What this covers
Kevin Kelly and Russ Roberts explore advice, technology, and artificial intelligence across this conversation, ranging from practical wisdom on time and money to fundamental questions about whether AI poses an existential threat. Kelly draws on his book Excellent Advice for Living to argue that the most useful guidance works by activating lessons people have already half-learned rather than introducing entirely novel ideas. The broader sweep moves from psychology and philosophy—how deep listening works, why controlling time matters more than wealth—into technology policy and the prospects of machine intelligence. Throughout, Kelly develops a counterargument to contemporary AI anxiety, one grounded in claims about how intelligence actually operates in the world.
The conversation turns most substantive when examining AI directly. Kelly contends that fear of a runaway Singularity rests on unfounded premises: that intelligence is unidimensional and scales like a simple ladder from animals to humans to machines. He argues instead that intelligence is multidimensional and alien, which is precisely why AI has value—humans working alongside artificial aliens can accomplish what neither could alone. On more immediate concerns, he proposes comparing social media's actual harms against prior technologies rather than imaginary baselines, and advocates the proactionary principle of evaluating technologies by evidence of real-world use. Contentious ground emerges around whether creativity requires consciousness, whether emotion does, and whether ChatGPT's errors expose a crisis in epistemology we must solve or simply a mirror of human fallibility we've always navigated. Kelly's optimism—which he presents as deliberate rather than temperamental—stands against the mythical, he argues, religious tone of superintelligence prophecy.
Kelly argues that compressed aphorisms work by crystallizing lessons we've half-learned rather than teaching new ones, and that AI fears are misplaced because machine intelligence is multidimensional and 'alien' rather than a single runaway scale that will surpass and replace us.
- Advice works best by activating a lesson already emerging in someone, not by introducing a wholly new concept
- Intelligence is multidimensional and not increasing exponentially, so superhuman AI takeover is mythical not evidence-based
- AIs are 'artificial aliens' whose value comes from being unlike humans, and they will help us become better than ourselves
AI should be understood as artificial aliens whose different substrate and thinking offer genuine value through collaboration rather than replacement.
- We cannot become better than human by ourselves—the barber paradox of getting outside ourselves—but we can do it with the help of the artificial aliens we create, who will help us go beyond ourselves; this is Kelly's central prediction and reason for excitement.
“We're going to make artificial aliens, and we're going to create these other beings, and they're going to help us go beyond ourselves. We can't do it by ourselves. This is the paradox, is that by ourselves we would be unable to make a better-than-humans, but we can with the help of these artificial aliens that we create.”
- AIs should be understood as artificial aliens—running on a different substrate, they will think and create in ways that are legitimate but 'off' and unlike humans; this alien difference is precisely their value, letting us prove and make things we could not by ourselves, working with them as a team rather than being replaced.
“in every case, they're alien. The way that they do things because they're running on a different substrate than what we run on, their creativity is legitimate. It's really creative, but it's off. It's different than us. ... That's the main reason why they're valuable, is because they're not doing it like humans.”
- Creativity is a lower bar than assumed—it does not require consciousness; machines and systems can produce at least a lowercase version of creativity, just as image generators do, showing creativity is more elemental and mechanical than humans believed.
“we have always reserved creativity as something that you had to be conscious first to be creative. And, it turns out, no, you don't need consciousness at all. You can make creativity, at least a lowercase version of it, in a machine and a system. So, creativity is not the high bar that we thought. It's actually low bar.”
- Emotion, like creativity, is a low-level elemental capacity that does not require consciousness or self-awareness; animals that are barely sentient have emotions, so artificial emotion will arrive without machine consciousness.
“the next surprise that we're going to have is emotion. ... Emotion is very elemental. Look at your pets. Look at any animals. They aren't sentient very much and they have emotions. And so, emotions, again, is a lower down, more mechanical thing.”
- There is no evidence intelligence is a single dimension that scales like decibels from mice to humans to AI; intelligence is multidimensional, graduated, and a huge possibility space whose ingredients we don't even understand, so the premise of one runaway intelligence scale is unfounded.
“this idea that intelligence is a single dimension like decibel... there's no evidence whatsoever that intelligence is a single dimensional--like, it's multidimensional. It's very, very graduated. It's a huge possibility space. And, we have no idea really what those ingredients in dimensions even are.”
- ChatGPT's autocomplete mechanism shows that writing a sentence or essay is a more primitive and elemental thing than we thought—not as big an achievement as imagined—just as chess turned out to be more mechanical than expected.
“what it means is that... actually writing a sentence or writing an essay is actually a more primitive and more elemental thing than we thought. ... Just like playing chess turned out to be a more mechanical thing than we thought.”
Technology policy must be grounded in evidence of actual use and real harms, not imaginary third-person fears or precautionary restriction.
- Basing technology-use policy on the rise of US youth distress is dangerous because the data is primarily US-specific, the US is unusual in many ways, and the correlation lacks comparison to baselines like in-person bullying.
