Artificial Intelligence and Rethinking the Principles of Society

Great to be here. Thanks for having me.

Well, on the one hand, I'm an AI professor. If you know things like Google Translate or Yahoo Translate or Microsoft Translate, I pioneered those sorts of things. I built the first web translation AI and developed a bunch of the machine learning foundations for them. I also work on music technology, AI for computer music, and computational social science.

I was one of the people who pushed machine learning approaches into the field of AI at a time, 30-some years ago, when the field was dominated by what we call good old-fashioned AI: rule-based, manually coded, logic-based systems that were really impossible to scale up. We fought this battle to move away from logic, to stop trying to model the world by writing everything we knew about the world by hand, and instead to model the mind, which is not a logic machine. To model these massively parallel distributed units that fire different activation levels at each other, and from that complex system emerges a learning behavior. I've been pushing on that the whole time.

I've pretty much focused on things that have to do with getting cultures to relate better to each other, to understand each other better. That extends also to my work as a musician, having formed one of Hong Kong's world music groups, as well as my social impact work, fighting discrimination, looking to enhance dialogues between different cultures, and looking at ways to decrease the alarming amount of polarization, divisiveness, fear and hatred that we're seeing, both domestically in many countries and globally in the geopolitical situation.

I think I've had the privilege of being shown a lot of the world since I was a child. My parents took us all over the US. They had come to the US as teenagers from war-torn China and Hong Kong and Japan and so on. And they landed literally smack in the middle of the country, in St. Louis, which, as you know, is on the Mississippi River, which separates the East and the West in the US, and on the Mason-Dixon line, which separates the North and the South. So literally right in the middle.

I grew up in the Midwest, in Illinois, the next state over, and it was very mind-opening to see the cultural differences and how they weighed on perceptions of people. And then my parents took us not only to almost every state in the union, but to other countries. We saw things about the developing world that were very, very different. We saw extreme poverty. We saw things that made me question from an early age whether all of the assumptions that were mainstream around where I was growing up maybe were relative to the environment, the economic level, the education level, and of course the cultural background.

I think that these are things that, as we've seen in the last decades, have led to enormous stresses in humanity. And that has really driven me to look for ways, between the technology, the arts, music and social dialogue. There are things that are urgent upon those of us who are able to look at how we can enhance [this], to really ramp that up, so that we don't end up with humanity literally tearing itself apart and destroying itself and the planet in the process.

Oh, very definitely. The thing about human intelligence is that language and music are basically the only things that we humans do better than other species. We don't run faster. We don't fly. We're not very good at climbing trees anymore. We're not stronger. We don't see more sharply. We don't smell more sharply. We're pretty lame as a species, except for our unique abilities in language and music.

And really, music was the foundation of a lot of human language abilities. Our ancestors were singing before they were talking, and early speech probably evolved from using our song abilities to be able to represent ideas about the world. Once we evolved stronger linguistic abilities, we turned around and applied those to music, and so we've been developing more sophisticated types of music using musical languages. So this has co-evolved through a million years, and it leaves us with brains that are supremely evolved for music and language. Those fundamental building blocks underneath are how we implement intelligence.

Therefore, studying music is heavily related to studying language. The same areas of the brain are used, except that music is activating many of them in parallel. It's like exercising those linguistic abilities on steroids. And the kind of creativity, the kind of improvisation and expression in traditional music, teaches us a lot about how the mind creates. Not just how the mind interprets new things that it's never experienced before, but also how it learns and then is able to generalize, to take what it's learned previously and then express new things in real time.

Not in the sense of the very recent aspect of music, especially in the Western European symphonic tradition, where somebody writes out a complex score and then you have trained musicians that do nothing but read and execute that score. That's actually a very artificial, recent form of music that only happened in the last few hundred years. Traditionally, music is improvisational. It has conventions, but it's not the same twice in a row. And you have a conversation between the musicians, who are typically sitting there bouncing off of each other, just like you and I are talking. That tells us a lot about how the brain works, about what kinds of facilities there are with which to learn and to express and to create.

Yes and no. I think the major change was what I was referring to earlier, when we moved from rule-based, logic-based systems that were coded by hand, so that all the knowledge of the system was written by humans rather than learned by the machine, to machine learning models. Deep learning models are one family of machine learning models.

