Video: Humanity Meets AI Symposium: Framing the AI Conversation

How might AI influence our understanding of humanity, morality, and meaning-making? How will religious traditions and communities adapt to or shape the ethical frameworks guiding AI development? In what ways can religious perspectives contribute to the creation of a more equitable society amid the disruptions AI will bring to labor, governance, and social structures?

Religion and Public Life hosted a symposium that explored the profound ways in which artificial intelligence is reshaping human society, with a particular emphasis on the role of religion, the transformation of societal structures and capitalism, and strategies to reduce inequities as society responds to the sweeping changes brought about by AI.

The symposium equipped our audience with the tools and frameworks to critically engage with the ethical, spiritual, and cultural dimensions of this transformation. In an age increasingly influenced by AI, this symposium helped identify practical opportunities to shape a more humane and just society.

FULL TRANSCRIPT

SPEAKER 1: Harvard Divinity School.

SPEAKER 2: Framing the AI Conversation February 27, 2025.

SPEAKER 3: We will get started. I'm going to hand things over to Swayam in just a moment. Professor Swayam Bagaria, who I think is probably the most interesting man in the world. His undergraduate is in mathematics. And then he did his PhD at Hopkins in medical anthropology. And for some reason is a Hindu studies professor as well. It's amazing. He got a tenure track appointment at Harvard in Hindu studies.

And of course, he just knows Sanskrit and stuff. He's really a polymath and brilliant. And I'm really, really thrilled to hear what he says. And he's actually the one person I don't even have a written bio for. I was just winging that. I was going to read the bio. But it's blank. Did I miss anything?

SWAYAM BAGARIA: No, that's great.

SPEAKER 3: That's his entire life story. Thank you.

[APPLAUSE]

SWAYAM BAGARIA: Good morning, everybody. I did not make my slides using AI just so that we get out of the way. So when James asked me to present at this conference, I thought he was asking me to moderate a panel. 10 days ago, he said, oh, you have to give a talk. Then I realized, well, I don't have that much to say. Then I had a third whiplash when he's like, oh, but you can frame the conversation. Then I'm like, yes, I can frame. That's what I can do.

And then this morning, I realized that he basically used ChatGPT to essentially say the things that I wanted to say as part of my talk. So there will be some redundancy. But I hope that you will see that there is a little bit of-- what happened? I'm an anthropologist, as he mentioned, by training. And I have tried very hard to bear the insights from my discipline for the purposes of our conversation today.

I had a few hesitations, only because of the kind of person that I am. And I'm going to share what those hesitations are. The first one is that I don't think you need to be an insider to talk about AI. But I do think you need to have a lot of technical competence. My rule of thumb is that if somebody who I know can speak better on this topic than I am, then I should defer to them. So I do know people who will be able to speak on this a lot more than what I have to say.

But I do think that a basic understanding of the architecture of how AI works, or how large language models work is necessary to be able to say something meaningful about it. Otherwise, I think a lot of the conversation happens to take place at a level in which people just from the get go are willing to agree to disagree a lot about a lot of things. So that's my first hesitation. So I'm just going to put it out there directly. I am a mathematician by training. I do data science. I have some understanding of how LLMs work.

Nevertheless, I'll still have that as a preface. Second is that, yeah, the speed of innovation. Because just last week, I don't know how many of you have been following this. But the first third-generation large language models were released. Things are happening at a scale and at a speed in which anything that you say seems to have become a little bit more redundant, just two weeks after.

So, in such a highly accelerated environment, how is it that you say anything meaningful about what is happening in the world, in AI out there? The third one is that there is also-- and I don't think anything specific to AI. But there is a lot of information out there and a lot of people having a lot of opinions on AI. I went into one of those rabbit holes of listening to podcasts about AI two weeks ago. And YouTube kept recommending me more and more.

And then suddenly, I thought, well, all the words that I wanted to use to try and understand what I was is being used by so many other people in so many different ways than I want to use it. So this is a feature of modernity, which is that there's a certain amount of endless production of the categories for self-analysis that modernity has. It's a feature by design. And in some ways, I think the best way to try and understand what is happening in AI is to use some of the concepts that are coming from within the community, and then to see how you can place them alongside other thoughts that you might have.

