Great post, sits at the intersection of 3-4 of my personal hobby horses. As a tech-savvy early 20s grad student with >1%ile ambitions but <10%ile conscientiousness, I've struggled quite a bit to find ways to force myself to stop wire heading on screens and get work done. The only thing I've found any success with is 100% impossible to bypass self binding. My phone addiction isn't a problem so long as I set the screen time settings to take away my browser and ability to download new apps, then have my girlfriend set the password. Unfortunately, I've found a way to break and bypass every blocker available for my laptop and desktop, so my distraction continues unabated on the bigger screen.
Short of getting rid of the screens at home and opting to do work only in libraries, it seems like an LLM assistant with the ability to carry out Oddyssean controls on my screen time according to its evaluation of pre-set criteria for acceptable levels and definitions of productivity would save me. I hope someone is working on such a defense-first product, or else maybe I'll build it if I...ever get around to it.
Maybe one equilibrium resulting from the world you describe is an elite 1-3% with the genetically endowed self control to resist the super-stimuli, another fairly elite 5-10% who lack the control on their own but possess the ambition/intelligence/wherewithal to outsource their self control to the machines, and everyone else living out Wall-E world.
Try a network level solution? It's kind of a fun thought experiment: how do you devise a system yourself that you can't escape using relatively open operating systems?
Here's just a spitball:
1) Router/switch is in a locked closed; you don't have the key/combination or the router admin password.
2) Router blocks all outbound DNS; no port 53 or 853 outbound connections allowed.
3) Block all possible DOH providers (this should be easy, since you only need to block the ones you already know about and just block new ones if you ever come across them).
4) Set up something like a pihole with a very limited whitelist.
5) Use kidde controls on the phone to block tethering.
6) Have a bash script running on the pihole that turns it off (or switches to a more permissive block list) at certain times.
I think that'd be pretty hard to work around---or at least really inconvenient.
Here's a positive angle: the prevalence of cheap LLMs will actually balance the playing field! Here's what I mean: right now, only megacorps can afford all of these thousands of PhDs. The only way to win that game is simply not to play. Nobody is really "smart enough" to avoid being manipulated by Facebook and Nestle's junkfood (actually I think being traditionally smart makes you more vulnerable); educated people just avoid it entirely, but of course that takes some degree of self-control and knowledge (and in some cases, nontrivial amounts of money).
In a few years, though, we might all have ten thousand PhDs in our pocket. Sure, the megacorps will have them too, and theirs will be better, but it'll probably like a 5:1 gradient vs 1,000:1 like it is now. That seems pretty great to me! You could just ask your AI agent "buy me ingredients from Instacart that allow me to cook seven meals this week; prioritize time over health and health over flavor" and completely beat all of the manipulation foisted on you by the designs of grocery stores and the design of the Instacart app. That's just orders of magnitude better than the process that most people follow today, which is going to the grocery store planning to buy tomatoes and ending up with some ultraprocessed Nestle crap or getting sucked into buying Twinkies based on some ad in their food ordering app.
Search is a great example. Until quite recently, Google was really the only game in town; "friendly" alternatives like DuckDuckGo just weren't very good. You _could_ opt out of a lot of the manipulative chicanery that Google foisted on you, but you had to know enough to install various browser addons, network-level blockers etc. But now, you can just pay $10/month for Kagi, which gives you better results and no attempts at behavioral manipulation. For just $120/year! Sure, _some_ people can't afford that, but not many. And for a mere $25/month, you get to pipe its results through the best frontier LLMs to get even more utility! And it's only Year 3 of the LLM Era!
You'll be able to tell an agent "review my Facebook messages and summarize me the ones that I've historically cared about, but ignore all ads, ignore all posts designed to elicit outrage, ignore all posts about politics."
One last point: all we really need to do is get productivity to the point at which food, energy, and shelter are essentially free. Once we do that, we've got the Star Trek economy, where 90% of the population can basically not work and have a perfectly decent life, but there are still things (leadership positions, Veblin goods, status goods, etc.) left for the remaining 10% to strive for, and that's enough to keep productivity marching forward.