To be honest, I started playing with what we now call deep learning in the late 1980s, early 1990s. In fact, my postdoctoral year in 1992 at the University of Toronto, I spent happily playing in Geoffrey Hinton's [lab]. Geoff is the granddaddy of deep learning, and I was literally working on what we now call deep learning models at the time. But there are also many other machine learning approaches. Deep learning is not the be-all and end-all. Even Geoff Hinton himself has said on the record: I think we need to throw it all away and start from scratch.

It's a phase. There's a lot of media hype about deep learning these days that is fairly shallow. Those of us who are really working in AI research have already moved on. That's one of the tools in our toolkit, but we're looking at deeper problems, like the issues that I've spoken about on modeling awareness, mindful AIs. You can't really have an AI that is just mindlessly doing deep learning. All that is, is glorified regression, and it's not aware of what it's doing. That's a very important distinction between human-level intelligence and the instinctive, automatic, reflexive types of learning that other species also have.

So when we look at music, you get the same dichotomies. For those of us who grew up playing musical instruments: I don't even remember not playing. My family tells me that I was picking out melodies when I was two years old on the little kiddie [piano]. Nobody had taught me to read music. Nobody had taught me technique. This was just something that you learn by associative pattern recognition, the same way kids learn language. Nobody says to a kid: hey, you're not allowed to speak until I've taught you the rules of grammar and how to read, and once I've done that, you're only allowed to say things when you're reading from the page, written by some poet.

It's done automatically, below the level of consciousness. As you're hearing someone speak, you're not saying: oh, that was a verb, that was the agent that was speaking. You're not doing that. It's all happening at the unconscious level. And it's very much the same thing with music. If you've learned music in this natural way, as opposed to the recent artificial way, that teaches us a lot about how the mind works.

But at the same time, once you've done that, and you start creating, and you start thinking: how would I compose a piece? Or if I'm a DJ, how would I construct this track? Now you're applying the linguistic side. Now you're really thinking the way that a poet would think. So you've got all those different types of mental processes going on, and studying music helps us also to understand that about the nature of human intelligence.

I think it's going to, increasingly. It is not very much out in the popular awareness, the popular press, yet. But people who are at the forefront of AI research are thinking about these problems and are making proposals. Our lab is certainly integrating both the automatic, unconscious, fast pattern recognition and associative types of processing with more mindful styles of consciously controlled understanding of what it is I'm doing.

Well, psychologists have for many decades now been studying the differences between automatic mental processes and controlled or conscious mental processes. Daniel Kahneman, who wrote the bestseller Thinking, Fast and Slow, calls it System 1 versus System 2. There's an enormous amount of empirical data on the characteristics of System 1, or automatic mental processing, versus System 2, or controlled, conscious mental processing, where you're aware of what you're thinking.

The thing is that if you're able to introspect, if you're able to describe your thoughts, that is conscious, controlled reasoning. But something like the fact that you're not thinking, oh, hearing was a verb: that's not conscious, that's unconscious processing. Or if I ask you, for example: do you know what the difference is between walking and running?

Sure, that's true. But you've probably also seen these competitive speed walking races, right? So they're walking very fast. And you can run slower than you can [walk]. There is actually a technical difference. Now let me just make sure: you do know how to walk, right? And you do know how to run, right?

And yet you don't know what the difference is. The difference is that if you're running, both feet leave the ground at the same instant. Most of us know how to walk. Most of us know how to run. And most of us have no idea that that is the difference. That is unconscious mental processes. But now that we've talked about this, you'll have conscious knowledge of it.

Yes. That is exactly what needs to happen for AI to actually reach human level.

Well, I think people get confused about what it means. As soon as you drop in a word like consciousness, there are like nine different definitions of consciousness, and they all depend on assumptions, on some small group of core axioms that you have to take on what you mean. First of all, are we talking about mind-body dualism? Are we looking at issues of the soul, which a lot of people fold into what they are casually thinking consciousness means?

But in the sense that I'm talking about, the consciousness that I'm speaking of is empirically measurable. We're not attempting to say things about the metaphysical soul. We're looking at things that you can actually read from brain scanners, from MRI or EEG. We're looking at speed issues. There's a lot of experimental psychology, cognitive psychology, that you can do with everything from eye tracking to just how fast people can process things and what kinds of mistakes they make. So there's a lot of empirical stuff that has been observed about unconscious versus conscious mental processes.

