De Kai: Why Empathetic AI Is the Mother of All Commons | Intelligence at the Frontier 2026

Transcript

Hey, thanks for coming. Um, this is my third talk today. So, like I've been talking about raising AI on the AI floor. Um, and I'd be happy to talk with you, but I want to skip ahead on this talk. This talk is actually titled why empathetic AI is the mother of all commons. And I want to explain that uh and what we're doing about the problems that I raised in the talks earlier today. So my perspective is coming from like 45 years of working in AI of which like more than 40 are in research since I started my PhD at Berkeley. And I think

one of the things that's really important to understand is that AI is actually not transformers, not LLMs. It's it's actually much deeper than that. AI is the study of the mind. We are modeling the mind. It is the ultimate mindfulness where we're understanding how the unconscious works. We're paying attention consciously to the unconscious mental processing. That is 99% of what our brains do. And we're describing it with mathematical precision so that we can actually implement simulations of that. And understanding AI is about understanding ourselves just

as much as it's about understanding engineering. This is really important to understand from the point of view of the continued existence of humanity. Anyone who was upstairs heard me conclude um with what I think is the clear and present existential risk of AI. Now, we all know that we've been hearing the Hollywood narratives and the DC narratives. You know, we've been hearing about rogue AIs taking over wanting to destroy humanity.

But here's the problem. the at the rate we're going, humanity is going to destroy humanity before AIs even get a chance to. And here's why. It's what I call the toxic AI cocktail. Because on the one hand, the most giant influencers in the world today are these things. You have a 100 AIs, 200 AIs in here watching you all day, learning to imitate, attention seeking, wanting your approval. It's your Reddit AI, your YouTube AI, your Open AI, your Instagram AI, your Google AI, your Microsoft AI, your Apple, we could go on all day.

And they are the most giant influencers in the world. They are deciding for every one of us hundreds of times a day, not only what pops up our screens, they are deciding what will never pop up on your screens. And that is driving and escalating the problem of polarization because they're dividing us, siloing us into different echo chambers and filter bubbles. We don't hear from other groups of humans and we naturally fear what we don't understand. And that fear drives hate and anger which slides into demonization and dehumanization.

And dehumanization is what allows evolutionarily members of one tribe to kill without remorse members of another tribe. And we've muddled through it for eons. But here's the problem. When that last went out of control in the 20th century in two disastrous world wars, we had not even invented the atom bomb yet. Today we have all kinds of AI enabled weaponry, weapons of mass destruction. And it's not just the drones that we're hearing about and the killer satellite, the the satellite targeting, the killer robots, the hypersonic missile navigation, cyber warfare.

It's the computational biology that has now democratized commoditized genetically engineered bioweapons. The Rand Institute has a great report this past year about this. What used to be a billiondoll endeavor in the 1990s is now a commodity that anyone can build in their living rooms out of commodity parts and ingredients. There's no file materials to track and trace. In a hyperpolarized world where everybody's thinking, "Oh, that's the other side is evil. We got to destroy them before they destroy us." In a world where you have all kinds of very angry people, people who have lost

hope, desparing people, would you take a bet that not a single human will hit their own personal weapon of mass destruction? That'd be a terrible bet. We can't stop school shooters. We can't stop 911. And yet, this is the situation we currently face. And Remember this. Humanity has survived to date only because of our ability to outrace the blast radius of the weaponry that we invent. I mean, nuclear weapons are terrible, but they take out one giant metropolitan area at a time.

Other humans survive and muddle on. But the blast radius for genetically engineered bioweapons is literally the entire species. COVID was a terrible thing, but it was a mild pandemic, and you still couldn't get away from it anywhere on the planet, let alone genetically engineered boweapons. This toxic cocktail where we unconsciously allow AIS to polarize us online and also give anybody who wants a weapon of mass destruction that could take out the entire species.

That is the immediate clear and present existential risk. Every morning we wake up, we've just rolled the dice again that nobody's going to hit launch. Could be a small group, a disgruntled individual, terrorists, or like even a country that is on the losing end of a war and decides to shift into asymmetric warfare. We cannot go on applying 20th century outdated thinking. And so one of the things that we're really focusing on and this like I don't know if you know this 56,000 TEDex talks per year chapter 12 in raising AI uh won

one of the few editors picks because once you see this you really cannot unsee it. Probably some of you have heard that there are three kinds of information disorder. Misinformation, disinformation, and malinformation. Misinformation and disinformation are falsehoods. They're deep fakes, fake news. The only difference is disinformation is being spread by someone you call a bad actor, which means someone you don't like because everybody is not a bad actor in their own heads. It means the other side.