Now, of course, there's the "violent young men" problem, which becomes a catastrophe for societies (lots of other great posts here about that). But I think we might be able to solve that problem with videogames. Seriously. Tyler Cowen wants to build cheap shanty housing in New Mexico with broadband for these people. Peter Suderman wrote a really compelling article about it all the way back in 2017 (https://reason.com/2017/06/13/young-men-are-playing-video-ga/). It's a bit of a nonchemical version of Wirehead City. I think there's really something to it!
Now this could all go horribly wrong, of course, in all the (totally plausible and perhaps even likely) described above. But there really is room for optimism!
I agree with pretty much all your takes - there's going to be opportunities galore for people to do better if they want to, and if they put effort in. I just think it's going to follow caste-splitting dynamics.
Because isn't "paying for apps" something like 80/20 in iPhone, with only 20% paying for apps, and isn't it ~95/5 in Android?
But yeah, I fully agree with your video game point too, I actually don't worry about any "violent young man" problem in the aggregate. Between porn and video games, we've got them pacified - and those are both part of my "Infinite Jest style VR heavens" for sure.
I think the future probably evolves into something like "Manna compounds" (pace Marshall Brain's short story here: https://marshallbrain.com/manna1) for a big chunk of the populace, where they camp out in dormitory compounds built on cheap land and get free food, lodging, and energy, and the video games / Infinite VR heavens.
And for the remaining ~20% (or whatever portion remains), I think the competition and race dynamics will actually get worse.
So to your point, there are gigantic multipliers on capability for relatively cheap in the near future. BUT I can envision a future where that also means is that there's a lot more headroom, AND each incremental hour is more valuable, which also means each incremental hour has a higher opportunity cost. If we're assuming people's own minds and time are still relevant (ie not fully dominated by AGI), they'll become an ever-more-valuable scarce resource.
So on the headroom point, much like during the Data Science revolution, there's going to be a lot of alpha laying around that can be picked up with extra effort and smarts. So sure, you can get an "as good as Ivy education today" AI tutor for your kids for $100-200 a month. But the people at the $2k a month tier are going to get hyper-Ivy quality that's like 50% - 200% better than today, and they'll be incomprehensibly farther ahead and are the only ones that can now get into the Ivies and Goldmans and FAANGs of that future. Educational Red Queen's Race has intensified.
Similarly, right now people have a ton of free time - the *median* person spends 7-9 hours a day on screens (in Zennials it's all phones, but in older gens, it's phones + TV's and computers). BUT in a future where AI has raised the headroom everywhere, each hour you spend on recreation has a much higher opportunity cost - now somebody who decides to work an extra hour drives 2-10x the productivity, revenue, or other value as before. Anyone who cares about promotion or growing their company is correspondingly incentivized to put a lot of formerly recreational time into working, because the returns are so visible and so high. Career Red Queen's Race intensified.
>Anyone who cares about promotion or growing their company is correspondingly incentivized to put a lot of formerly recreational time into working, because the returns are so visible and so high.
This is such a great point, it's almost worth its own post. I'd never really considered it before. Part of the reason videogames are so addictive is that they provide reinforcement at _just_ the right time. Right as you're exhausted and about to give up, you defeat some boss and lightning shoots out of your crotch. But in life (especially with high-effort/high-payoff jobs), you have to work in many cases years before getting any real positive feedback. To the extent that productivity gains also make the returns more visible . . . that would be incredible! I do think there are plenty of people who probably could be more motivated but who just don't have the patience to work on something for five years with basically zero feedback until they break through.
Another interesting thought experiment is whether diminishing returns will hold for LLMs. With most of technology, the performance band heavily compresses over time. The cheapest Android phone isn't really that much worse than the top-end iPhone or the Pixel Fold measured by productivity. You can still visit the same websites, use the same apps, etc. It took a bit for that to happen; back in the 90s, lots of people had older (like "purchased-just-two-years-ago-older") PCs that literally couldn't run significant software packages.
