Making Sense - Sam Harris - June 04, 2026


#479 — When Robots Take Over


Episode Stats


Length

14 minutes

Words per minute

139.328

Word count

2,083

Sentence count

123


Summary

Summaries generated with gmurro/bart-large-finetuned-filtered-spotify-podcast-summ .

In this episode of Making Sense With Sam Harris, we talk to Vinod Kosala, a venture capitalist, technologist, and venture capitalist. In this episode, we discuss the risks and benefits of artificial intelligence (AI) and how to deal with them.

Transcript

Transcript generated with Whisper (turbo).
00:00:00.000 you're listening to making sense with sam harris this is the free version of the podcast so you'll
00:00:06.400 only hear the first part of today's conversation if you want the full episode and every episode
00:00:11.360 you can subscribe at sam harris.org there are no ads on this show it runs entirely on subscriber
00:00:17.980 support if you enjoy what we're doing here and find it valuable please consider subscribing today
00:00:22.940 i am here with vinod kosala vinod it's great to see you thanks for coming on the podcast
00:00:29.160 It's great to be here.
00:00:31.280 So how would you describe your background in business and tech before we jump into all
00:00:37.440 things in your wheelhouse here?
00:00:39.320 You know, I can do the business part, but I've never really had an interest in business.
00:00:44.460 I'm sort of just very curious about tech and the impact it can have.
00:00:49.480 So most of my focus is what technology is coming down the line and what impact can it
00:00:55.320 have and what does it take to do that?
00:00:57.340 So it's more about making things possible.
00:01:00.680 I like to say, I like to imagine the possible and then try and make that happen.
00:01:06.140 That's sort of my main goal in life.
00:01:08.980 But you were a technologist first and then a venture capitalist second, correct?
00:01:14.680 Yeah.
00:01:15.000 In fact, funny thing is I've never actually in 40 years called myself a venture capitalist.
00:01:20.460 Oh, sorry to insult you to your face.
00:01:25.640 So what do you call yourself?
00:01:26.960 I would say I'm in the business of helping entrepreneurs build companies with new technology.
00:01:33.400 And I'm only curious about technology.
00:01:35.840 The other business applications don't interest me very much, or business-only applications.
00:01:42.840 So we're going to talk about AI and its social and even political implications.
00:01:50.100 But I guess a big-picture question for you to start.
00:01:54.020 What most concerns you about the economy at this moment?
00:01:59.740 Is there anything you're worried about?
00:02:01.120 Yeah.
00:02:01.780 Look, when I look forward 10 years from today, two big things stand out.
00:02:08.680 One, we have AI progressing rapidly, and I'm not worried about its capability.
00:02:16.320 Almost anything we want it to do that's economically valuable, it'll be able to do over the next
00:02:23.080 10 years.
00:02:24.020 It doesn't matter whether we are talking about robotics or AI or any, frankly, function that the human brain can do.
00:02:33.700 What worries me a lot is if we're going to maximize the impact of that AI for good purpose, societal good,
00:02:44.700 we are going to have a lot of disruption and change.
00:02:49.460 And my biggest worry, to get to your question, is AI may not be permitted because of that disruption.
00:02:59.200 So politics is most likely the biggest impact on AI the next decade, more than anything to do with technology or capital or data centers or power.
00:03:12.180 So you're fearing regulation that proves unwise?
00:03:17.360 Does that summarize this concern?
00:03:19.960 I would say a simple idea, for AI to be fully effective, we should have large-scale job
00:03:27.180 displacement.
00:03:28.740 If we let things happen, just happen to be most efficient in a capitalist sense, we'd
00:03:36.140 probably get to 50% unemployment or underemployment by 2035 or so.
00:03:43.340 that obviously cannot happen without a lot of political pushback. So the answer is we have to
00:03:52.620 do something radically different for us to accept the level of disruption. And that worries me
00:04:00.380 because politicians will take short-term advantage of things like job displacement.
00:04:07.140 My nightmare would be Bernie Sanders or AOC get elected president.
00:04:12.080 That'd be about as bad as Trump getting elected president.
00:04:15.500 So you could get A.I. being slowed down by politics of job displacement and fear.
00:04:25.080 A.I. is as feared, as popular among general people as ISIS.
00:04:30.460 And if you look at that perspective, then it's worrisome.
