AGI Already Happened... And Almost Everyone Missed It w/ Dr. Alexander Wissner-Gross

When you're talking about robotics, are you referring to how Tesla has decided to stop making most of their cars and wants to build a million Optimus robots next year?

That's I would say a symptom, not a cause. This is an industrywide phenomenon. Tesla is doing an excellent job of embodying that with this recent, I would say, courageous and one might say founder mode style pivot from Model S and Model X over to humanoid robots in their Fremont factory.

But yes, I would say that is emblematic of a broader shift toward humanoid robotics with ultracapable vision, language, action models that again just following the law of straight lines and capabilities consistently going up and to the right.

I think we'll find ourselves in a world in the near-term future where physical labor is also too cheap to meter.

So does that mean I guess maybe you can disambiguate that for us a little bit, right? Is that is that are you talking about the physical labor that we're doing right now? Are there any particular area sectors that you think are going to be affected sooner than later?

And I'm just talking, you know, we obviously cover a lot of the Mac 7 and everything and you've seen that these rumors Amazon wants to get rid of there are 300,000 workers, all that kind of stuff. Are you talking about basically anything physical? Like are are you talking about robots painting my house?

Yes. Is there any you're talking about every every type of physical labor? I mean the entire economy. I mean one one can cherrypick particularly vulnerable subsectors to physical automation or cognitive automation but I think in the fullness of time it's the entire economy as it's currently constructed.

What's the biggest barrier to that then? Is it is it the cost of production? Is it the actual chips? Like what is what is kind of holding back that that development?

Regulation I think. So, I spend substantially all of my time in the Boston area. And here in Boston, there's a big food fight going on about whether Whimo robo taxis can be brought to Boston.

The primary barrier there is arguably regulatory. It's no longer a technical capability argument, even though some would perhaps try to frame it that way.

I think the jobs I would almost say the question to ask is not which jobs or which labor categories or job functions will be automated first. I think maybe the more interesting question is which will be automated last and those right now if if present trends continue that will be automated last are those that either are protected by laws and regulations or those that demand such extremely fine tolerances and compliance that for whatever reason but this is largely in the end a social construct that it's very painful a march of the nines in terms of reliability and compliance will be required to fully automate that labor.

So one can imagine scenarios where ironically and Hans Moravec has spoken about this quite a bit in in terms of the Moravec paradox where the things the tasks that humans find easy robots and automation finds difficult and vice versa.

I think we maybe find ourselves in a world where large chunk of human cognitive labor and human physical labor is relatively easy to automate with a combination of models, frontier type models that we have right now on the cognitive side which are relatively difficult for humans.

And then the physical labor which is relatively easy for for humans relatively low bar unskilled labor ends up being harder but not that much harder. I think at most, call it conservatively, 3 to 5 years before most physical labor tasks that even a skilled human could perform will will just be like a special case of some vision language action model on top of a humanoid robot.

So, Alex, does that mean that we will then have UBI? Is that what's going to happen to like people who have labor jobs right now and and most of the population? Is that the way is that the solve for I guess continuing the economy as we know it?

I think it's a totally separate discussion. So, so I want to distinguish between technical capabilities that that is what the AIs and robots that we produce and that produce themselves will be capable of in the next few years and what the human economy looks like, what the social economy looks like and what we do about potentially a yawning capability gap between human capabilities and human economic faculties and the automation.

I think these are to they're not totally independent problems. Obviously, they're coupled, but I think they need to be discussed independently.

So to the question about UBI, my modal hypothesis is that as we saw at the beginning of the 20th century with the parade of isms, probably the world economy will try every social economy experiment that that we can conceive of. So I think you'll see and are already seeing UBI experiments in different places.

UBS universal basic services. So just to distinguish UBI income it's arguably sort of a demand side solution to what happens when we hit some form of post scarcity. UBS universal basic services more of a supply side solution.

So under UBS, take like Amazon Prime or or some sort of flat rate subscription where you get a bundle of services. Now imagine scaling that up by a factor of 10 or 20. So maybe individuals in the near-term future pay either out of pocket or via subsidy $200 per month and get a bundle of every necessity of living, health care and food and shelter and utilities and information and entertainment.

So that that's the UBS, universal basic services scenario. There's also UB, universal basic equity. That looks a little bit like sovereign funds like what we see in Alaska or Norway paying out dividends from some sort of sovereign fund that is able to invest perhaps in the broader market or in some assetbased class and distribute some fraction of the dividends to to people.

So I guess to to wrap up my answer, you asked specifically about UBI. I don't think UBI should be treated as the totality of a quote unquote solution to post scarcity. I think UBI plus UBS plus UB taken as a whole. I think even that is only a fraction of the solution.

