Scott Caravello: We were never going to get fully away from it. Last episode we talked about how we both really like social media videos of house cows. But how, you know, given how realistic AI-generated content is, it can be tough to tell if they're real. It really is a house cow. But after the episode, I knew of a legit account, so I texted you a link to an article about it and you never responded. So I want to know, what, what gives? What gives?
Katherine Forrest: OK, all right, so here's the problem for me and house cows. I go down the rabbit hole, so to speak, and it's hard to come out. So I went down that rabbit hole. I started looking at house cows account, yours and others. I found one called Cow_Hub, which I recommend for those of you in our audience who might be interested in house cows. And so I just want to actually turn the tables here and say I want a couple of hours of my life back because what you did was you actually put me down the road of the house cows. All right, so if that's the biggest bone you've got to pick with me, I'm going to count myself lucky.
Scott Caravello: That is, yeah. Yeah, yeah, yeah.
Katherine Forrest: Great. Before we actually head into our main topic today, which is going to be talking about the Bill Gates essay that has created quite a stir recently on AI, I wanted to just mention to our audience that we had spoken a couple of weeks ago about the OpenAI cyber incident that involved Hugging Face and that there would be more information probably coming out. Well, there are two things I wanted to recommend to our audience for them to take a look at. We're not going to spend any time on these. The first is put out by OpenAI as a technical report, and that's called, and you can find it on the OpenAI website, OpenAI-Hugging Face Incident Technical Report. But I also want people to be aware that there's also an interesting letter which is on the web at openai.com/collective-cyberdefense. So taking a look at both of those might be very useful. And with that, let's talk about this Bill Gates essay that has been all over the news this week. It came out on August 26th, and we're only on August 28th right now as we tape this. And so for those of you who haven't heard about this, Bill Gates published about a 6,000-word piece on his own website called something like Bill Gates Notes or something like that. And he talked about, well, that's a quote, “the turbulent AI era is here. The choices we make now are critical.” And what makes this essay really notable isn't exactly what it says, although there's a lot of meaty stuff in it that is worth pausing on. But it's really, you know, who's saying it. And we've had this now a couple of times where different folks who are very embedded in the tech industry have made a variety of statements about AI. And Bill Gates is now, he's said a couple of things in the past, but this is his most complete statement yet. Let me just give him one more quote. And then I want, you know, sort of have you react to it, Scott. But one of the early quotes, it sort of appears when you pull this up on the web, right at the top in a banner, it says, “the transition to the AI era will be one of the most turbulent times in human history. Right now we are not preparing adequately for that transition.” So that's a pretty big statement from a pretty big guy in the tech industry.
Scott Caravello: Yeah, exactly. I mean, but that sentiment itself isn't uncommon. We see similar statements like that all the time. But to your point, Katherine, about who's saying it this time, and that's because he's spent years talking about all the upside for AI, talked about its potential to cure diseases, about the ability to close gaps in health and education. So it is like, it is a shift in tone that says something about where we're in, because basically he's saying for the first time that his feelings about this technology are complicated. And he even says that he'd get behind a credible plan to slow down AI development globally if one was actually put forward.
Katherine Forrest: Right. And it's interesting because there now are several recent, very recent calls for that kind of slowdown. And so he's adding his voice to it. But the essay is built around a series, not just one, but a series of these quote-worthy sentences that are really interesting that talk about why, I guess, he's suggesting a slowdown, and he sort of builds to it. And part of his building to that is, here's a quote: “AI will either be the greatest equalizer ever invented or the worst source of injustice.” That's a pretty big statement. And he then goes on to say, and is that where we end up isn't set in stone yet and we have a chance to make choices, and those choices are choices we need to make right now.
Scott Caravello: So, let's start with one of the other key pieces, really one of the key pieces that gets at why he would support this plan, the slowdown. And that's why he thinks at this moment is really different from past technological revolutions.
Katherine Forrest: Right. And the usual reassurance we hear for why the AI revolution isn't going to be as disruptive as some may think is because there have been past waves of automation that were predicted to do all kinds of things. But in fact, they've often been job creators, not job eliminators, despite people's concerns. And also the fact that we really just don't know how the AI sort of revolution, if you want to call it that, the AI transformation, how it's going to come out. But Gates says that this comparison to the past actually falls apart this time because the past played out over generations. So when there were past technical transformations that people were concerned about, say, the Industrial Revolution, they played out over generations. And that the new jobs that were created, and that were predicted sometimes to be created, or that were just in fact created, actually needed human brains. What's different now is that AI can do the thinking part. And so if jobs are eliminated, the concept is there's not at all a certainty that the new jobs that will be created will need, you know, a human brain, because AI can see, listen, speak, reason, and eventually do some physical work too. It's doing that with robotics now all over the place, so it doesn't just hit one industry and let the others sort of soak up the workers or even necessarily retool. What he's talking about is that there could be a sort of simultaneous hit to multiple industries at the same time without necessarily a need for a replacement with a human brain.
