Loading summary
Transcript34 lines
- [00:00]
Tim Miller
Hey, y' all, it's Tim Miller and I'm pumped to be here with Karen Howe. She's got a new book out called Empire of AI, about Sam Altman and OpenAI. And there's just so much news on this front, and it gave me nightmares last night to read the halfway through that I've read. And so I'm sure this conversation is going to make me feel better. Call me. I don't know. Karen, how you doing?
- [00:19]
Karen Howe
Great. How are you doing, Tim?
- [00:21]
Tim Miller
Maybe not. I'm doing all right. You know, I started it, I don't know, kind of during the second half of the basketball game last night. So when would that have been? Like 10 o' clock. And by midnight I was like, oh, my God. I I my concerns about AI apocalypse increased about 30% the more I learned about Sam Altman. So that's not good. We never overlap. Despite being gay and in the Bay Area together, I was never invited to his dinner salon. So your insight is, is the closest I've gotten to it, but I want to start here. So for folks who are watching this who have, I think there's just me varying degrees of interest in Sam and in AI. Let's just start with, like, why he is such an important figure and why OpenAI is so important when it comes to this brave new world that we are currently stepping into with regards to artificial intelligence.
- [01:11]
Karen Howe
Oh, I think he's such an important figure because he is. OpenAI is a manifestation of him. And OpenAI led the entire AI industry into approaching a particular type of AI development that is now kind of eating the world. And Allman is also very much a product of Silicon Valley. And so I don't want to give him too much credit, but certainly the way that OpenAI introduced the world to AI through a ChatGPT moment really pegged all of AI development, most of AI development today, to a particular conception of AI that is really built on this idea of scale at all costs. And, you know, it's not a coincidence that Silicon Valley, which has been trying to scale at all costs for a while, would end up developing AI in the same exact way. But Altman is the conduit of that entire enterprise where he kind of took all the ideas that he kind of grew up with as he was rising as an entrepreneur and then investor in Silicon Valley, and channeled it into this specific organization that became the firing shot, the opening shot of the global AI race that we're in now.
- [02:28]
Tim Miller
Yeah, so I want to get into all this extra. But just you said Something that was interesting there that also piqued my interest when I was reading it, which was you can have this notion that the type of AI we've gotten to, which is what we're trying to maximize as much compute as possible and make these apps as all knowing as possible, et cetera, was the result of a series of choices and then it didn't have to go this way. And you alluded to that in that answer. Like, what do you mean by that? Like, what are some other ways that people who were early, because you've been covering this for a long time, who are early in the AI research world or some other kind of proposed routes that this could have gone?
- [03:08]
Karen Howe
Yeah. So the AI research field has been around for a really long time. It was found in the 1950s at Dartmouth University. And a side note to this that I think is important to understand just generally about why there are so many different types of AI technologies is the term artificial intelligence was coined as a marketing phrase. It was coined by this assistant professor at Dartmouth called John McCarthy and decades later he said, I invented the term artificial intelligence because I wanted more money for a summer study. So essentially he picked the name to rebrand research that he was actually already doing under a different name. And one of the reasons why AI has just become so confusing, but is also, as you mentioned, I argue that it is very much driven by human choice is because this conception that these scientists are trying to recreate human intelligence is inherently flawed because there's no scientific consensus about what human intelligence is. And so throughout all of the decades of AI development, there have been lots of different debates about what AI should look like, what it should do and who it should serve, ultimately actually based rooted in the fact that different scientists have different answers for what human intelligence is. And so when I started covering AI in 2018, I mean, there was just such interesting research happening across the board. Like there were people that were exploring how do we build AI systems without any data? Like, how do we build extremely powerful AI models that can run on device, like on a smartphone? How do we try and capture knowledge? Researchers often call it common sense knowledge about the world without actually just trying to extract it from human experience, but hire experts to construct expert databases that contain this knowledge. So there were so many different variations, and all of that kind of died on the vine when, when OpenAI started working on what ultimately became ChatGPT. In the AI world, the ChatGPT moment was the GPT3 moment when for the first time they unveiled this model that was trained on 10,000 chips. And up until then, the largest models were trained on like maybe a couple hundred chips, and those were already gargantuan models. And when they did that and said, no, we are going to go for colossal. We are going to pump extraordinary amounts of data and extraordinary amounts of computational resources into this, that's when all of the other companies were like, oh, we're going to play this game too.