“it's primarily the United States and there's so many weird things about the United States that I think it's dangerous and misguided to decide on the policies of technological use based on how U.S. children are using it.”
- When evaluating a new technology's harms, one must always ask 'compared to what?'—comparing social media's effects against prior media like cable TV, talk radio, or newspapers rather than against an idealized baseline.
“what I always say about technologies, a new technology, is we always have to say about new technology is: compared to what? ... There's some new harm in social media: compared to what? Compared to Fox News? Compared to cable TV?”
- Technology policy should be based on the proactionary principle—constantly evaluating technologies based on evidence of how they are actually used and what actually happens—rather than the precautionary principle, which restricts based on imaginary harms to a third person.
“I don't believe in the precautionary principle at all. I think it's a really misguided idea. I believe in the pro-actionary principle--that's my term for it--which is things where we're constantly evaluating things based on evidence of how they're actually being used and what actually happens, rather than some imaginary harm because of a third person.”
- Social media is barely 7,000 days old and is still an infant; we don't yet know what it is good or bad for, so policy should be based on accumulating evidence rather than imaginary harms.
“Social media is hardly 7,000 days old. It's still an infant. We don't really know what it's good for or we don't really even know what it's bad for yet. But, let's base it on the evidence rather than the imaginary harms of a third person.”
- When people complain about technology like social media, they are often really complaining about capitalism—the consumeristic, monopolistic, profit-driven overlay—rather than the technology itself.
“when a lot of people are complaining about technology, they're really complaining about capitalism. That's what they're unhappy with. It's not the technology itself, but this capitalistic, consumeristic aspect of this.”
- Technology panic cycles are characterized by 'third person' complaints—people are not saying the technology harms them personally, but expressing concern that it may harm someone else (e.g., artists worried about other artists losing jobs to AI rather than themselves).
“most of the complaints are not about people. They're third person complaints. So, the artists--no artist is saying, 'I lost my job to AI.' They're concerned about some other artist who may lose their job.”
Superhuman AI takeover scenarios lack evidence and should be treated as low-probability risks, not drivers of everyday policy.
- Whatever metrics we have for AI capability are barely increasing even linearly, while the compute, chips, and CPUs needed to produce gains are increasing exponentially—so it takes more and more resources to move the needle a little, the reverse of runaway exponential intelligence.
“whatever metrics we have are not increasing exponentially. They're hardly even increasing linearly right now. What's increasing exponential is the amount of resources or chips or CPUs that you need to make that stuff. So, it's kind of actually reverse, where it's taking more and more and more compute to actually move the needle just a little bit.”
- The fear of a recursively self-improving superhuman AI (the Singularity) that displaces humanity rests on no evidence; it should be treated like asteroid risk—a greater-than-zero chance warranting a small dedicated effort, but not something that should influence day-to-day AI policy.
“In my investigations of that kind of logic and that argument, I could find, again, no evidence that that was happening. ... I equate it to, like, asteroids... there is a greater than zero chance that could happen. But we're not... altering what we do based on the fact that we could be wiped out by an asteroid.”
- If an AI has 'a mind of its own' and does things we dislike, we can unplug it; the doomsday claim that takeover would happen instantly or behind our backs ignores the many evident precursors that would have to occur first.
“It's very easy to unplug things. Part of their argument is that it happened so fast that we don't have a chance. And, that's this exponential thing that I don't see happening. ... there are so many things that have to happen before that would happen that would be very evident that that's where it was going.”
- The vision of superhuman AI intelligence is mythical if not religious—a belief not grounded in evidence about AI or the world—and resembles the way medieval theologians described a God-like being.
“this vision of the super-human intelligence is mythical if not religious. It's a belief. It's not really based on evidence of what we know about either AI or the rest of the world.”
Personal wealth and success come from controlling your time, pursuing distinctive work, maintaining long vision, and deliberate self-improvement through discomfort.
- Choosing control over your time rather than maximizing money is itself a form of wealth; Kelly lived on little money but had freedom over his time, and only late in life realized he had been wealthy all along because of it.
“it was only later on, late in life, that I kind of realized that that was a kind of wealth to have control of your time versus money; and that I actually had been wealthy all my life because of that.”
- Rather than aiming to be the best, aim to be the only—pursue work so distinctive that it's hard to even name, where no one else could or would do it; this removes competition, lets you work at your own pace, and is identified by which ideas keep returning to you that no one else will take.
“Don't aim to be the best. Aim to be the only.”
- People overestimate what can be done in a couple of years and underestimate what can be done over a decade; taking the long view on investments, friendships, and relationships gets a lot done, while trying to get rich quick fails.
“we overestimate what can be done in a couple of years and underestimate what can be done in a decade or doing it for 10 years. ... if you take a long view for investments, for friendships, for relationships, you can get a lot done. If you're trying to get rich quick, it's not going to work.”
- You tend to behave as the average of the five closest people around you, so deliberately surrounding yourself with people you respect and admire—not for what they know but for how they behave—makes you better.