Scott Caravello: Right. And so then he also adds on to that the fact that the technology is adapting to us. And that's another way to sort of differentiate the, what's projected out of the, this AI transformation versus other historical technological revolutions. Because in the past when that disruption came, people moved into new roles. And this time the technology is going to itself move into the roles. It could move into new roles that are created and then can do it cheaply.
Katherine Forrest: Right, and that's his argument. There's still a lot of people who say that AI is in fact going to create lots of opportunities for people to reskill and to take on new jobs. You know, when we transformed from a horse and buggy world into an automobile world, we may have eliminated those blacksmiths who were shoeing the horses, but we also created a lot of jobs in the automobile manufacturing industry. So, but, you know, his argument is a powerful one when it's coming from someone like Bill Gates. And then this also leads to his sharpest claim about jobs. He says the ones that are most at risk are entry and mid-level jobs. And he points out specifically, you know, sales, customer support, software engineering, things like paralegal work as being among those that are going to be first in line. And then he fills in a lot of the detail about that in the essay itself. And he talks about that in the essay in terms of what's going to happen with jobs and why he doesn't think there'll necessarily be an absorption that will automatically come through new jobs. But he also fills it out in a whole bunch of interviews that have happened over the past several days, for instance, with the MIT Technology Review, where he talked about the time that AI ends up, you know, sort of predictably becoming cheaper and being responsible for fewer and fewer errors, that will cause a transfer of AI into the human role. In other words, one of the things that's preventing AI from taking on more human roles right now is that AI is still error prone in certain ways. And so as we get to a point where those errors are removed, we're going to see even more shifting of jobs and roles and substitution because of that. And he says the real tipping point then, and the one that we should be looking out for, is when we start to realize that, like the frog in the water, that AI is actually not really creating errors, that in fact, it's creating fewer errors than humans, and that time is coming faster than we think. And then you pile robots on top of that, and their cost is, you know, dropping too, and their capabilities are increasing. And he thinks that smart robots competing with people for physical work is going to create also all kinds of job shifts. Think, you know, construction, hospitality. And he predicts that all of that last piece is going to happen by the end of the decade. So that's only a couple of years away.
Scott Caravello: Yeah. And so I do kind of want to talk about that robot piece a little bit more because I think we still see all these videos of robots out there kind of failing horribly. I don't know if you've seen the one recently, Katherine, about the robot in a weightlifting competition in China that lifts the barbell over its head and then stumbles and it's a total mess.
Katherine Forrest: And does it fall apart?
Scott Caravello: Yeah, like it falls to the ground and then kind of goes crazy. I would find the link and text it to you, but I don't think you'll respond, so I'm just not going to.
Katherine Forrest: Don't do that because it's no. Do you know what happened exactly? OK, right, unless it's fuzzy, got some relationship to house cows, 'cause I don't want to see it.
Scott Caravello: Fair enough.
Katherine Forrest: But, you know, I also saw the sort of robot Olympics that happened in China recently where they had the robots, right?
Scott Caravello: So, I think it was part of that.
Katherine Forrest: Oh, OK, right. And so, you know, these robots running on a track, they actually sort of have now exceeded Usain Bolt's fastest time. Some of them fell apart on the way, but some of them didn't. And Gates has something to say about all that as well.
Scott Caravello: Exactly. But so he says that it's going to be this kind of “vicious cycle,” those are his words. Because once the adoption starts and you get savings to lower prices, competitors are going to feel a lot of pressure to do the same. And then even if existing players in a market aren't adopting the technology, startups will. So it's kind of sounding the alarm, because even though, you know, the technology isn't there yet, it's going to be. And he sort of thinks that competitive dynamics in our economy are going to ensure that the disruption happens once robots get to the point where they can in fact perform tasks cheaper than human labor can. And he then goes a step further and connects all of that to the broader societal impacts that come with large-scale unemployment.