- [05:53]
Tim Miller
The other options that you just described, the other routes, using experts, paying people, like, that's all, that's all hard steal. I mean, not that, not the way OpenAI is not doing that hard, but, you know, there's something to be said for stealing every, all the existing data out there. You know, rather than using expertise.
- [06:10]
Karen Howe
Absolutely. I mean, it. Absolutely. It's. They took, they took the path that was easily, most easily accessible to them. Silicon Valley companies had been, they've already been sitting on massive piles of data because they've been accumulating for a really, really long time. So this is already their competitive advantage in how they could develop AI. Like, why, like you said, why hire the experts? Why pay people when you've already got these troves that you can just tap into? And really, the only kind of bottleneck at the time when OpenAI started doing this approach was capital. They needed a boatload of money to do this. Gobs and gobs. And it just so happens that OpenAI has one of the greatest fundraisers of our time right now as the CEO. And so, yeah, so it very much was explicit, intentional choices that led them to take this path.
- [07:03]
Tim Miller
So this takes us to Sam, who is head of this. And for people, I guess the just really quick, short backstory is Sam and Elon actually were kind of involved in the origins of this when it was going to be a nonprofit. And we'll kind of get into that a little bit, but just Sam himself. I kind of assumed, being an outsider before I started reading the book, that he was like, really a really skilled engineer. And he comes off a little bit non social to me. So I find it weird that he's such a good fundraiser. So did I watch his interviews and I'm like, I don't know, does this person make eye contact? But I assumed that he was a really good engineer or that he had, he had founded some really successful tech thing that I just hadn't heard of because it's like a niche tech thing and it's like none of that. Like, he founded like a Foursquare competitor that failed. I liked Foursquare. I was a Foursquare user, something called Looptop, that didn't do any good. And then he got Ben into vc. So, like, why is he the point person for like, the future of humanity? I really. I still can't understand.
- [08:06]
Karen Howe
Yeah, so, yeah, so he did. He was at one point technical. He did study computer science, but then he dropped out of Stanford before he finished his computer science degree to found sleep. But it really comes back to, you know, the core skill set that he has that everyone says is what makes him unique is he's really, really, really good at telling stories about the future. And he also has a loose relationship with the truth. So when he's in a room with someone, the things that come out of his mouth to paint this vision of the future are more correlated with what he thinks that person needs to hear to then want a piece of that future than what he believes or the ground truth of the matter. That's what makes him a really good fundraiser, and that's also what makes him really, really good at rallying a lot of top talent towards a particular goal. And so Those were the two main assets that he brought to OpenAI was the capital and the top talent. He was able to recruit first Elon Musk to the venture, then he was able to recruit Ilya Sutskever, who became the chief scientist, who was at the time one of the most famous AI researchers in the world because of a contribution he made as a grad student that essentially set off the deep learning revolution and started turning AI into a commercially viable technology. And he recruited other people along the way. That I document in the book that he recruited people from Google to then help with developing ChatGPT. And he recruited Microsoft to be the investor and main backer and to develop the supercomputers that OpenAI wouldn't be able to develop. So I think his skill is channeling, consolidating resources and channeling it in a specific direction. But there's also a lot of frailty and weakness in the ventures that he creates because he's really good at telling personalized stories to people when he's sitting with them one on one. But once you have a company and you need everyone to be on the same page moving in the same direction, that's where Altman really falls short, because he will tell different people different things based on what he thinks will motivate them, and the shared picture that would be the foundation of the company starts to fall apart.
- [10:36]
Tim Miller
Right, so the prime example of what you're talking about here, this where Altman's Good at mirroring what other people, what he thinks other people want to hear is, is in his recruitment of Elon Musk initially. And so Elon, for all of my concerns about him and a lot of other ventures around AI, you know, if you go back, he kind of still does this now, but less. Less which I want to get into. But if you go back half decade, like, Elon was kind of apocalyptic about AI, like, very concerned about worst case scenarios. Yeah, he still is. Yeah. So concerned about worst case scenarios. We need this to be in good hands, you know, like, we need to be considerate about this. Is why OpenAI, when it started, was a nonprofit, you know, that we need this to be in the public interest and the public good. It's like an Oppenheimer type situation. And so Elon, as you tell it, like Altman, it kind of reflects that same opinion back to Elon as part of the recruitment process. But now, I guess if things have gone along, it's no longer a nonprofit. Altman, when I listen to interviews of him, sounds downright optimistic, almost. What's the word? Almost like there's a paradise waiting behind us with AI. So what happened there, do you think? Was Altman lying to Elon to get him on board? Has he changed his view? Has something happened that made him change his view?
