care really about your employee's cognition or cognitive ability, whatever, if you

care about your business only, which some people do, then you should be doing this as well. Right. Because in the long term, if, for example, you're using AI for all that, you know, strategic thinking and competitive advantage, you're effectively destroying it. Your competitors have access to the same tools. Right. So how do you inject your authentic thought? How do you inject your own your own kind of,

what you bring to the table, the human capital, that's the value there, right? How do you inject that so that it is useful? And of course, it requires some institutional changes, right? Like for example, when we learned that, like I love the gym, you know, going to the gym metaphor, because, know, it requires you to actually go for it and exercise and suffer.

Like no pain, no gain. And yes, technology can be super useful. Like all those machines are great because it allows me to basically do whatever I want. It allows me to track my progress. It allows me to wake up and to keep track of my food, whatever. Technology can be amazing, right? But in the end, you have to do the work, right? And if you are doing cognitive surrender, that is basically taking a robot.

to the gym and having it lift. And of course, if they're paying you for the amount of weight lifted, by all means, have a whole army of AIs and go for it. But do not fool yourself. You're not exercising

Mm-hmm.

yourself. And you are becoming weaker. So maybe you actually need to do

a bit of both. Even if you work in a high

very competitive, hyper-efficient environment. Maybe you do that, but maybe in your own time, you basically use your own time to develop your own cognitive abilities with or without AI. mean, you could very much do it without AI. Just take a book and read something new. But you could also use AI for this. But yeah, it requires some institutional changes for sure, I think.

If you had to design an MBA for this generation of managers, what would be the three things you would teach them?

I, the, the, the, yeah, the seductive allure of the efficiency trap, right? Like the thinking that more and faster is necessarily better for longer, right? That I think leads to burnout and leads to some sort of, especially now with cognitive surrender, it deals with

cognitive decay, and at some point things are going to implode. So efficiency trap, would say.

I will say to learn and to become comfortable and familiar with discomfort, right? That cognitive dissonance that like, don't get it. Like you're reading a paper, you're reading something new and you're like, what is this? Like that feeling, right? You need to become

more comfortable with it. You need to almost, almost appreciate it. Right. It's almost like it's almost when again, the gym, when you're going and you're failing that terrible, wonderful feeling or failing because that's precisely when the gaining is the most prominent.

So becoming more used to or familiar, even liking it, liking this discomfort. that would be another one. And the third one would be

Not to be afraid to reinvent yourself. Like it is very likely that in our lifetime, for our generation, but certainly for the next generations, they will have to reinvent themselves at some point, right? is going to, know, technology in general is going to be, it's already very pervasive. It's already disruptive. And it is very possible that at some point your career is going to change dramatically and you will have to shift, right?

Our parents, our grandparents used to have a career for 40 years and that was their job and they did it for 40 years and that was fine. I had to change my career a couple of times and it was so hard for me to reinvent myself. All my identity was so mixed with what I did. That was so unnecessarily hard. So I would also teach that. It's just like...

you can do whatever you want and you should be ready to adapt. Do not construct yourself and your sense of identity on what you do. You should do it more on what you believe in, what you think is more valuable. So those three things I would say.

On the on this topic, we we've we've talked quite a bit about the the framing of effort, you know, like this idea of you know no pain, no gain. And obviously there's so many things in life that require that kind of effort. for a specific goal which can be different from the effort itself. but but one thing that I find

Perhaps equally important is also, you know, developing what I would call the taste for the effort, right? Which is like an actual liking for the activity, right? The you could think of someone, you know, who's practicing music, right? And maybe to the outside it look like they're just doing hard work, but maybe they are actually enjoying the practice, right? Or same, you you can imagine you know, mathematicians who whose brain never

stops, they're always thinking, you know, about the next theorem, they to bed, they keep thinking about it because it's

It's it's what they like, you know, in in some ways. The I I I I guess one one aspect of cognitive surrender that worries me a bit is that people missing the opportunity to discover that they would actually like specific type of efforts. You know, the I I know artists

Absolutely.

are talking are talking a lot about this in the context of pride, for instance. Like one thing that makes artists enjoy their work very often is that they they take pride in what they do.

Whereas and that that's why they enjoy the effort,

Yeah.

right? Whereas if if if they feel that AI has done ninety percent of the work, they they just don't have the pride anymore. So they don't enjoy the the effort itself. You see what I mean? I don't know if you want to I don't know, if you agree or not with this.

Yes, absolutely.

I do agree. And there's a couple of things there. I experienced something writing one of my essays during my master's. And I wanted to come up with some sort of summary for some papers. I needed to read a bunch of papers, and I needed to understand whatever. And then for the first time, I asked one of these AI's to do it for me.