So there's this open-endedness of the analytical task that I think is very important to acknowledge. And the fourth one is that often when I look at what is happening technically in AI and what is happening in terms of the domains of ethics, the latter seems to be not slow. But they don't seem to have the grasp that a lot of the technical specialists of AI do.

So, for example, the AI Bill of Rights was created two years ago. All the procedures that are recommended there, such as creating failsafe mechanisms, creating kill switches. These are we have known this for the last three years. The point is not that, we know something more in the ethical domain of AI now. It's just that we have failed to find a way to implement those things. Even as the speed of change is so incredibly high.

So having mentioned my hesitations, I just want to say a little bit about what I call the fog of AI, a little bit of fog of war kind of a thing. Because there is also, I think, this other sets of issues, which makes conversations about AI a little bit confounding. So the first one is a term that was coined by a computer scientist, Arvind Narayanan. He is at Princeton. He wrote this book called Snake Oil AI. I recommend the book.

He has this term called AI hype vortex. And by that, he means something very, very simple. He means that often, we tend to classify highly disparate things under the umbrella of AI, which if we just disaggregated, we would know that these things are not the same. So the software that the bank uses to evaluate your credit score is not the same as the ones that are used to generate images on DALL-E. Nevertheless, these things can be called AI.

I'll give you one example from my own experience. In India, there is something called the Kumbh Mela which is going on. I don't know if you've heard of it. It is this massive sort of festival that happens every 14 years. People go to the river, and then they take a dip, and they feel better. One person at the business school has funded a startup, which basically creates tea. But a robotic way of creating tea. But that is being considered AI in India, which is like, no, it's not AI.

So just be very safe. So I think this sort of phrase, this AI hype vortex in which we tend to classify many disparate specific purposes applications as AI is something that we should keep in mind as we go forward. The second one is also if you hear any conversation about what is the impact of AI on humanity, people often confound goals. So, just one example. If you ask somebody, well, what will be the impact of AGI? They might often say things like, well, it will lead to the curing of so many diseases.

So many drugs will be discovered. It will make drug discovery very easy. Those are specific applications of AI. So there are narrow agents, which is, again, a term, which was coined by this professor at UPenn called Ethan Mollick, who is a very big supporter of co-assistants, AI co-assistants. And he said we have to, at some point, distinguish between what is a narrow agent AI and what is AGI.

And often, we also have to distinguish how do we often rationalize what AGI is by proposing the goals that narrow AGI achieves or narrow AI achieves. So that's the second feature of the fog of AI conversation. The third one, which humans. And this is not a-- I mean, we all know that AI has biases. But there's something else that I want to show you. This is a graph from a study which anthropologists lead to see when an LLM does respond, which human being is he responding as.

And what they did is that there's a set of questions called the World Values Survey. The World Values Survey are basically questions that have been administered across the world for the last 40 years to determine all kinds of things about human beings across cultures, things about their morality, about the sense of meaning, purpose, et cetera. So this questionnaire was administered to ChatGPT 3.

And essentially, they created something called a cultural distance framework, which is how far is a culture from what we consider to be American culture. Now, again, I know national cultures has heterogeneity, blah, blah, blah. Nobody is doubting that. But you will see that on the y-axis is correlation between ChatGPT and humans. And the x-axis is cultural distance from the United States.

You will find a lot of places where people answer questions on the World Values Survey in a way very different to how these large language models answer these questions. So, in some ways, there is a certain-- when we are talking about humans, we are talking about what is called in anthropology, weird humans, which is basically like educated, the phrase, the acronym that is very famous right now. So just keep this in mind as well as when we talk about the fog of AI. Because when you're talking about AI, we are talking about two countries, the United States and now China.

But even in China, for example, the way in which they employ or they plan to use AI is highly different. And I was trying to find use cases for DeepSeek. And it's very interesting how the first use case that they have found for DeepSeek is to enhance provincial governance. So, again, the way in which we plan to use AI and the way in which other places plan to use AI is also highly different. And then there's the fourth point about the fog of AI, which I do think is well-known, which is that there's an unevenness of its effects on short-term, mid-term, long-term.

And we often confuse those three.

So when we talk about abundance, we are not talking about the short-term. But we often use it in the conversation as if it were an anchor that would allow us to think about our expectations in the near future. And it clearly isn't. Economists have a way of thinking about the bifurcation of the timeline future, which is short-term, mid-term, long-term. But I think you have to be a little bit more granular to think about which effects arise when and what will happen in the interim, such that the secondary effect can't arise or the latter effect can't arise.