- [11:56]
Karen Howe
Yeah, I think there's kind of two ways to answer this question. One is like, just how Altman operates in the world, which, through the reporting of my book, what I realized is he will bring in people and resources and create certain structures based on what he needs, what the objective that he needs to achieve at that time. And then once those people, those resources lose their value because the objective changes, he then sheds them and shifts. So the nonprofit, which originally was a mutual idea that Musk and Altman sort of had, was something that was particularly helpful for recruiting Musk to the venture and also for recruiting other talent. Because Altman is very strategic. He plays the long game. He likely understood at the time that he didn't have the capital to compete with Google, which was the main monopoly on AI talent. So he couldn't wave around millions and millions of dollars in salary to poach Google researchers. So the thing that he could compete on was a sense of mission and purpose. And by creating a nonprofit, that was a really great way to highlight why OpenAI was different from Google and had that mission and purpose. And that was what then hooked Musk in and then hooked Ilya Sudskever in and many other researchers who moved from Google to OpenAI. Once he had that talent and once Musk had already lended his brand name, it became less necessary for Musk to be there and it also became less necessary for the nonprofit to stay a nonprofit. The next objective at that point was how do you win? How do you build a lab that is going to be number one and beat Google at that point. Now you need capital. And in order to raise capital, you can't stay a nonprofit. You need to create some kind of for profit fundraising vehicle. And that's when he nests the for profit and the nonprofit. And Musk doesn't want to be part of it anymore. And it's fine because he's already taken the value utility from Musk being part of the project and doesn't need it anymore.
- [14:14]
Tim Miller
It's weird to find Musk as the principal person story. The other thing during this period that is like that has changed is there's an email from Altman to Musk that you have in the book that's obviously we comply with, aggressively support all regulation. This also speaks to their concerns about, right, like about the downside risks of AI. It should be, should be a nonprofit. It also should be regulated. Like now the whole like tech bro, Teal Andreessen, like kind of orbit that was pushing Trump through not regulating AI is like one of their core precepts.
- [14:48]
Karen Howe
Yeah, yeah.
- [14:50]
Tim Miller
What changed with that?
- [14:52]
Karen Howe
So that sort of gets to the second way that I could answer your question, which is, okay, the rhetoric that AI companies often engage in is it takes one of two forms. Either it is this technology is really dangerous, or it is this technology is going to bring us utopia. But ultimately they are two sides of the same coin because the conclusion from both versions of the rhetoric is AI is extremely powerful. And therefore we, the people who are saying this should be the one to control it. So, you know, like, when Musk was really deeply concerned about existential risk, I don't think it was rhetoric in the sense that he believed that this was a problem, but also he leaned into it because ultimately there is this deep seated desire to want control over the technology. And Altman tapped into that desire, both the fear and that desire, when he proposed to Musk, hey, I know you don't like the way that AI development is going right now. And it seems, seems to me that the best way to counteract that is by just building our own organization. You know, like he taps into the fear and then he taps into that desire of wouldn't you want to run an organization that then gives you more control over this technology? As the public discourse has sort of changed and shifted based on what is on the minds of people and policymakers, those different narratives get wheeled out in turn. You know, when people are very fearful, they wheel out the yes, and you should be fearful. And this is why we're being so cautious and careful. And this is why we don't want this technology to go into the hands of China.
- [16:39]
Tim Miller
Right.
- [16:39]
Karen Howe
When people are feeling more optimistic, they wheel out the and you cannot imagine the prosperity that we will soon see. We don't even have the words to articulate the profound positive transformation that's going to happen. And that's why you can't regulate us. We need to put pedal to the metal to accelerate it. It always goes back to the same goal, which is that they just need to continue moving forward with no obstacles in their way.
- [17:04]
Tim Miller
Like one tangible example about this is the climate side of it. You had a funny exchange with a couple of the guys about how the same paradox is working right? Where they're like, AI is going to fix the climate problem. Eventually the super intelligence and AGI will figure something out that'll fix it.