And it did a great job. it read the papers. It gave me a summary and everything. And I was like, am I done? Like, have I finished the summary of the, have I finished the literature review? Like it, it felt, I felt like I could never do such a good job as the AI. So I felt bad. I felt inadequate and I feel like, what is, what is the point? What am I?

why am I doing this if this thing can do so much better? It could probably write my dissertation for me. So there's this feeling of devaluing what you can bring to the table. And that is really dangerous because if you believe that, then of course you are more likely to adopt AI and basically forget and devalue yourself and your own effort. Because there is something beautiful

in suffering through something and accomplishing something for yourself. Right. Even if my summary was not as beautiful or polished, it is mine. Right. And I suffered through and it took me two days to get this damn data colonialism concept. Right. But now I get it. Now I can relate it to my life. Now I can see it. Now I can project. You know, it's part of me. I supposed to read in about it.

in a very superficial way and then it's beautifully written, whatever, but I'm not going to even recall it. Right. So placing the value again on the process, on the thing that is happening to you as opposed to the result. Like, yeah, the result is great. yes, my my my blog or whatever my essay could have been written better. Sure. Whatever. Right. But that is not the most important thing.

Another thing that I would say is that some people seem to naturally prefer or like this sort of effort. They enjoy a challenge, right? Richard Feynman used to say like, scientists, we love not knowing, right? It's just like, yeah, like what is this mystery thing? Some other people are not as inclined. Some other people feel a bit weird when there is like uncertainty or like, you know, they don't know. However, this is a skill. It's not that you're like you're cursed or blessed.

with this or that, you know, a proclivity, you could, you can develop curiosity. You can develop resilience for failure. You can, you can develop this taste for effort, right? Like you can actually do things. For example, I love going to saunas and I saw the cold plunges and I hated them because they're so uncomfortable, right? But every time I did it, I was like, this is so good for your body. This is so great.

This is like I was going in. I was going into the water. This is so great. This is like I was convincing myself, right? I did it enough. Now my brain, I managed to convince it and now I actually enjoy them and like them, right? And so in the same way, that's kind of what I was saying. In the same way, when we confront a student with something hard as opposed to, this is hard. this is so hard. Wow. So cool. It's an attitude.

It's, know, I don't know if you like Dragon Ball, the anime. Yeah, like Goku. OK, so

It's been a while, but yeah.

this is a random reference, very millennial. you know, Goku, when he confronted a new adversary, like a very strong person, right, he was excited. He was not afraid. He was like, yeah, this is exciting. You know, this is exciting. And, you know, I

Mm.

get to fight with you. You know, that is the sort of attitude.

And I think that this could be cultivated in young people, especially. They're fearless, right? They haven't learned, you know, we can actually teach them to value this sort of thing. And I think that's really, really important.

think there's also a very important aspect of where do you derive meaning from, right? I mean, if

yes.

you take artists, for instance, I think maybe I missed getting this wrong, but I Henry Fonder had basically a lifelong fear of going onto the stage and he still did it and it was fulfilling him tremendously, but to a point where he's very often had to, or regularly had to throw up before actually going live, right? Where you would feel like, why is anyone doing to us in his right mind? And the answer is simple because it gave him a lot of meaning and satisfaction.

And I think this is one of the aspects that you are also alluding to that might be if you are not in that struggle, you're missing out on the potential to get meaning from it.

Yes, and meaning, well, like meaning human flourishing and technology, they interact in very weird ways, right? Like we have this promise of a utopia when AI is going to give us leisure time and is going to give us, you know, satisfactory meaningful lives. And that is this definitely not necessarily the case, right? yeah, a lot of people take pride, as Bruno, you were saying, in what they do.

regardless of how menial or mundane it may be to other person, right? Like it is my work, it is what I'm contributing, et cetera. And yeah, like what I take what you say, Philip, like it doesn't have to be enjoyable. And sometimes because it is not enjoyable, it is more meaningful. Right. Like

Yes.

like Victor Frankl was was it like with the whole, you know, his his his approach to meaning.

and how people can find meaning even in the most horrible places. Yeah. And what happens if AI or technology takes away something that traditionally has been the source of meaning to so many people? Work, right? What happens then? And yeah, as you said, with my dissertation, it was hard.

And I think, you know, it probably could have been much better, but it is mine. And I learned a lot. Right. And I take a look at the scars that I got from it.

yeah, I like it. You know, it's just like in that is meaning. And that is that is so important for well-being, for for flourishing, for just happiness in general, which is something that, of course, technology has also impacted. Right. Like mental mental health.