Katherine Forrest: Yeah, well, you know, before you even get there, because that's a very serious point. I'm still on the robots that are like running down the track at the Olympics, the Chinese Olympics. And one thing that Gates actually says is don't get fooled. Don't get fooled by these little clips where you see the robots falling down and falling apart and doing ridiculous things, because you should understand that robotics are actually improving right now by leaps and bounds. And I am a big believer in that. In fact, we did a couple of episodes on robotics and I thought, and I still think of, you know, 2026 as the year of the robotics where we're starting to see the dexterity of robotics at lots of different places. You know, we're seeing Boston Scientific having extraordinary dexterity. And, you know, of course, there's the SpaceX and, you know, Optimus, its robots with extraordinary dexterity, and Unitary, the Chinese robotics company, having extraordinary dexterity and capabilities. So as Gates says, as Bill Gates says, don't be fooled by these little clips, your favorite clip of the robot sort of falling down. Those are all over the place. But think about the fact and look at the fact that the capabilities are improving by leaps and bounds. So let me go on to the second bucket that he talked about, which was safety and security. And this is where his background, his tech background, the Gates tech background really shows on the cyber side. He says that the bar is basically dropped. Even somebody with pretty limited skills can now use AI, some of these most capable AI frontier models, arguably to do some bad deeds, and that the resources that are needed keep going down. Now, I would also say that we also have companies working on alignment strategies, very strong alignment strategies, and staying ahead of that. But what he's doing is he's sort of sounding, you know, the alarm, which is, look, this stuff is becoming very capable and we may end up lowering the bar or the kind of skill set that you actually need to take bad acts. And, you know, that may well be true. And it's something that a lot of people are worried about. So we've got this situation where we have these more and more capable models that may result in both incredibly good things and also some dangerous things.
Scott Caravello: Exactly. And so then he makes a very similar point when it comes to biological risk, right? The fact that you could use AI to potentially create a novel pathogen. And to your exact point before about alignment strategy, right, there is a ton of work that goes into making sure that the systems that are made available to the public are safe and in fact don't do exactly that. But it's still unsurprising to see Gates flagging this risk, especially given all of his years working on global health and pandemic work through the Gates Foundation. And then further to that point, right, his worry is the same, that AI is going to give us life-saving drugs and vaccines, but it's at the same time going to make it a lot easier for someone to design a deadly new disease. And so it's a similar contradiction to what you were mentioning on the cyber side, that the helpful capabilities and the dangerous ones can be hard to pull apart. And so then moving on to the third big area that he flags, which is the psychological and social impact of AI, right?
Katherine Forrest: There's a big debate right now about human and AI social interaction, which we're not going to get into. But folks should go ahead and read the essay themselves and also listen to the MIT Technology Review interview, in which Gates expands on the point, several of these points, but that he believes that AI may be difficult to actually steer in a reliable way. And he then harkens back to how they're trained and reward hacking behaviors. We actually need to do a whole episode probably on what reward hacking is. But in short, it's AI trying to achieve something that we are, as humans, are calling a reward, which sort of in a human way, sort of in an anthropomorphized way, suggests sort of an endorphin hit. We're not at all sure that AI thinks of reward hacking as an endorphin hit. But reward hacking is a phrase, a term of art that's used in the AI world to talk about AI being incented to actually achieve certain metrics or to accomplish certain tasks. And he is concerned about that and about alignment. So, you know, it's interesting.
Scott Caravello: Yeah. And he says that he's surprised about the level of concern that he's seeing outside of the AI industry. But he is very clear about the fact that in his view, we should be looking beyond industry to look at solutions to these problems he perceives, right? We shouldn't count on industry to lead the charge. And part of that is because some of the issues are outside of their expertise, and in a democratic society, it's not really their role to solve all these problems anyway. So with all that said, what does he propose for a path forward?
Katherine Forrest: Yeah, well, he's got three ideas which I think are controversial. And, you know, people should look back at what OpenAI put out in April of 2026 in terms of its industrial policy paper, which you can get on the web, because they also had some ideas that they were putting up at that time. And he offers up a robot and token tax and says that, you know, when AI or a robot does work that a person used to get paid for, there might be some sort of tax that is then paid and that that money from the tax would go towards retraining or providing a stronger safety net for the people who are displaced. And, you know, this actually isn't all that new. He's floated a robot tax years ago and he's still a big believer in it. And it's interesting because what we haven't done is provide exactly the same kind of tax on something like automation. There have been robotics used in manufacturing for years and we haven't, which have displaced certain workers, and we've retrained workers, say in the automotive industry, to do different things. But there hasn't been a tax on it. So I'm, I'm not really sure how it differs. But of course, this is just one of those ideas.
Scott Caravello: Yeah. But we've also seen some similar proposals put forward, I think at the state level. But it's again, very interesting to see a voice like Bill Gates getting behind the concept. And so we'll see if it gets a little bit more discussion as we go forward. As part of that, he is also making the point that it's quite difficult to implement, that you wouldn't want the proposal taking such a broad approach that it's also capturing the purely good stuff, right, AI that maybe makes medicine or education cheaper, and you want it, you would ideally want to scope it to AI that's just swapping in for the human job, or what Gates is talking about as, you know, kind of the negative impact of automation.