- [17:23]
Karen Howe
Yeah.
- [17:23]
Tim Miller
But in the meantime, we need to extract as much resources as possible that's going to exacerbate the climate change problem in order to get there. And there's this race between how much we exacerbate it versus when the brilliant supercomputers are going to fix it. What is your sense for. You talk to a lot of folks who are not spinning you the real level of concern there as far as the climate element to this and just I guess that whole kind of conversation.
- [17:55]
Karen Howe
There'S no concern at the top at all. You know, multiple people about their impact on the climate. Of their impact on the climate. Yeah, multiple people told me OpenAI sources told me this has never once been mentioned in an all hands company meeting. There are actually people who mentioned to me that on the policy side of OpenAI, this came up as a this will eventually like that shoot. The environmental shoe is going to drop. People are going to realize that these have environmental consequences and we're going to have to deal with it as a PR crisis. But it was never mentioned as a maybe we should actually think about how we are developing these things to accelerate climate change and accelerate other kinds of environmental issues, like people's access to fresh water resources. Never, never once happened to the point.
- [18:49]
Tim Miller
Where do you feel an ethical issue like with the climate side of it, like it's hard for me to wrap my head around like how real the climate like, you know, should I not be asking ChatGPT to make me funny pictures because I'm killing a tree.
- [19:03]
Karen Howe
It is becoming a huge, huge issue. Like there are projections that say that at the current pace of data center development, to support these AI ambitions in five years, at the end of the decade, we will need to slap the equivalent of 2 to 6 California's of energy demand onto the global grid. And all of that energy demand will, the majority of it will be serviced by fossil fuels. Because these data centers have to run 24 7, they cannot run on renewable energy. The sun and the wind don't blow and shine all the time. And Altman even said in his Senate testimony most recently that in the short term this will most likely be serviced by natural gas. And then of course he says, but, you know, in the long term we'll figure out fusion. But the, the problem here is the AI industry always does the same dance of there are real things happening now with concrete evidence that you can point to of harms and then they wave a wand of. But in the future they will be fixed by something that we don't actually have evidence will ever happen. So they're trying to use a figment of imagination to make okay, a current reality. And so yeah, it's absolutely, I mean, it's a huge problem. And it's not just climate. We are also facing a huge freshwater crisis globally. Now. There are many communities around the world that now have trouble accessing drinking water. Like when I was reporting this book, I traveled through Latin America. In the book I read about Chile and Uruguay and I also went to Colombia. All of those countries are currently facing a mega drought, a historic mega drought. So is the US the southwestern US where which has become a huge hub of these data centers, is facing the worst drought in a thousand years, according to MO research from Nature. You know, like in Colombia, I was literally visiting and tried to go to the National Museum in Bogota. And the National Museum was literally shut down because they could not service water to the bathrooms. And in Uruguay, like they, the government is mixing toxic chemical water into the drinking water supply because they just need something to feed the tap so that when people open their taps, something comes out. And there was this. Yeah, no. And the last thing I'll say about this, there's this amazing Bloomberg story that just came out that looked at all of the water consumption that's happening for these data centers all around the world. And what they concluded is it's not just the total volume of freshwater that we need to be concerned about it's the distribution of these data centers. I think they said like 2/3 of these data centers are now in water scarce areas. So it is a huge environmental and climate problem.
- [22:14]
Tim Miller
We could do so much and there's a lot there in the book. People should go check it out. Empire of AI Just two really quick things that caught my attention. The sister, he has a Sam Altman who's like going to be the richest man in the world, probably maybe the first trillionaire, has a sister who is in insecure housing apparently.