I think the real solution is making sure that human capabilities and human economy continue to be well coupled to the machine economy. And so I spend a lot of my time thinking about how we augment human capabilities to make sure that the human economy and the AI economy maintain a strong enough coupling that to the extent that we need the U's and the B's UBI, UBS, UB that those are on the margin sort of bandages to to keep the entire coupling going and to keep the social economy from collapsing.

But I'm not yet convinced that those are the front and center solutions or should be the front and center solutions.

Link: Get the biggest AI moves and what they actually mean for investors twice a week straight to your inbox. The link is in the description.

Link: Crypto taxes are a nightmare. You've got trades across 15 exchanges, DeFi positions you forgot about, NFT flips, staking rewards, airdrops, and somehow you're supposed to report all of this to the IRS. Good luck. Cue the solution. Sum. You may know it by its old name, cryptot tax calculator. The Sum platform connects to over 3,500 exchanges, wallets, and crypto projects, including full support for DeFi, NFT staking, and airdrops. It finds deductions you'd miss, reconciles massive transaction histories without losing track, and generates IRS ready reports that will help you pay the least tax possible. Oh, SUM is also the official tax partner of Coinbase and MetaMask, rated 4.6 out of five on Trustpilot. Turn crypto tax chaos into confidence. Get started for free at milkroad.com/sum. That's sum.com. Milkroad listeners can also unlock 20% off their first year subscription with code milkroad 20.

I want to get your thoughts on AGI because that's also something that I feel is is talked about a lot across a lot of different circles. You see it if you go on X, it feels like AGI is being discovered every day in some new place.

I'd love to get your thoughts on on when that's coming, how it's going to affect us, and even how it plays into kind of like your last answer about about what that human to AI relationship is going to look like.

Yeah, I think AGI is coming at least 5 years in our past. I think we we hit AGI no later than summer of 2020. Now, AGI is a term that was in part popularized by Nick Bostonramm, part coined/popularized by Ben Girtzil. It it's become somewhat mushy as a term at at this point.

The way I construe it is the ability for AI to demonstrate generality in terms of its capabilities. And I've argued and I would continue to argue that we hit as a civilization AGI no later than summer of 2020 when open AI published their paper language models are few shot learners or I guess it was large language models or or few shot learners which coincided and was about coincided with and was about GPT3.

So I would say GPT3 summer of 2020 is when we hit AGI. The rest like the rest of history between 2020 and now has been relatively from my perspective incremental scaling, incremental features, relative relatively small but important additions, capabilities, the addition of reasoning obviously was an important step.

But these were all I think in my mind these pale in comparison to the big unlock which was discovering that we could achieve general intelligence by training models to predict next tokens over general human knowledge.

Like that's the big surprise. If if we could send a message back in time 20 or 30 or 50 years to this entire AI industry that that has been developing since the mid1 1950s at the very latest that has been wasting arguably a bit of a hot take wasting time on different approaches, different artisal algorithms.

So much time wasted. If we could just send back in time the message, look, take all of human knowledge, store it, and and these are concepts that would be familiar, say, to Vanavar Bush with his MEMX, uh, sort of a proto Wikipedia, if you will.

These would be very familiar concepts in the 1950s, probably in the early 19th century or early 20th century, rather. Store all of human knowledge in one place and then build a model that's really good at predicting the next word. That's all you have to do.

and and you know maybe parathetically it's it's well established in in computer science that the ability to compress information is dual to the ability to predict next tokens or next words. So doesn't matter how you formulate it but just do that do that really well and you get more or less AGI for free.

So many decades arguably wasted pursuing fruitless trajectories. We could have just done it. It was very simple.

So you're telling me that you think with with GBT3 that we had AGI and that basically the the the start of AGI is this chat GPT model that basically is able to predict the next word or kind of like feed back the information you've given to it and and respond to you actively right even predates chat.

So so I'm talking about GPT3 before chat GPT even existed. ChatGPT remember started out as just a wrapper around GPT. I'm talking about the GPT3 model which predated a conversational interface.

Got it. Okay. But you're telling me that basically you think you think that that was AGI and that from here we're just adding things to it. And I'm just I'm asking you that because I feel like that's significantly different than what most people think AGI is going to look like which is some kind of massive scientific discovery that it's like, hey, we've cracked it and now there's this intelligence beyond us.

But you're kind of giving us a a slightly different view that it's really just taking everything that we've learned and letting it kind of feed back to us or at least kind of add a little bit to it.

Link: Want to stay ahead of the biggest technological shift in history? Subscribe now to get insights straight from the sharpest minds in tech and finance. Quickly, you'll note this show is for educational purposes only. Nothing here is financial advice. Investing always carries risk. Never invest more than you can afford to lose. Thanks for tuning in. See you in the next one.