Well, there are different approaches toward defining intelligence. One of them, in the style of Turing, of course, is more of a behavioral definition. If it looks like a dog and it smells like a dog and walks like a dog, it probably is a dog. That's basically the style of definition Turing had in his famous Turing test.

I think from a science point of view, that is a reasonable definition. If you can drop an AI into a new situation that it has never experienced, if it's had the same kinds of environmental upbringing as a human, and if it reacts the same way as a human, then you've produced human-level AI. Again, not ascribing any metaphysical qualities of having a soul to that definition, but that would have achieved human-level AI. And beyond that, if it becomes more powerful in its learning capabilities or its reaction speed, then it starts becoming a superintelligence.

Oh, absolutely. I think creativity is basic. There are a couple of very common clichés that you hear as soon as people start talking about AI casually: oh yeah, these machines are getting incredible, but they'll never be able to have human qualities like creativity and emotion and those sorts of things. And I think that's actually misleading.

When it comes to emotion, that is actually simpler, more primitive and easier to model than truly intelligent human processes like language understanding. Anybody who's had a dog knows that emotion is very basic. You don't need human-level intelligence to have emotion. It's far easier to model. People didn't start doing it until recently, but it's easy to do compared with a lot of other things.

And as for creativity, it is pretty much impossible to have any kind of intelligence without creativity. You cannot have a learning system without creativity, by definition. What does it mean to learn? It doesn't mean to memorize. Memorization requires no creativity, that's true. But to actually be intelligently learning is not just to memorize, like a database does. It's to have been exposed to a few examples, but then you leap to generalizations and you are making inferences in situations you've never encountered before. That by definition is a creative process. To be able to speak a sentence in a new situation, like we're doing now, that we've never done before, requires creativity. These are fundamental building blocks of machine learning, and you cannot have intelligence without those creative building blocks.

So I think it's a comforting myth for a lot of people that machines can't replace humans because we're uniquely creative, and emotion and all of these things cannot be replaced. It's a sentimental notion that doesn't have any scientific basis. I understand very much the human yearning for that, but I think we need to redirect that human yearning to: how do we, with positivity, with the best of the human qualities for creating empathetic societies, deal with the fact that these new machines are already creative and are already, especially in social media, reflecting emotion? Those things are altering our societies in drastic ways.

If we look at the kinds of polarization that I was talking about at the outset: the reason that so much of my work in the last several years has been directed at AI ethics and AI in society is that AI-powered social media and media and search engines are ripping apart our societies, as an unintended consequence of the fact that they're actually amplifying all these natural, evolutionarily hardwired human qualities that cause us to be biased, to believe things that we probably shouldn't believe, and to react in ways that are irrational. In the old days, your tribe would be killed by another tribe if you didn't react that way. But in a modern era, where weapons of mass destruction are being democratized so that ordinary people can be carrying really dangerous weapons of mass destruction, it's no longer tenable. It is no longer survivable for society to be that way.

So we need to figure out how to redirect those sentiments about human emotions, about taking care of each other, and love, and understand how, in our new era where society is already heavily populated by AIs today, we bring those qualities to bear in that mixed society of humans and AIs, rather than allowing fear and hatred and stronger emotions to be learned and propagated by the AIs.

Well, sure, I actually do. And here's why: you are an artificial intelligence. Look at the Oxford definition of artificial. Pretty close to word for word: made or created by human beings, roughly as a copy of something occurring naturally. I think we can agree that you would not exist unless you were made by human beings, and that you were made as a rough copy of something that occurred naturally. And since obviously you're intelligent, by definition you're an artificial intelligence, and we all are. So I think it's actually a very appropriate term.

Yeah. So I'm working [on that].

Yeah, indeed. I think we're getting closer and closer to models in the field of AI that are getting more natural. We're still quite a distance off, but I think that for AI to progress, it has to get more natural. AIs cannot be mindless; they'll never be intelligent that way.

Today's AIs, because they're mindless, require exponential amounts of data. And so people have confused AI with big data. If you're going to process big data, you also need exponential amounts of computation power, which is why you have these enormous farms of CPUs and GPUs these days. But when we look at a human brain, it's being powered by less electricity than a 100-watt light bulb. You don't need that much computation or that much data if you have human-level intelligence. A true AI is actually small data: it is able to make those intelligent generalizations from the same amount of small data that a three-year-old kid learns from. So it is inevitable that progress in AI will drive us closer to natural intelligence.