So we often think of effects as not compromising the way in which we will lead to subsequent effects. And I think the unevenness is something that we have to keep in mind when you're talking about the fog of AI. So what am I going to do today? Having given those qualifications, very academic thing to do. There is a philosopher who I really like. His name is Wittgenstein. He has this idea called picture thinking. What is picture thinking?

It's like when we think of an object, we often get captivated by certain images in our mind. And those images are something that confuse us to the point that we aren't able to click out of them. So the way in which I have arranged my thoughts for today is I have focused on five pictures of AI that we just can't get out of. So when we talk about AI, we can't not talk about work. When we talk about AI, we can't not talk about leisure. When we talk about AI, we can't not talk about meaning.

When we talk about AI, we can't not talk about what I have called the anthropoid techniques of the human. And I'll tell you what that means. And when we talk about AI, we can't not talk about language. And this is, by the way, a pretty expansive coverage. But there are other things that AI effects. But we don't necessarily think those things as being central to the way in which we think about what is AI or what are the pictures that we associate with AI.

So I'm going to quickly go through all these five pictures. And essentially, what I'm going to do is I will synthesize a lot of the insights from scholarship on these issues and present them as dilemmas to you. So this is the open-endedness of the task that I had alluded to earlier. Picture one. Work labor. I think the question here, when we talk about automation, when we talk about unemployment, the question is not so much what will happen to the labor markets, which is, of course, central. But how do we reclassify work as being not about labor?

And let me just give you a little bit of my thoughts on how I came to this specific proposition.

So, if you are an economist, you will have heard what the first phrase means, the lump of labor fallacy. Anybody has any idea of what that means? It basically means something very simple, which is what James was alluding to that the labor market is not finite. It will keep expanding. Different kinds of jobs will keep being created. So we shouldn't be so worried about the fact that AI will take away jobs because it will create new types of jobs, new kinds of jobs that we can't anticipate yet.

A more sort of pragmatic approach to that question of the lump of labor fallacy. You can see it in the way in which organizations actually absorb AI. So there's one example. When we think about gig work, we think Uber invented it. But there's this book. I again recommend reading it. He says that gig work was incorporated by organizations only because something short-term contracts became a thing before something gig work was an object.

So, essentially, you needed three decades of permanent contracts, becoming short-term contracts in organizations, which then allowed them to absorb this idea of gig work as something that is more permanent. So, organizations change first. And the absorption of technology comes later. And if you, again, look at it in the realm of AI, this is, again, called the diffusion problem, which is that for economists, the main bottlenecks often for AI is human beings.

It's like you will not allow for this technology to be integrated in the way in which it should.

Because diffusion happens in a lot more slower pace than technological innovation. The third one is, again, given that organizational absorption of technology happens in a lot more slower pace. how do we then evaluate the effects of-- or the gradual effects of unemployment over time? And here I just want to allude to the person who won the Nobel Prize recently in economics, Daron Acemoglu.

And he basically maps out scenarios that a lot of what will depend, or a lot of the future will depend on what kinds of tasks are automated. So there is a future in which AI will automate those tasks which are rote. And if it does automate those tasks, then the macroeconomic implications will not be that drastic. But as soon as it starts automating some higher-order tasks-- because no job is just one task.

A job is often like a bundle of tasks.And that is what makes something like incorporating AI a little bit more harder. You're not automating one task, you're automating a job which seems to be different. And he says that, well, you can automate road tasks. And the effects on macroeconomics or inequality will not be that high. But if you start automating higher-order tasks which involve decision-making, which involves a certain people management, then the effect will be a lot more higher. So there are different scenarios that can play out.

And economists usually tend to simulate them. But we should at least just keep this in mind that when you're talking about AI and work, it is not clear. How is it that the future of work will transpire even from the information that we have right now? The fourth and/or the last one that I just want to mention, because for me, labor is a particularly modern category of analysis. We have always worked. But we have not always labored.

If you want to see this distinction, if you have read Marx, he has two texts. One is 1844 manuscripts, which is where he comes up with this idea of alienation, which is that workers often get alienated from the products of their own labor. That is what capitalism does. This is where he has a certain idealized image of labor in which we are Homo Faber. That is the term that he comes up with.