00:04:35.000 So if memory serves, and I think this is implicit in what you just said, you're not very concerned about the so-called alignment problem, the idea that we'll build superhuman AI that could be unaligned with our interests and pose some kind of existential risk to us.
00:04:52.340 I wouldn't say I'm not concerned. First, there's two versions of the alignment problem, the way it's traditionally defined, which is you align with human interests. And this is what all the labs are trying to do. The more egregious version of that is an AI gets so dominant, it decides to take over the planet.
00:05:15.020 Look, none of these risks can be completely discounted.
00:05:19.460 So I won't dismiss that.
00:05:21.760 But my much bigger concern would be strong AI in the hands of Putin or presidency.
00:05:28.720 On a relative basis, which is the larger risk, maybe even the largest expected value in terms
00:05:36.580 of impact slash, clearly, I worry about the West falling behind what we would consider
00:05:44.260 bad actors on AI and them using it to dominate us. Right. But in order to weight the risk in that
00:05:53.520 way, you must put a pretty low probability on the existential concern. Because we're clearly in an
00:06:00.320 arms race with China, but if you thought there was a reasonable chance that we were racing toward a
00:06:05.880 cliff, you probably wouldn't be worried about China winning. But it sounds like you think that
00:06:11.560 The expected value on China winning is worse than the expected value on the low probability
00:06:17.760 that we're headed toward a cliff.
00:06:18.880 I assign a low probability to existential risk.
00:06:22.820 Right.
00:06:23.340 Okay, well, let's talk about, let's just launch into it and let's talk about these concerns.
00:06:27.740 So in success, you're not somebody, it sounds like, who's given to the usual economic happy
00:06:36.340 talk that AI will not be job-displacing or job-canceling because everyone will just find
00:06:43.180 new jobs. By analogy, the Industrial Revolution moved everyone from the farm into the factory,
00:06:52.920 and then the Information Revolution moved everyone from the factory into the service economy,
00:06:59.120 and everyone found better jobs, and nobody's putting horses to work anymore, and we're all
00:07:03.600 better for it. You seem to agree with many people, including myself, that if we build AI that truly
00:07:10.980 is general, which is to say that it is a replacement for human cognition across the
00:07:17.940 board, well then by definition, these crazy productivity gains are labor canceling.
00:07:24.880 I absolutely agree. I think the thing to keep in mind is everything we've had before,
00:07:31.080 and people say it's always happened that more jobs get created.
00:07:35.060 My personal view, everything we've had before
00:07:38.280 is a tool for humans to use and to leverage.
00:07:42.160 Back a long time ago, we did the steam engine
00:07:44.620 and then the electric motors and the amplified human muscle.
00:07:49.100 And everything else since has amplified the human brain.
00:07:53.080 But when the external AI brain is better than the human brain
00:07:57.520 in almost every function,
00:07:59.260 I don't think we can say the future will be the same as the past.
00:08:04.760 We'll invent new jobs.
00:08:06.720 Now, there's a finer nuanced version of that we can talk about.
00:08:12.960 If, in fact, we get large-scale job displacement, which I think is likely, within a decade,
00:08:21.120 and every corporation in America will be failing if they haven't quadrupled their revenue per
00:08:28.180 employee, or for a given amount of revenue, have one quarter the number of employees.
00:08:33.440 There is another version.
00:08:35.040 So in that world, what do humans want?
00:08:39.160 I think the key driver will not be utility.
00:08:42.740 I can buy this coffee mug I have in my hand from China pretty cheap.
00:08:48.280 It'll be for human preference.
00:08:50.520 That means if it's made by a human, I have provenance, I have a story attached to it,
00:08:56.720 I'll prefer it. So human preference may play a large role, in which case we might end up,
00:09:06.200 and I think the most likely scenario for 2035 is far, far fewer corporate jobs,
00:09:14.100 but 50 million more micro-entrepreneurs in the US. Let's just take one geography.
00:09:20.620 And these micro-entrepreneurs, because of AI, don't have to know anything other than
00:09:27.120 their skill.
00:09:28.180 I grow the best flowers.
00:09:29.660 I carve the best wood.
00:09:31.000 I bake the best muffins.
00:09:32.960 I'm best with dog walking.
00:09:34.860 And humans will prefer that.
00:09:36.660 And so human preference, not utility, starts driving it because all utility is in a hugely
00:09:43.820 deflationary economy going to zero.