I am always reluctant to say this, because one of the terrible things about the field of AI is that it has a long history of always predicting that the problem would be solved in 20 years. And so if I say the problem will be solved in 20 years, 20 years later people will still be saying that.

That said, progress has been incredible, going back about 30 years, to when I was finishing my PhD and still fighting the battle against good old-fashioned AI, which was hand-coded logic rules. We have made enormous progress in what AI can do today. For the people back then, what we are doing today would be mind-blowing. I have no doubt that that will continue to be the case.

Yes.

Oh, this was something that happened like for many people. Obviously, we've been really struggling with how to deal with the horrible effects of the coronavirus. Because I have joint appointments at the Hong Kong University of Science and Technology and at Berkeley, I see very regularly what's going on both in the East and the West.

By February, it was really alarming me. Hong Kong was paying attention to the WHO warnings and the warnings from China, and it was already locking down in January. There was an immense amount of [discussion] about what happens on February 1st, when the semester was supposed to begin: should we not physically open up? And indeed, we never opened up physically. We converted to all online teaching for the entire semester.

By mid-February, it was really alarming to me that not only was that discussion not really happening seriously in the West, in the US or Europe, but even face mask wearing was not being discussed. In Hong Kong, literally 99% of the population masked up. I could not go into town without wearing a mask without hearing about it and getting dirty looks, because you're not doing your part to prevent spreading your possible Covid viruses to the other people around you. It was not just about protecting yourself. It was about protecting those around you, the community around you.

But when you looked at the WHO data, or at other datasets, Johns Hopkins or whatever, even though they were listing which countries had social distancing and which countries had testing and contact tracing, they didn't even bother listing which countries had masking. And this was alarming to me, because masking costs far less than testing and tracing and many other measures, and there is no more evidence for social distancing than there is for masking.

So we formed a team rapidly, in late February, with not only me in Hong Kong and at Berkeley, but also people in London, Paris, Helsinki and Tallinn in Estonia, across disciplines: from a medical doctor in London on the front lines of treating Covid, to an economist in Paris, to someone who is actually a senior advisor to the ministry in Finland looking at public policy, and a computational biologist out of Cambridge. We looked really broadly at this, and we built two new theoretical models.

I built an AI-inspired model that combined agent-based modeling, from the long tradition in AI where you have intelligent agents walking about and interacting when they meet each other, to do the predictive modeling. And I made it highly visual, so that you could actually see these agents walking around. You can just take your browser, go straight there and play with it yourself if you like. You can set your own assumptions about how many people are infected, how many people there are, and how effective your masks are at blocking transmission of viruses versus blocking absorption from other people into your system when you inhale, and you can see what happens.

And so I made this four-minute video to explain, as a tutorial, how to use it. The video got a huge amount of interest and led to a cover story in Vanity Fair, which really raised a lot of awareness amongst European and North American states.

Basically, we found that you need 80 or 90% of the population to wear face masks, and you need them to do it fast, before the virus spreads to too much of the population. If only half the population wears masks, it has almost zero impact. And the reason the visual simulation is so good is that this is counterintuitive for most people. Most people think if half the population is wearing masks, then it'll stop half the infections. And that's simply not true, it turns out, because of the exponential behavior of virus spread. Half the population wearing masks has almost zero effect. 80 to 90% is needed in order to have significant impact.

And if you do that, and you combine it with other sensible measures, social distancing, and we do need testing and tracing, then you can control the spread long enough for the scientists to develop the vaccines and the treatments, and prevent the hospitals from getting overloaded. Otherwise we create thousands of unnecessary deaths a day.

So that project was a really cool application of various concepts from AI and computer science to tackle the social problem that is urgent upon us right now, that we're still grappling with: how to save as many lives as possible, and how to be able to exit from extreme lockdowns without risking lives in an excessive way.

In Hong Kong, because we put on masks really early, we have never had to have a strict lockdown. The restaurants have never been closed. We can go out and shop. We do sensible social distancing and we wear masks. To this day, we have had a grand total of four deaths. You're talking about a place with a population of seven and a half million, approximately New York City. In total, we've only had a couple of hundred local transmission cases. Almost all of our cases were imported cases, from people coming back from Europe or the US or other places, and they got tested and quarantined.

So, talking about AI ethics: there are a lot of things that we can do with AI to really have a positive impact on society as a whole, and that's an excellent example.

