Reading The Age of AI

When I heard that Henry Kissinger's new book, "The Age of AI: And Our Human Future", had come out, I bought it. The book deals more with politics than with technology or economic trends, so there was a lot of interesting material. Kissinger, one of the most influential secretaries of state in American history, wrote it with former Google CEO Eric Schmidt and MIT dean Daniel Huttenlocher. Drawing on four years of discussions, they stress that AI's progress needs constant questioning and regulation.

In particular, the number of people building AI keeps growing, but only a handful work on its social, legal, philosophical, psychological and ethical questions. If content that AI generates, transmits and filters stirs up hatred and division, it could become a serious threat. So social order has to be kept through legitimate means, and that requires keeping a balance between AI and humans.

The book is a philosophical look at how AI will change the role of humans, and at how the international order will shift once AI is applied to military technology. It gave me a good chance to think again about human value and meaning.

First, the authors argue that AI's emergent behavior may affect us most fundamentally. Emergence means producing new results no one predicted. AlphaZero beating opponents at chess in a completely different way is the classic example. They also say a new mechanism was discovered when deep learning was used to find the antibiotic halicin.

Humans get stuck in existing patterns of thought because of instinct, prejudice, authority and self-interest, which makes it hard to try new things. AI is free from those mechanisms, so it's more likely to try something new. AI has the potential to work on the world of consciousness, and on things that are hard to pin down scientifically, and to offer solutions. As AI opens up new possibilities, humanity may come to understand things it couldn't grasp for thousands of years. Of course, AI won't be free from human biases like confirmation bias either. Still, unlike humans, AI has the potential to explore the world freely and try new things without being trapped in old patterns of thought.

For important decisions like national security, though, human judgment is more appropriate than AI's. This argument is backed up in the book "After the Singularity". It argues that ritual and principle give humans the responsibility to make important decisions, and that AI can't take that over. AI can perform brilliantly in structured settings like board games, but in complex, ambiguous situations like national security, human judgment is still needed. Humans can make moral judgments through ritual and principle, and AI doesn't have that ability. In national security, for example, AI can't accurately read an enemy's intentions or make an ethically sound decision. So those decisions have to be made by humans.

AI can, of course, be used as a tool to supplement human judgment. It can analyze huge amounts of data and surface information people might miss. But AI can't replace human judgment. It can perform brilliantly in structured settings like board games. For important decisions like national security, though, human judgment is more appropriate than AI's.

And as AI uses its creativity to make new things, we may end up in a situation where it's hard to tell reality from fiction. With progress in generative neural networks, AI has gone beyond being a simple tool and become a source of possibility. But this progress can also have a side effect: it gets harder to tell whether something was made by AI or not. As text and image generation improve, I've felt AI's potential and its risks more sharply.

Now that programs can be automated and generated from simple code structures, the programmer's role is bound to change. As AI gets cheaper to build, it's expected to replace humans in most future work. But since AI still can't copy physical movement very well, it's more likely to replace work done with the head. New jobs that never existed before are expected to appear. And since AI will keep being developed to do better work, this change won't be temporary. It will continue over the long term. So along with these new jobs, we need more conversation and debate about employment and about what AI replaces.

Yuval Noah Harari and experts around the world predict that AI will replace many human jobs. So the world may need a completely new economic model and social system, like universal basic income.

For all sorts of reasons, from human nature to today's economic structure to biases like confirmation bias, people are conservative about the current economic system. But in the span of human history, today's economic system has only lasted a short time. So we have to think about what new economic model we need for a stable human life. AI is a tool that can bring huge change. We'll feel both hope and fear about AI at the same time, and we'll need to talk about how to improve our economic model.

Lines that stayed with me

To finish, here are a few passages I found striking.

Keep in mind that today's innovation is only the first step in what AI will achieve. AI will break through every limit that's a problem today before we even notice. (Bill Gates)

In the world AI brings, decisions will be made in three ways: 1) by humans, 2) by machines, 3) by humans and machines working together.

In the Enlightenment, Descartes's "cogito ergo sum" (I think, therefore I am) gave rise to the belief that the thinking mind is humanity's most important ability, and that this is why humans deserve to stand at the center.

Four centuries later, we have to ask ourselves: if AI can "think," or imitate thinking, then who are we?

AI will push humans out of almost every job. Creating new jobs isn't the answer, because AI will eventually do those jobs better than humans too. We need a completely new economic model. (Yuval Noah Harari)

According to Kant, human reason can understand reality deeply but is necessarily incomplete. Human perception and experience filter, organize and distort everything we know.

We urgently need to build a sophisticated dialectic so that people readily question the interactivity of generative AI. The point of that questioning isn't just to justify or explain AI's answers. It's to examine them.

We need to build the ability to test AI systematically, with a shared skepticism, to judge whether its answers can be fully trusted or how far they can be trusted. That means consciously reducing our unconscious biases through strict training and constant practice.

Over time, people may come to prefer digital assistants over humans. Other people don't pick up on our tastes as readily, and there's more "disagreement." If so, we may rely less on human relationships. Then how will the important experiences of childhood take shape?

AI is a tool humans can control. The problem is that someone won't set the safety limits we set. (Sam Altman, CEO of OpenAI)

If a free society relies on AI-powered network platforms that generate, transmit and filter content across borders, and those platforms stir up hatred and division, even unintentionally, that society faces a threat it has never faced before and will have to police its information environment in ways it never has.

This is an urgent problem, but solutions that rely on AI raise serious questions of their own. We always have to weigh human judgment and AI automation on the scale and look for the right balance.

By the early 2030s, we'll be able to connect the brain's neocortex to non-biological intelligence. Our intellectual limits will disappear, and human intelligence will extend beyond the brain into the cloud. (Ray Kurzweil)

Until now, no technology has been dual-use, easy to spread and potentially enormously destructive all at once. Railroads carry goods to markets and soldiers to battlefields, but they have no destructive potential of their own. Nuclear technology is largely dual-use and can produce terrifying destruction, but it needs complex infrastructure, so governments can control it fairly reliably. Hunting rifles are everywhere and can be used by both soldiers and civilians, but their limited power means they can't cause destruction on a strategic scale. AI breaks this pattern.

This is an age in which human intelligence teams up with artificial intelligence to take on work at the national, continental and global level. To understand this change and build an ethical framework to guide it, scientists and strategists, politicians and philosophers, clergy and CEOs all need to bring their efforts and wisdom together, at the national level and at the global level too. Now is the time to define how we will work with AI to explore reality.

Originally written in Korean on Brunch, November 24, 2023.