We are, as human beings, are human beings who like to labor.

But if you come to this latter part of his career, which is Grundrisse, or capital, he thinks of labor as just a category that reconfigures what we think of work. And the reconfiguration happens because capitalism starts to put a premium on labor power. So, essentially, the question here is not when work gets automated, what kinds of work will be lost? But can we think about work as not being labor and see it connected to certain different kinds of ends that we often associate with in capitalism?

So that's the first picture thinking that I want to put across. This brings me to the question of leisure. Leisure is something. Again, it's a great thing to have or to anticipate, very bad thing to have. Because people often don't know what to do with themselves when they're not working. So just, again, I want to mention Keynes's very famous essay called The Economic Futures of our Grandchildren, in which he anticipated that come 2030, we will have 15-hour work weeks.

And what will we do otherwise when we are not working, we will have leisure. And we will be like, I don't know, painting, cultivating hobbies, fine-tuning the art of conversation. Who knows? The thing is that, and there's a lesson here, people often leave leisure unspecified. They are not able to say what will replace the kinds of activities that you do if you're not working, which will intrinsically remain meaningful. And here, I want to just show you one photo.

Sorry.

This is the older version of the PowerPoint. No, don't worry about it. So do any of what the superfluous man is? This is a character, which became a very, very typical portrait of a person in Russian literature. Turgenev, Pushkin, they all have some version of the superfluous man who basically they narrativize in their novels, short stories. This guy was somebody who was very wealthy, had very rich parents. And essentially, did not want to participate in working.

So he would gamble, drink, and essentially, live a life of cynicism. So that is the superfluous man. And often, the reason why I'm mentioning this is that you have examples in the history of literature of what happens to people when they don't have work, and they have all the abundance that they actually want to be able to live their lives. So what separates a cynical future of leisure from something that we actually enjoy is not so clear to me. That's the second picture that I wanted to emphasize.

The other one, again, the affluence without abundance. This is an essay written by an anthropologist called Marshall Sahlins. And he wrote this very famous essay called The Original Affluent Society. And his argument is that he draws on data on one specific Tribe called the Kung. And he says, well, they are not abundant, but they are affluent. Because relative to the kinds of resources available to them, they have a great life. They work for 15 hours. They eat very well.

The diversity in their nutritional habits is very high. They have a lot of time just chatting and hanging out and dancing. And he calls them affluent, but not abundant. And I think one thing that we can think about is when we're thinking of leisure, when we have abundance, and we have affluence as well. And this is, for me, a very interesting sort of thought experiment. Because as James was mentioning, even if we do have abundance, it does not kill our impulse to have positional goods.

We still want a bigger yacht. We still want a bigger house. And that is what economists call positional goods. So social signaling does not end. Status games do not end even when we have abundance. But in an affluent society, which might not be abundant, social signaling or status games might have a lower premium. So that's the second picture that I want to just mention to you as something that we should think about.

This is a quote I just thought I would put it on my PowerPoint. It's by a very famous jurisprudence, a legal scholar, Richard Posner. And he basically had this to say about a book that he was reviewing by two authors called the Skidelsky's. I'm going to read this out. The Skidelsky's have an exalted conception of leisure. They say that the true sense of the word is activity without extrinsic end. The sculptor engrossed in cutting marble. The teacher intent on imparting a difficult idea.

The musician struggling with a score. A scientist exploring the mysteries of space and time. Such people have no other aim than to do what they are doing well. That isn't true. Most of these people are ambitious achievers who seek recognition. And it is ridiculous to think that if people work just 15 or 20 hours a week. They would use their leisure to cut marble or struggle with a musical score. If they lack consumer products and services to fill up their time, they would brawl, steal, overeat, drink, and sleep late.

Again, this is just one example of how this idea of leisure. The reason why it is unspecified is because we actually don't know what we will do with our time. And something analyzing boredom as an evolutionary concept becomes highly interesting to me in this setting. The third picture I want to go to is meaning. Now, as James said, we have spiritual poverty and material prosperity.

The solution to that is often what we call meaning. Not very clear what that means, but there is a definition that we can work with. This is a definition that is provided by psychologists. They have honed in on three concepts that compose this idea of meaning. One is coherence, which is that we imagine that the world is a totality, that there are relationships within it that hold, and that we can use to ground our own sense of being. The second is significance, which is that essentially, how do we use this idea of the world that we have?