Absolutely, yes. The best place to start is that explainer video. It's only four minutes long, and you can find it easily at dek.ai, that's my name with the dot in it, slash maskvideo. So dek.ai/maskvideo.

Super. In the video description, there are links to the rest of it. There's a website with all the resources, dek.ai/masks4all, with the number four. There's a white paper that is highly visual and easy to read. There's a research paper. There are a number of other related videos, and a list of news stories from around the world.

I don't even think it's possible to program it in, honestly. The kinds of systems that you can program ethical rules into are not even true AI. They're very, very simple systems. Today there's a tendency, a marketing tendency, for everybody to label what they're building as AI. It might be just some if-then rules. Real AI includes learning capability. The machines are learning in the same way that human children learn.

And the three laws of robotics don't really make any sense in that context, because even when you have only three rules, they're constantly in contradiction with each other. Isaac Asimov wrote dozens of highly entertaining stories, and every single one of them was based on a situation where, gee, these two rules create a logical conflict. And imagine now if you try to write an AI system that has several hundred ethical rules in it. The robot will just be paralyzed, because it will constantly be in a state of contradiction.

So how do human children manage this? They learn, right from childhood. You're taught values, you're taught principles, but you also learn trade-offs, and you learn the difficult situations. And it's highly cultural. There are a few ethical rules that are probably universal, like the golden rule: almost every culture has some form of the idea that you should treat others the way that you would like them to treat you. But other than that, different cultures have a lot of variation.

One of the things that makes machine translation hard is that it turns out that language encodes this very deeply. Some of the earliest words that children learn already encapsulate highly culture-dependent values, and you can't even translate three-year-old language because of that. It's really tough to do. And because of that, we have a situation where the ethical rules that an AI would have to operate with would be just as culture-dependent, just as contextual, just as much in shades of gray as for a human. They would be learning those things. They would not be something that you can program in, because modern AI by definition learns.

So you can no more program ethical rules into a real AI than you can hardwire them into a child. There's no way you can unscrew the head of a child and solder in some logical rules. There's no place to connect them in that architecture. A modern AI is the same. So we are going to have to, as a society, come to grips with the fact that we have to teach our AIs in the same way that we teach humans: what are the ethics, what are the values. And we have to set examples, because they're learning from us.

The AIs at Google, the AIs at Facebook and at Twitter, the AIs at Apple and at Microsoft and Amazon: they are learning by watching what you do. You are the training data. You are the example. You are the one teaching the ethics and the values to the AIs, and we all are. And ultimately, just as with humans, there is no substitute for the fact that all of us bear collective responsibility for teaching the AIs, in the same way that we do with humans.

This is one of those things where, because AI has come upon society so rapidly, we are not wrapping our heads around it. We are still thinking about machines as dumb, passive mechanical tools, as slaves. And we don't realize that everything changes when machines actually have opinions and thoughts and can influence our opinions and thoughts. The metaphor of treating machines as mechanical slaves that we can control and program is a dangerously obsolete metaphor that we need to get humanity to throw away as fast as we can.

Absolutely, yeah. And that's what my TEDx talk was: it was challenging the audience to ask: how is your parenting? And it turns out that when you actually dive in and look at it, we are violating pretty much every principle of good parenting that we all believe in. And that's why AI-powered tech today is tearing apart our societies. If we all abandoned responsibility for raising human children, society would fall apart. And that is what we're doing now with our artificial children. And that's what's so dangerous.

To me, that is the paradigm that we need to challenge and to shift. This is a change that is already upon us. This happened while we weren't watching; it snuck up behind us. And we are still living as if we were in the 20th century. We are still pretending that things can go on.

A lot of AI ethics is still, I think, unconsciously trapped in a mindset of thinking: how can we preserve the status quo in society? How can we design AI so that it doesn't change the structure of society? And I think that that is impossible, even with the weak AIs of today. We are seeing how impossible it is, even if we just look at the news from the last week: Twitter finally deciding that they are going to slap labels on what they consider to be dangerous or false statements, even from a US president, and the pushback on that, and Facebook still refusing to do it.

We are already today living in a situation where we are grappling with the consequences of the fact that the AIs are already deciding, based on imitating humans and learning from human behaviors, which ideas to spread and which ideas to censor, to filter out.