If price and performance scale linearly (or worse), then yeah no question these things will provide massive benefits to people with assets to begin with. But if they asymptote out like tech historically has, then I think it ends up being very egalitarian (at least measured on money/class; it'll make the returns on intellectual curiosity and conscientiousness even more exponential).
Of course, I was one of the naïfs who thought that stuff like Khan Academy would be a massive boon to high IQ/low SES kids, and that really never quite panned out.
It resonates particularly because I am one of these PhDs and have worked for one of the big online advertising companies in the past. Of course, there were many good reasons for that: I find their products provide value; there were many learning opportunities; I worked on internal tools and was far removed from anything related to monetization... but all things considered, I certainly contributed to the dynamic described in this post.
I made a switch in 2020 and have worked for effective nonprofits in the global health space in the last five years. This has made my life a lot more meaningful. The salary cuts have been more than outweighed by increased happiness.
If anyone reads this and becomes curious, I'd be happy to answer questions or provide mentoring.
> This has made my life a lot more meaningful. The salary cuts have been more than outweighed by increased happiness.
Nice! Congrats on making that transition successfully, I think it really makes a difference in terms of quality of life. The time we spend working is such a large part of our days, to be spending it towards higher ends that you feel are intrinsically worthwhile really means a lot.
I admire your offer of mentorship, too - kudos. A great way to pay it forward.
This Thursday I'll be reviewing the first clinical trial on a gen-AI therapy chatbot, which was published last week (10.1056/AIoa2400802) and yielded happy findings.
The authors acknowledge financial interests in the success of their bot, but even if they end up running one of those trillion dollar enterprises, mental health treatment is arguably shielded from the usual monetized AI influences, because the clinicians and clinical practices that inform the bots are grounded in countervailing principles.
For example, every therapist knows that feeling better momentarily is not necessarily a sign that you're healing in any meaningful sense. A moment of relief desirable, but it's not reliable evidence that the therapy is working.
So - in theory - gen-AI therapists won't be designed to prioritize little dopamine hits for users. Maybe the non-gen AI crap that's cheap or free will end up being ultraprocessed food for the soul, but the better products might be genuinely helpful. (In which case, some people will end up seeking relief from gen-AI for the miseries it creates.)
Depends on the monetary model. If therapy bots are one-payment fix the problem type things the incentive is to fix the problems as quickly and efficiently as possible. Maybe mental health insurance could help solve this? If everyone is always paying the same theres no reason to maximize therapy time.
If the model is subscription or frequent returns then the incentive is to fix the problem just enough to keep the consumer coming back. This is the conflict of interest that therapists are always needing to fight against.
This makes the authoritarian regimes of the 20th century look unrefined and primitive in their systems of control. People are being convinced to put on their shackles
Great post, sits at the intersection of 3-4 of my personal hobby horses. As a tech-savvy early 20s grad student with >1%ile ambitions but <10%ile conscientiousness, I've struggled quite a bit to find ways to force myself to stop wire heading on screens and get work done. The only thing I've found any success with is 100% impossible to bypass self binding. My phone addiction isn't a problem so long as I set the screen time settings to take away my browser and ability to download new apps, then have my girlfriend set the password. Unfortunately, I've found a way to break and bypass every blocker available for my laptop and desktop, so my distraction continues unabated on the bigger screen.
Short of getting rid of the screens at home and opting to do work only in libraries, it seems like an LLM assistant with the ability to carry out Oddyssean controls on my screen time according to its evaluation of pre-set criteria for acceptable levels and definitions of productivity would save me. I hope someone is working on such a defense-first product, or else maybe I'll build it if I...ever get around to it.