So, for example, if I say the world is sad or the world is tragic, my sense of significance comes from the fact that I'm going to recuse myself from participating in it. And my purpose is basically to live my life out. That is still meaning. Because it gives you a coherent sense of the world. It establishes your significance in it, and it gives you a sense of forward movement which comes from purpose.

When they say that we have a meaning crisis or there's meaninglessness, what they mean is that if there is a shock to our lives, it can come in the form of losing our job, getting divorced, whatever. We are not as resilient as we would be in response to those shocks compared to if we did have meaning in our lives. So meaninglessness gets expressed when there is a lack of resilience. So how do you notice if somebody has succumbed to meaninglessness?

You look at how they respond to a shock. And if you see a lack of resilience, that's how you know that essentially things can become very, very bad for them. And the reason why they say or the reason why they argue that meaninglessness is something like an epidemic right now is because he says, we just do something called fluid compensation, which is we take one domain. Let's say there's a God-shaped hole within us.

And we try to fill it up by using elements or beliefs from another domain. Let's say aesthetics, or making money, or I don't know, friendship. And it just does not happen. We are trying to compensate for the lack of beliefs in one domain by using beliefs from another domain. And that mapping or that matching often falls apart. It does not stick. The power of stickiness is pretty low. The way in which, for example, you see this idea of meaning coming up in AI is I found a nice example of it.

I'm sure you have all heard of Nick Bostrom. His recently unpublished essay is on something called cosmic purpose or cosmic host. And he says, well, it is our duty as human beings to work in response to something called a cosmic host or a cosmic agent, the preferences of which we can't understand. Yet we still see it as our evolutionary duty to reach that specific agent. So, again, if you have read Nick Bostrom, he does a lot of philosophical tinkering and a lot of game theory to think about what this cosmic agent might mean.

But essentially, he wants to say something as simple as we should not be speciest. Our jobs as human beings is to reach something beyond us. And if we just start with that premise, then you can positively posit an idea called cosmic host, which can anchor our sense of purpose in this world, which seems very, very fragmented and very meaningless. I don't know if you buy that idea or not.

But for me, this question of, can you find a purpose? The source of which is not yourself, emerges even in the realm of AI. It's just that religion gets replaced by a certain idea of a cosmic host. If that is a little bit clear. So this is the third picture that I wanted to leave you with. The fourth one, anthro-- what happens to human beings? Anthropoid techniques is essentially a phrase that combines two realms of what does it mean to be a human being.

And we are anthropological creatures. We have an evolutionary lineage. We have a phylogenetic tree. And the second one is we are also technical creatures. And often, these two might come in tension with each other. The question here that I want you to think about in this picture thinking is, what is distinctive about humanity? And let me give you some examples from the discussion that I have tracked in AI and in anthropology about this question.

So the first one is we are by default transhumanists.

This is an argument a lot of people make. I'll give you one version of it. There is a neuroscientist, a cognitive scientist, a neuroscientist called Andy Clark. Andy Clark and David Chalmers wrote this very famous essay called The Extended Mind. Their argument was that essentially, our mind is not our brain. Our mind is everything that we do on the outside that could be considered our cognitive operations. Though it involves external instruments.

If, for example, when you use a calculator or when we use a paper and a pencil to do a math some, the question that they ask, is this paper and pencil something that I'm using to just support my cognition or is it a part of my cognition? They will say latter. So, for them, mind is something that by definition is extensible. And it keeps on looking for more and more things to incorporate within it. In some ways, this is something that this phrase I really like from Andy Clark in his book on Transhumanism.

He says, we are all electronic versions. We are somebody who have not yet figured out what is it that we actually are, or what capabilities will we realize once we do incorporate and merge with all these gadgets around us. There is a counterpoint to that which our critics of this paradigm.

And the term that I have thought about is technological dandyism, that we often just add things to our sense of what we can achieve or our sense of mind without necessarily thinking about what they might be adding or taking away from our sense of ourselves.

There's another one that I just want to mention very quickly is a recent argument made by a Columbia Professor of Philosophy, not the one that you mentioned, James. But essentially he says-- and this is of religion and philosophy.