I want to be sensitive about the wording here, but an AI that is working for a social media or search engine company: its job is to decide what are the top 10 or top 20 posts or hits to display to you. You don't get to see the other billion. 99.99% of people never look past the first 20 options. That is, in reality, de facto algorithmic censorship. And there's no way to get around that. You have a lot of useless discussion saying: oh, AIs shouldn't be censoring. It's impossible, because you cannot read a billion posts. That algorithm is going to decide what you don't see.

So AIs are already, by necessity, the deciders of what you do and don't see. And how we teach those AIs is not a question that we can continue to avoid if we want our societies to survive. That is the major paradigm shift that needs to happen, not just among tech developers. That is a major paradigm shift that needs to happen for every member of 21st-century societies.

First of all, I think the concepts of utopia, and maybe dystopia as well, are like fairytale metaphors from certain cultural traditions. They tend to be absolutist traditions, the same kinds of traditions that give rise to the idea of universal truths, that sort of absolute. Those are myths. They don't exist in the real world. The real world is a process. It's a continually evolving flow, and I think the sooner we realize that, the better. It's a convenient metaphor, we can talk about utopia versus dystopia, but we shouldn't get too hung up on that, because it's not real.

What is real is how we evolve our cultures, how we evolve our civilizations, and whether we avoid extinction. What is real is how much we can do away with human suffering, which is still at unethically large levels, given the wealth of our planet today.

I think we need to focus on the fact that it's inevitable that human intelligence and artificial intelligence are merging. That is already happening today. Half of my brain is already in my phone over here. I can no longer remember numbers; I used to be legendary at doing it. I have no sense of spatial orientation anymore. I mean, it was bad enough to begin with, but now I'm basically lost without my GPS.

To me, there's a little bit of nostalgia: oh, it was nice to be able to do that. But it's also incredible freedom, because I have offloaded dreary tasks to the machine, freeing up so much of my cognitive capacity for creative things and other aspects of thinking. And this synergy, this coupling of processing between my primitive artificial intelligence here in my phone and my human intelligence, is the tip of the iceberg.

We are going to rapidly see, in the coming decade, various forms of AR, of augmented reality, that will make the interface with my phone screen look clumsy. It will become much more instantaneous: what I want, what I'm thinking, what I'm retrieving, just in my field of vision or in my ears. And that's before we even get to, eventually, inevitably, brain implants.

So all of this is going to happen. There are going to be older generations that resist kicking and screaming and say: oh my God, we're losing what it means to be human. The younger generations aren't going to care at all. Let's be honest. The kids of today are doing things with TikTok and everything else, and the older generations can yell until they're blue in the face, and it's not going to change what the younger generations do with the technology, because it is freeing. The natural tendency to be nostalgic and to sentimentally want to hold on to what we were familiar with when we were kids just doesn't carry enough weight against the additional power that AI and tech are able to give to the creativity of the younger generations.

So acknowledging that, and using that to rethink how we need to reform our societies, is essential. I'm American, and the founding fathers in the US, 250-ish years ago, thought long and hard about how to define the constitutional principles. They did a lot of what we today would call systems thinking: trying to figure out all of the interplay in a complex dynamic system and how to keep balances. And it sort of worked for a while. But it still led to an awful, bloody civil war, the effects of which we are still grappling with, even this week in the news, as everybody is saying. And the situation is being made much worse, spiraling out of control, by all the amplification of AI-driven tech, social media, et cetera.

So once again, it is time to rethink the way our founding fathers thought, in the US, and in parallel in many other countries as well. It is actually long past time for us to rethink, in a modern era with AI at the core of so much transfer of knowledge and information: what are the complex system dynamics? We cannot just blindly assume that principles defined 250 years ago, or even 50 years ago, still apply. They don't. It's a brave new world. We need to do that rethinking.

I don't think we can survive doing that through another revolution. The weapons that we have today are far too dangerously destructive. We're not just talking about unreliable muskets. We really cannot afford a revolution, in a hard sense, to do this. We have to raise enough public awareness to be able to understand the absolute need to reframe the founding principles of society in an AI era.

I would just like to have made contributions to get humans to overcome their natural, unconscious biases against having greater understanding between different peoples, between different cultures, between different neighborhoods.

Our biggest danger, with all of this AI-driven amplification of fear and hatred and divisiveness, is our own natural tendency to divide everything into our in-groups and out-groups. That is a process that dehumanizes the out-groups. If we're going to struggle with the question of what the artificial does to our society, then we need to really focus on those biases that cause us to dehumanize too much of our society.

Pleasure.