Whereas misinformation is being spread by your crazy uncle, your fringe friend. Malinformation is also being spread by a quote a bad actor. But it's not a falsehood. It's a halftruth. The omitted half though makes it even more misleading because our brains just fill in the missing blanks with whatever we unconsciously imagine that would confirm what we already wanted to believe. And we mistakenly think, "Oh, I've got even more evidence for what I wanted to believe." And that hardens our positions.

We have two distinctions going on here. If you have two distinctions, what does that give you? That's a 2x two grid. How many quadrants does a 2 x two grid have? Not three. What happened to the missing quadrant? This is ironic because the same people who study how AI amplifies information disorder online have fallen into their own cognitive bias trap unconsciously, which is the rule of threes. You you know this one. every slide has to have three bullet points, that kind of thing. It's not totally illogical, right? Unconsciously, we just

assign greater credibility to things that come in threes. This is the actual reality. And that missing quadrant there didn't even have a name for the last 20 years that people have been studying how AI amplifies information disorder. What is it? This is the even more misleading halftruths that are being spread not by a bad actor but by your crazy uncle or your fringe friends. I call that neg information because it's the negative space of information that's been hidden from us in negligent ways.

And that is really important for us to look at because AI is amplifying NEG information. We don't know what we don't know. We're not aware of what the AI algorithms are not making us aware of. But that is the most crucial question for humanity in our entire evolution. Because if we keep on allowing AIs to make decisions in ways that drive us to not understand each other and throw us into conflict where everyone is holding a weapon of mass destruction that could take out the entire species, we're toast.

And it's natural for us not to see this because there's nothing in our memories that we can refer to as an analogy. This is another cognitive bias called the availability heristic. There's nothing available to us in our memory that we can draw a parallel to. And so I want to take this to the topic of this whole gathering funding the commons. What is a commons? Well, it's a resource like knowledge or the internet shared by a community and the community manages it through some kind of collective practice. Now, these practices follow norms and rules that are developed by the community for

sustainable management and equitable access. But I want you to think about this. Communities form around ideas that capture viral attention in the 21st century. What determines which ideas do and don't spread? Well, that's AI. Over the last 20 plus years, machine learning and deep learning has taken control of what ideas do and don't spread, do and don't capture attention. The AI algorithms behind social media recommendations, search engines, news feeds, and chat bots hundreds of times a day are deciding what will never pop up on your screens. And I call this algorithmic censorship.

without judgment. Just as a fact, none of us can review billions of new items per day, per minute. We depend on the algorithms to decide what we will never know. And we don't even know what we don't know. It's so insidious that we don't even ask. It doesn't occur to us. I am sure that you'll constantly hear that humans need to watch out. we don't lose our autonomy to the AIs. But we already lost that battle 20 years ago.

Ever since Facebook and YouTube and Google, Amazon, Reddit came out. Losing control of AI is not about the physical world, even though we tend to visualize Hollywood's killer robots demolishing our cities. We lost mind control 20 years ago. Let me put this in perspective. AI is determining which communities form and which communities do not form. In other words, AI is the 21st century's knowledge commons.

That makes it a metacoms that has become the mother of all commons. And here's the crucial question we're not asking. What norms and rules are these uber powerful algorithmic sensors using when they decide what ideas do and don't get attention? When they decide what communities will and won't form, who's determining the criteria, the principles, and values by which these algorithmic sensors make those decisions?

Right now, it's a handful of obscure machine learning engineers or tech executives. It's not the people. It's not even elected representatives who are of the people, by the people, for the people. Guys, in a commons, the norms and rules for sustainable management and equitable access are supposed to be getting developed by the community. And in the mother of all comments, it's even more important that the community is deeply involved in de developing the norms and rules for algorithmic sensors. Because the mindless decisions our algorithmic sensors have been making over the past 20 years have ripped apart our social fabric domestically and

internationally. This is not survivable when AIs have commoditized weapons of mass destruction. The elderly sensors that are making such consequential decisions have to be mindful of how humans feel about the consequences of the AI's choices. And to do that, AIs need the cognitive empathy, not the emotional empathy, the actual conscious cognitive empathy to be conscious of how their decisions make humans feel. AIS need to be mindful when they're making those decisions because they're influencing us in such powerful unconscious manipulations of our psychology.

We don't have the bandwidth to monitor all the algorithmic sensors decisions. That makes it even more important that we have difficult, honest, meaningful conversations in a collective process to answer the most existentially crucial questions humanity has ever faced. What are the criteria, principles, and values by which AI should be deciding which each of us will never know? >> How do we nurture that in the AIS? Establishing that mother of commons is the focus of my life's work and of the team of our new empathetic AI institute.

We're here because I believe in the commons. I believe we should decide together. We need to build the collective movement to center these questions. And we have a plan. Please join us. Please join our email newsletter where we're going to engage and ask everybody here's how you join. What can you do? Uh it's just a substack for ease. Uh please join. All my socials are over here as well. Thank you very much.

>> Thank you Kai. [music] >> [music]