And the book is called After The Human. It was published two months ago. I recommend reading it. And in this book, he basically makes an argument that human beings are they are bad. So we do want an AI. We want an AI that fixes our mistakes. So, for him, it's actually a good thing that we are becoming transhumanists to the point that we should actually just become completely transformative and abandon the human in transhumanist.

And there is some confluence of biology and technology that will allow us to be able to achieve that. So, for him, he's a professor of religion, teaches Christianity, or Christian philosophy, and basically, says we human beings are irredeemable. And we should want something to replace us, not because it is some kind of a cosmic host that we are responsible to, but because we are just very, very horrible.

So, again, I think all of these arguments, they essentially play on what is it that we consider to be human beings and whether we are speciesist. And again, I'm going to leave you with that question in an open-ended way to think further about how this question might arise as part of this conference. The last one that I just want to quickly mention is language. And this is something that I'm actually very sympathetic to.

A month ago, MIT Technology Review published an article in which they profiled eight people just having a conversation with a chatbot. They often use these conversations to figure out everything. What is my kink? What is it that do I do? How do I parent a child? It's in the December issue of MIT Tech Review if you're interested. People will be like, OK, this is the most weirdest use of AI because you're actually not helping anybody do any task.

You are filling a hole that exists in people's life prior to the oncoming of this technology. I would argue the opposite. I think this is the best use of AI.

A non-goal-oriented use of AI, in which something like, how does it help us relate to our existential drift is, for me, the most ideal use of what an LLM can be. I'll tell you why that is the case. This phrase, stochastic parroting is a phrase that linguistic anthropologists come up with to talk about what do these LLMs do? I'm not sure what they have against parrots. Parrots are very nice creatures.

But essentially, I think they're wrong for two reasons. First, our language is very, very redundant. Our human discourse, we often use very repeated phrases. It's not like, I don't know, we're not being Shakespeare in our daily lives. We are being very ordinary people who repeat the same kinds of sentences, the same kinds of phrases, the same kinds of references on a daily, weekly basis. There's a lot of redundancy, which allows for these LLMs to actually be like, hey, I have a sense of what is it that you want.

Because there's a lot of self-patterning in our own discourse that we do. And the second one is that there's nothing wrong about stochastic processes. I mean, that is stochasticity under constraints is what evolution is in some ways. So, for me, the fact that LLMs are talk to you. But it seems like the language is a little bit, I don't know, not heartfelt. It doesn't seem personal. It's not something for me, which is a problem.

Because we ourselves are not that deep. And this is where essentially-- again, I'll go back to Wittgenstein. His argument against a private language. For Wittgenstein, it is impossible for us to have a private language, all languages public. Because if a language was actually private, we would not be able to speak about it. Nobody would know what the hell are we referring to.

And there's the second argument that I want to provide in support of this. And I know that James juxtaposed his personal voice with the AI-generated lecture that he read. And he told me that oh, well, you can perhaps sense the difference between what he said, or what he wrote, and what the software wrote. I wasn't able to. There is a difference. The difference is in orality or the way in which voice indicates a certain text, as opposed to the content of the text itself.

There's a second argument that I just want to give you in relation to this is that there is a linguist called Roman Jakobson. And 60 years ago, he basically said that most of our conversation is not referential. Most of our conversation is what he called phatic. Phatic conversation is basically conversation that does not communicate any information. It just maintains social connections. Think about conferences. How many have you been to and you have realized you have learned nothing new?

But you have reaffirmed that these are the people who think about these things in a particular way. And that is phatic communication. And for Jakobson, so because so much of our conversation is phatic communication, it is very, very fine to have a language, large language model talk to you. Because essentially, what you are reaffirming is not whether this chatbot is providing you with some information about yourself, but the fact that you have somebody to talk to 24/7 on a daily basis on demand.

And I'm going to give you one quick example of this. I don't know if you have heard of ELIZA. It's a very famous thing. Weizenbaum created this model, the first chatbot in 1970s. And all ELIZA did was-- it was supposed to be a therapist. You give me a question. I'm going to repeat the terms of your question to you as part of a further question. That's it. The technical term for it is called lexical entrainment, which is that I use the words that you provide me to make you feel as if you are heard.