- [22:29]
Karen Howe
And yeah, so Altman's the oldest of four siblings and his two brothers, his three brothers and a sister. The two brothers followed him in his career. So when he struck out in Silicon Valley and started doing really well for themselves, they joined in, he became an investor, they became investors, and they've all become very, very rich from this career path. The sister was the black sheep of the family. She was the artsy, fartsy sibling. She really didn't want to go into tech, she wanted to be a writer, she wanted to be a comedian, she wanted to do other types of art. And she had always intended as, when I interviewed her, what she told me was, you know, she had always intended to support herself. She had never expected, you know, her rich brothers to support her in any way. The problem is she then started having a series of really intense health problems, physical health, and she struggled with mental health problems her entire life as well. And as she started having more and more exacerbated health problems which worsened after the death of their father, who she was the closest to in the family, it made it extremely difficult for her to continue working. And so she alleges that when she appealed to the family for support, and she provided me a lot of documentation showing email exchanges and text messages that they withheld money from her that should have been hers. She learned that her dad had left her pool of money and she tried to access it to support unemployment and support the ability to heal her physical and mental health. And her mother and brothers stepped in and they, they frame it completely differently. They dispute the way that she characterized it. They said that they, she already had some money and they were really worried about exacerbating her mental health challenges by giving her access to more money. But essentially the effect was that she ended up in a place where she had no money and she turned to sex work to start paying her bills. And she then ended up in a period of years where she just faced a lot of housing insecurity, food insecurity, economic insecurity. And ultimately, one of the reasons why I highlight her experience in the book is I think Annie is very much more representative of the way that the majority of the world lives than Sam is. And her life is an interesting case study of the impact that AI has on people who live like Annie, which is most people.
- [25:18]
Tim Miller
Yeah.
- [25:19]
Karen Howe
In that when these companies talk about AI, they talk about it solving poverty. Like Altman has said, AGI should solve poverty. And like, AGI will cure cancer and bring accessible, affordable healthcare to everyone. And AGI is going to make all your, you know, all your problem, all your economic problems go away. And the problem is she was facing all these intersecting challenges, health challenges, economic challenges, mental health challenges, and she wasn't getting any. Any benefit out of AI. And in fact, actually, she was having trouble accessing economic opportunity online. She was trying to monetize a podcast, she was trying to monetize you a YouTube channel. And she consistently felt like she was being shadow banned. And when I talked with researchers and experts about this, this particular issue, they mentioned, you know, because she was involved in sex work, that that's how the Internet works. They use AI systems to track sex workers, even in platforms that are completely unrelated to their sex work, to limit their distribution.
- [26:30]
Tim Miller
In some ways, you're making this a policy thing, but, I mean, there's a human element to this. I mean, he's unbelievably rich. I understand that these things are complicated, but, like, I think it reflects a little bit about how much he's going to think about the impacts what he's doing on people that are going through issues, how he thinks about this when it's his sister going through the issues. I'm sorry, I'm running out of time, but I just really quick. So he partnered up with the guy that designed the iPhone, Joanie, I've. And they have new products that they're, like, teasing. Looks like it's going to be a smart brooch. So we'll call it a super smart brooch. Can you give us just, like, one minute on what you think they're planning with the broach?
- [27:09]
Karen Howe
Altman recently said in an interview, like, his strategy is to try and do as many things as possible to create as many products as many surfaces through which people can interact with their technologies. So hardware is a very logical step under this strategy. They've thus far been using the hardware that people already have. You know, they're creating apps that install on your phone and apps that are on your computer, but they want to move into wearables, maybe Smart speakers, I don't know, they want to add more hardware to your life for you to essentially create more service area for them to collect data on you.
- [27:47]
Tim Miller
Yeah, they want to listen to you all day so that you can then ask the smart bro to be like, hey, I had a meeting earlier. Like, remind me what they said and it will have been recording everything.
- [27:56]
Karen Howe
Yeah, exactly. And you know, the way that Altman frames it is he has long really loved the movie Her. And people that I spoke to within the company said that the reason he loves it or the reason he says he loves it is because it's this seamless AI experience that just exists in your life. It's like constantly there, gather, you know, like intimately understanding, you know? Um, but the cynical take is that it's just more ways to collect more data on you, because ultimately that is one of the key ingredients to training their larger and larger models.
- [28:32]
Tim Miller
Boy, I gotta tell you, I'm like, some viewers won't believe this because there's so much bad stuff happening. I'm always negative, but, like, I'm an optimist by nature. Actually. I was like, a very excited tech person. Like, I was going to south by southwest in the mid-2000s and, like, so excited about all these new things that were coming. And I have, like, my. My will towards techno optimism has been stripped from me page by page in your book and also in some other things that are happening out there in the world. But anyway, people should read it. Anyway, it is. It is called Empire of AI. It's Karen Howe. Thank you so much for spending time. Good luck on the book tour and let's stay in touch as I'm sure this stuff is going to be in the news.
- [29:13]
Karen Howe
Thank you so much for having me, Tim.