“you are kind of [?] to behave the average of the five closest people around you. And so, I did try to surround myself with people that I respected”
- Embracing discomfort—as in travel, where you leave things behind and put yourself out—is good because it forces you to look differently, think differently, and reevaluate what you think is important; personal growth can be gauged by how many uncomfortable conversations you are willing to have.
“that discomfort is good because it's forcing you to look differently, to think differently, to reevaluate what it is that you think is important.”
- Success requires walking the line between never giving up and giving up when it's time, since you can't get things done unless you give some things up; discerning the difference is a skill, often requiring friends to help you tell which situation you're in.
“you've got to walk this thing of never giving up or else giving up when it's time to give up. And, then having your friends help you just discern what the difference is. Because you can't really get things done unless you give things up.”
AI development raises deep epistemological and ethical challenges requiring new infrastructure for trust and better-than-human values we haven't yet defined.
- ChatGPT's tendency to produce both highly trustworthy and confidently wrong answers exposes a profound, multi-decade problem—how we decide what to believe and what counts as evidence of truth—that forces us to build infrastructure for distinguishing the trustworthy, ultimately addressing a deeper epistemological problem we had already but never solved.
“there is one thing that ChatGPT has uncovered that I think is actually very important... this issue of whether you can trust the answers or believe the answers. And, that's a really, really fundamental problem that spills over to the general question of how we believe anybody and what we take as evidence of something being true”
- Although we train AI on flawed human experience and ethics, we will demand the AIs be ethically better than humans—less racist and sexist—which is codable in principle, but the deep challenge is that we don't actually know what better-than-human ethics looks like.
“we are going to demand that the AIs be better than humans. That's what we're demanding. ... The challenge is we don't know what better-than-human looks like. ... that's the exciting challenge of growing up, is to make our AI robots better than us because in the end, they're going to make us better humans.”
- AI is technologically inevitable—it will arrive in some form and no amount of regulation can stop its coming—but society retains an enormous choice over its character, ownership, and regulation.
“a lot of the changes like AI, AI ahead of us is inevitable. We're going to have it in some form or another, and no amount of regulations or anything else is really going to stop it. We have an incredible choice about the character of these AIs and who owns them and how they're regulated”
- Network effects will operate in AI—certain AIs improve as more people use them, forming natural monopolies—so the largest AIs will become services delivered like a utility, which may warrant utility-like regulation, public ownership, or treatment as a commons.
“network effects will be at work in AI. ... the more people use them, the better they'll get. ... you will see these very large AIs explode through use; and they'll have kind of a natural monopoly--that it becomes a service. ... it might become a utility, I think there could be some regulation or maybe even some utility-like public ownership of it”
- Current AI relies on neural nets needing billions of data points, unlike a toddler who learns cat-versus-dog from a dozen examples; once few-shot training arrives, training sets will be tiny and artists will clamor to be included to be counted among the few influential ones—making intellectual-property objections a distraction.
“almost all the AI that we have so far has been on neural nets, which require billions and billions of data points to train. ... a toddler will learn the difference between a cat and dog with just 12 examples. ... at that point, every artist will be clamoring to be included into the training set”
Optimism and wisdom emerge from deliberate practice—constructing evidence-based perspective, crystallizing emerging thoughts into aphorisms, and deep listening.
- Kelly's optimism is not merely temperament but a deliberately constructed, evidence-based perspective grounded in reading history—comparing what humanity had in the past to what it has now and the momentum of that improvement.
“my optimism is actually not just my disposition is, is actually a carefully constructed perspective. It's a deliberate perspective. And, I am more optimistic now than before because I've been reading more history.”
- Advice in the form of short aphorisms works best not by introducing an entirely new concept but by crystallizing a lesson you have already begun to learn but not yet articulated—soldering an emerging thought onto your brain and giving you a handle to grab when you need it.
“it works best when you have already kind of learned the lesson but haven't articulated it yet and don't have a way to handle it... You're actually having it soldered onto your brain in some ways, the thought that was emerging already”
- When confronting a two-sided black-and-white controversy, ask 'What is the third side?'—triangulation can break the dilemma and reveal a solution that the binary framing hides.
“when there's a controversy about two sides... I always say, 'What is the third side?' There's some third side to it that triangulation can break that dilemma and make it easier to see maybe there's a solution in there.”
- Aphorisms and mantras inculcate tiny habits that can grow into larger ones; they are reminders that help you begin a habit, not magic pills or incantations that produce behavior simply by being said.
“they're not magic pills that you take and then you suddenly do them because you've said them, incantations and stuff. They are habits that you have to have.”
- Listening is a superpower; when listening to someone you love, keep asking 'Is there more?'—at least three times—because only after the obvious and cliché answers are exhausted do you reach the real story or heart of the matter.
“you need to go at least three times, but you would say, 'Is there more?' ... You've got to go all the way and let them go through the obvious stuff and the cliché stuff. And, then at the third, 'Is there more,' then you'll hear the real story.”