And guess, when he actually gives disbursed this chatbot to his students, a lot of them felt heard. This is 1970s, which this is a chatbot trained on one technique, lexical entrainment. I think chatbots now are a lot more, I would say, sophisticated. So I do think that this is a use case, which is often derided. But which I actually think it's quite novel and quite needed in some ways.

So those are my five pictures. Essentially, these are like five things that I hope you think more about as part of this two-day conference. It is not to give you any answers. I do think that it is very hard as I said. Because of the fog of AI, it is very hard to give any specific answer to any specific question, philosophical, or technical. And I'm going to just end this with giving you a little bit of a counterargument to my own argument. Because I want to be like, you know what?

Maybe things, maybe AI is not-- we should perhaps think about culture a little bit more when we think about AI. And these are my version of the precautionary principles. The first one is, does AI actually augment us or does it make it duds? We often think of AI as co-assistance. And it will help us do certain things. But maybe it just prevents us from doing the things that we should have done on our own.

Norbert Wiener is the founder of cybernetics. He had this quote, the world of the future will be an ever more demanding struggle against the limitations of our own intelligence. I'm going to show you the abstract of one paper that was published by UPenn a couple of months ago.

This is the abstract. Not the abstract, but the title. The impact of generative AI on critical thinking, self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. Essentially, in this paper, they say, you know what AI does is not assist us. It just makes us very lazy. So is something like large language models as part of organizations?

Is it just making us duds or is it actually complementing us? I think that's an open question. So we should be a little bit cautious about we know when, for example, Ethan Mollick had a post, blog post last week in which he the title was The End of Search Search, the Beginning of Research.

And he was using the Gen 3 models to basically be like, hey, look at how much research can you do? He can do it, definitely. I'm not so sure if everybody else in the world does the same thing. Second precautionary principle. Intelligence versus niche construction.

Are we just intelligent people, or are we also something more? There is a paradox, Moravec's paradox, which was coined in 1980. This paradox basically was about something quite simple, which is that, how is it that we can create code that can help us beat other people in chess, but we can't simulate the sensorimotor movements of a 1-year-old baby?

And in some ways, the sensorimotor mediation of intelligence is as important to us human beings as just linguistic generativity that you often see in large language models. And again, I'm not saying that other people are not working on this. Yann LeCun is working on something of this kind.

He is the chief AI scientist at Meta. But I do think that we human beings, we are not intelligent, we are adaptive. And we do a lot of niche construction, which means that we modify the environment in which we find ourselves as often as we try and be intelligent. Perhaps the former is actually a lot more central function of what does it mean to be a human being. At least that's what my discipline anthropology would say.

The third one is-- I think I'm going to run out of time. So I'm going to quickly go over these. Wisdom. Wisdom is not knowledge. And it's not information. Wisdom is essentially, the definition that you find of wisdom right now in positive psychology, is that it is a morally directed metacognition.

So if there are five people in the room, they all have their own knowledge.

But wisdom is essentially what coordinating mechanism can you find. Such that the knowledge of those five people can be expressed in the best possible aggregate whole. And I don't think that is so easy to do. We often need a different thinking, perspectival thinking. We need participatory thinking, which is not just about intelligence, which is not just about cognitive knowledge, or cognitive inclinations.

The next one is culture. We all have fantasies about replacing our culture's wholesale. I think that's a bad idea. Culture has a lot of redundancies built into it. It is the reason why evolution is slow. The reason why adaptation is often long drawn out is because we know that we will screw up. And somehow that acknowledgment is built. It's designed into the way in which culture functions.

So any project that wants to have this massive uptick, whole scale replacement of it, I think is destroying a lot of the redundancies that are built into it. There's a lot of self-correction that happens as part of culture, through social learning, through niche construction. And we don't want to completely abandon those mechanisms that are built into it by design. The last one, I'm going to skip these two. Preexisting problems and post hoc solutions. I think it is very, very good to remember that a lot of our problems that we have right now is pre-existing.

And we often see AI to come and solve them for us. It's like I study mental health. I study cognitive sciences or mental health. And it's like you're essentially treating the symptom and not the problem. And the sources of a lot of our problems is something else. So maybe I think addressing those sources might be as important as just finding one more novel solution to it. That's it. I don't have anything more. Thank you.

[APPLAUSE]

SPEAKER 2: Sponsor, Religion and Public Life at Harvard Divinity School.

SPEAKER 4: Copyright 2025, President and Fellows of Harvard College.