And yeah, like that is a question that is so important. You know, how to use this technology, not only for our cognitive abilities, for our own flourishing, for our for our own wellbeing. Yeah, that is a super important point.

There's on a yeah, on a related topic, so the I think I think one of the danger we've seen with technology since the mobile phone and social media and so on is the kind of globalization of the of narratives, right? Like the the idea that we all compete on the same kind of Twitter sphere and so on. And I worry

Mm-hmm.

sometimes that, you know

sources of meanings used used to have more plurality to it, right? That you you could have a source of meaning from your community, from your family, from from whatever. But if everyone is in the same platform and there's only one, you know, one dimension, one scale, it means there's very few let's say there's less opportunity to to, you know, define different path to kind of meaning, right? and

Yeah

one one fear a lot of people have specifically with

recent AI models is that because they are trained in kind of one specific kind of culture, typically North America, they might,

Mm-hmm.

you know, have a relatively narrow set of ideas about what you know meaningful life looks like. And people using it from other parts of the world might find it less just just like simply biased, right? Towards a relatively like narrow, narrow set of of

you know, definitions of good or meaningful life, if you will.

And

at best, they would find out or they would realize that that is not what they agree with. Right. But at worst, they could just adopt it as their own. And this is the homogenization of expression

Exactly. Yeah, yeah.

and culture. Right. And, you know, I have a very, very personal example here. Like a cousin of mine was having trouble with with her dad.

And she has been talking to ChatGPT extensively about this problem, as if it were some sort of therapist. I didn't know about this until recently. And the sort of advice that ChatGPT was giving her was very American, very individualistic, very much in

Mm.

the vein of you're an independent woman and you should be doing your own thing. That sort of message.

Mm.

Whereas in Mexico, we have a huge deference for old people, right? Like the abuelitos are usually in our house and you know, it's a whole thing.

Mm-hmm.

Of course, the advice that this thing was given her was culture mute or like was very, it didn't actually know the context. We didn't

Yeah, yeah, yeah.

have that context. And the kind of advice that it was given her,

was actually quite damaging for the sort of relationships that typically, you know, we have in a Mexican family and it is causing a lot of problems. Right. So yeah, that, is, that is a problem with AI and with using AI because by definition it, it just, you know, responds well, depending on the training data. Right.

And the variety of the training data is not, you know, it's limited in for many things. And of course, there are certain culture, cultural values, there are certain things that are more, there are kind of more apparent and there are others that are completely ignored right? Like there are cultures and their views and their things that are not represented at all. And of course, you know, this can have repercussions.

Because as we know, culture is a soft power, but it's a very powerful power. The way

Yeah, yeah.

cultures influence other parts of the world is also a political power. And this is a very insidious way of all of a sudden, we are pushing all of this to literally billions of people at once, without them even realizing.

that they may be giving up their own values.

Makes a lot of sense. I'm a I'm aware we've taken a lot of your time. We're getting to the hour. But I I I guess on this we've talked a lot about both positive and kind of no negatives during this hour. I I wonder on the on this kind of cultural question if you have any I don't know, any note of optimism. Like what what is the what is the path to world where AI gets kind of better at

you know, augmenting either people's individual taste or more like, you know, community scale or cultural scale, taste. Wha what what do you think is the path to to to a better future?

I think,

yeah, I think if I would, if I had to say something like succinct, I would be,

Mm-hmm.

I would say something like be critical, right? Like, like use AI for your own goals and to your own advantages and not only short term, but also long term, right? Make sure that you're being intentional.

as to why you're using this and what is it that you want to get out of it? Like you have to, before using AI for anything, it's just like, do I really need AI for this? What is it doing to me? Do I really need to, you know, like write every single email with AI now? Like, that a, why am I doing that? Like, is that really saving me time? Is it worth it? What is it that it's doing to me? So being critical, being intentional, being judicious.

Once you're there, typically things just improve dramatically, right? It's just like a good parent. If a person is concerned

Mm.

about being a good parent, know, chances are they're going to be a better parent than someone that is completely, you know, unconcerned. So become concerned. this, this is some, this is wonderful. This is a great thing, but it is also quite dangerous and potentially dangerous, right?

So use it to your own advantage, use it to your own flourishing. And so, you know, do not be afraid and use it because you will be competing with people who know how to use it. So you better learn how to use it, but you also should be informed and be judicious, be intentional.

Excellent. On this word of wisdom, thank you so much for your time. Really a pleasure to talk to you.

Thank you very much.

No,

worries. My pleasure. This was great.

Thanks for watching. We've got more episodes in the oven, so please like and subscribe to get notified when they're out. And please check out tastebench.com to read about the Taste Benchmarks and the creative engine we're building towards.