Katherine Forrest: Right, right, right. By the way, I want to correct something I just said, which is I said that the OpenAI industrial policy document was in April of 2026. It wasn't. It was June 9th, 2026, and it's called Industrial Policy for the Intelligence Age. And we talked about it during an earlier episode. So you can get that on the openai.com website, June 2026. Moving on, the second idea in the Gates essay is one that I found really pretty interesting. And he sort of coins a phrase calling things human reserve, and that we ought to have a human reserve category of work. And that would be work that is particularly reserved for humans, even if a machine technically could perform that work. And so he gives a personal example of when his father was going through Alzheimer's and there was a caretaker and how he did not believe that his father's care, even if a robot could have been capable of providing the same kind of care, that that robot would have provided the kind of care and the kind of caring that the human caretaker that the Gates family had did. And then in his interviews, in Gates's interviews about this essay, he also sketched out this idea further, floating the idea that child care, jury duties, and other things might also be human reserved. So, you know, he's trying to find ways of having humans right now say, yes, there may well be things that AI is going to be able to do, but we don't have to have AI do them, particularly human-to-human important contact jobs.
Scott Caravello: Yeah, but I also really like how he talks about how it wouldn't necessarily need to be a one-size-fits-all solution, right, where that we all collectively agree as a world which jobs to reserve, and that it could vary across country. So take Japan, for example, right. They have a shrinking workforce and not enough young people to care for the old. So maybe they'd welcome a robot in a similar scenario to what Gates was talking about when it came to caring for his aging father. But that, of course, then leads to a bunch of practical questions about what happens when one country lets robots do something and another country doesn't. And so what does that do for international trade? But these are questions that he asks and not necessarily things that he's trying to solve with this essay. Moving on to the last big piece of the potential solution, and it's the one that he calls the top priority. It's building a new, whole new system to manage this transition both at home and internationally. His point is that none of our current institutions were built for a technology like AI that moves this fast and affects this many areas at once. And so he's making a case for new national and international bodies that can approach this transition. And so, for example, on the international side, he invokes regimes like nuclear inspections, international aviation, and ozone protection treaties. And while he says that this international institution for AI would still be unlike any institution we've ever created before, it could borrow elements from these three examples and more. And then at the national level, he suggests coordinating bodies that can set priorities across all the different federal agencies. And he also says, I think importantly, that some cooperation between the U.S. and China is going to be necessary to pull the whole thing off.
Katherine Forrest: Boy, that's going to be tough. The geopolitical overlay onto all of that really makes this difficult. And then we're in a, but also at a point in time when we don't have a long runway to get all of this accomplished. And so tying all this together, the crux of the Gates concern isn't just that the risks of the AI transformation are big. It's that, in his view, and there are others who share this view, and there are others who are working on solutions, but in his view, there isn't enough of a plan, or as he would say, there's no plan at all. And so some would say there is a plan, there's a partial plan, there's the voluntary, some of the voluntary commitments that are coming out. Some would talk about some of the other sort of efforts and initiatives that are underway. But he says, you know, there's no plan at all. And it keeps coming back to the fact that he doesn't see, from his perspective, leaders or experts or communities actually confronting the entire array of what might be coming down the road at us. And then he sees the gap that he's trying to get people to close with this huge disruptive force as really quite a big gap.
Scott Caravello: And we should add some balance here though, because he does still talk about some of the things that he's genuinely optimistic about when it comes to AI, speeding up drug and vaccine discovery, helping farmers in low-income countries, personalizing education, and so on and so forth. But his view is that we do eventually come out better off with AI. He just thinks that there is a turbulent stretch to get through it, and right now we're not preparing for it.
Katherine Forrest: As you and I have been saying, it's going to be interesting to see what kind of regulation ends up taking hold, because there are some efforts around, and the companies themselves are working on both voluntary commitments and self-regulation and all kinds of things. So this is very much a wait and see. But I take his point. It's a hugely important point that we're sort of watching the clock tick here. OK, that's it for today, everybody. If you've got a house cow, send in your photos to me and Scott. We'll talk about them and we'll do a little contest. Whoever's got the hugest house cow, send you a Waking Up with AI fleece if you're the one, although not with real wool. We'll make sure that it's some form of non-real wool kind of fabric. But that's it. I'm Katherine Forrest.
Scott Caravello: And I'm Scott Caravello. Don't forget to like and subscribe.