00:09:46.020 And so you have this human preference element, which will allow many more people to be their
00:09:52.760 own boss, be a micro-entrepreneur, do the thing they love most.
00:09:58.100 I think that's one and probably the most likely scenario.
00:10:03.160 There's other scenarios possible, but lower probability, I would guess.
00:10:07.420 So, but under those conditions, you're still talking about fewer jobs and an economy wherein
00:10:13.020 and many people have lost their jobs not to find them replaced.
00:10:17.480 So there has to be some redistribution of wealth.
00:10:21.140 I mean, I want to talk about wealth inequality here specifically,
00:10:23.780 but this is a picture of kind of a vast concentration of wealth.
00:10:27.860 What is the unemployed radiologist or the unemployed truck driver to do under these conditions
00:10:34.640 when they want to hire all of these bespoke services where human provenance is still so attractive?
00:10:40.740 So let's talk about two separate things. Where will income come from to survive? Separate from
00:10:48.680 what will humans want to do? Humans will want to produce goods and services that other humans
00:10:55.440 prefer because they're provenance, because of that human skill at baking or pick your favorite.
00:11:02.560 Now, that will be a source of income. It'll be a source of pride and dignity for every human being
00:11:09.900 because they're their own boss.
00:11:11.820 And we know people are happier when they are their own boss.
00:11:15.320 Let's look at the other side.
00:11:18.380 The first thing I would say, the vast majority of jobs in the US, and these jobs are better
00:11:24.320 than jobs in most other parts of the world, like India, are not really jobs.
00:11:29.380 I call them servitude to survivals.
00:11:33.220 And that servitude, if you're working as a farm worker for 40 hours a week for 40 years
00:11:41.460 or an assembly line worker for 40 hours a week for 40 years or till your back gives
00:11:47.420 out and you have injuries, is not human dignity.
00:11:52.080 People like to equate it to that, but dignity comes from doing what you love, not the servitude
00:11:58.940 to survive.
00:11:59.680 it's just a new form of slavery to capitalism. It is better than no other job, but it's not
00:12:06.620 exactly human dignity or the most respectable job or one that gives purpose or meaning to these
00:12:14.300 human beings. If they're these micro entrepreneurs, there will be dignity. There will be
00:12:20.700 pride in doing that job. So that's that part of it. So let's talk about the income side of it.
00:12:27.620 there's two periods to talk about, the 2030s and the 2040s. I think starting 2030s,
00:12:38.020 there's a lot of services that become free. An AI doctor will be a dollar an hour,
00:12:43.840 which is compute costs. AI robot, maybe $2 an hour, add an extra dollar an hour for hardware.
00:12:51.500 So almost all labor trends to that cost, resulting in free doctors for most people, free AI education or tutors, personal tutors, free legal services, free financial advisory services.
00:13:10.000 All these things become near free services.
00:13:14.680 And I think we have a hugely deflationary economy for utility goods.
00:13:21.500 That's my best guess today, and the government may actually provide that.
00:13:26.780 In fact, I have a project to start providing some of these services for free in India because
00:13:34.940 the cost is so low.
00:13:36.320 You develop them once, and then there's the cost of tokens, which is declining very, very
00:13:41.220 rapidly.
00:13:42.260 So basic services, with the one exception of housing, which we can get to, are going
00:13:49.500 to be very free and accessible.
00:13:51.500 I believe. Then there's the question of in the transition period.
00:13:55.960 Just tell me how a massively deflationary consequence to unlimited free intelligence,
00:14:03.380 how do we navigate that with respect to the initial massive concentration of wealth that's
00:14:10.200 going to go to all the investors such as yourself who funded this technology? How do we navigate
00:14:16.580 from where we are now with, you know,
00:14:19.960 that concentration seeming to begin
00:14:22.320 across a landscape of labor displacement.
00:14:26.140 Let me get really, really precise.
00:14:28.960 Members can hear the full conversation
00:14:30.580 by subscribing at samharris.org.
00:14:33.160 Subscribers get a private RSS feed
00:14:35.000 you can use with your favorite podcast player.
00:14:37.720 If the president can exempt himself from tax audits,
00:14:43.560 there's just no ethics.
00:14:45.020 You can easily buy this president with a few million dollars.
00:14:50.020 Values globally have been destroyed.
00:14:53.020 Protocols of international behavior is completely dead.