Maybe one equilibrium resulting from the world you describe is an elite 1-3% with the genetically endowed self control to resist the super-stimuli, another fairly elite 5-10% who lack the control on their own but possess the ambition/intelligence/wherewithal to outsource their self control to the machines, and everyone else living out Wall-E world.
Try a network level solution? It's kind of a fun thought experiment: how do you devise a system yourself that you can't escape using relatively open operating systems?
Here's just a spitball:
1) Router/switch is in a locked closed; you don't have the key/combination or the router admin password.
2) Router blocks all outbound DNS; no port 53 or 853 outbound connections allowed.
3) Block all possible DOH providers (this should be easy, since you only need to block the ones you already know about and just block new ones if you ever come across them).
4) Set up something like a pihole with a very limited whitelist.
5) Use kidde controls on the phone to block tethering.
6) Have a bash script running on the pihole that turns it off (or switches to a more permissive block list) at certain times.
I think that'd be pretty hard to work around---or at least really inconvenient.
Here's a positive angle: the prevalence of cheap LLMs will actually balance the playing field! Here's what I mean: right now, only megacorps can afford all of these thousands of PhDs. The only way to win that game is simply not to play. Nobody is really "smart enough" to avoid being manipulated by Facebook and Nestle's junkfood (actually I think being traditionally smart makes you more vulnerable); educated people just avoid it entirely, but of course that takes some degree of self-control and knowledge (and in some cases, nontrivial amounts of money).
In a few years, though, we might all have ten thousand PhDs in our pocket. Sure, the megacorps will have them too, and theirs will be better, but it'll probably like a 5:1 gradient vs 1,000:1 like it is now. That seems pretty great to me! You could just ask your AI agent "buy me ingredients from Instacart that allow me to cook seven meals this week; prioritize time over health and health over flavor" and completely beat all of the manipulation foisted on you by the designs of grocery stores and the design of the Instacart app. That's just orders of magnitude better than the process that most people follow today, which is going to the grocery store planning to buy tomatoes and ending up with some ultraprocessed Nestle crap or getting sucked into buying Twinkies based on some ad in their food ordering app.
Search is a great example. Until quite recently, Google was really the only game in town; "friendly" alternatives like DuckDuckGo just weren't very good. You _could_ opt out of a lot of the manipulative chicanery that Google foisted on you, but you had to know enough to install various browser addons, network-level blockers etc. But now, you can just pay $10/month for Kagi, which gives you better results and no attempts at behavioral manipulation. For just $120/year! Sure, _some_ people can't afford that, but not many. And for a mere $25/month, you get to pipe its results through the best frontier LLMs to get even more utility! And it's only Year 3 of the LLM Era!
You'll be able to tell an agent "review my Facebook messages and summarize me the ones that I've historically cared about, but ignore all ads, ignore all posts designed to elicit outrage, ignore all posts about politics."
One last point: all we really need to do is get productivity to the point at which food, energy, and shelter are essentially free. Once we do that, we've got the Star Trek economy, where 90% of the population can basically not work and have a perfectly decent life, but there are still things (leadership positions, Veblin goods, status goods, etc.) left for the remaining 10% to strive for, and that's enough to keep productivity marching forward.
Now, of course, there's the "violent young men" problem, which becomes a catastrophe for societies (lots of other great posts here about that). But I think we might be able to solve that problem with videogames. Seriously. Tyler Cowen wants to build cheap shanty housing in New Mexico with broadband for these people. Peter Suderman wrote a really compelling article about it all the way back in 2017 (https://reason.com/2017/06/13/young-men-are-playing-video-ga/). It's a bit of a nonchemical version of Wirehead City. I think there's really something to it!
Now this could all go horribly wrong, of course, in all the (totally plausible and perhaps even likely) described above. But there really is room for optimism!
I agree with pretty much all your takes - there's going to be opportunities galore for people to do better if they want to, and if they put effort in. I just think it's going to follow caste-splitting dynamics.
Because isn't "paying for apps" something like 80/20 in iPhone, with only 20% paying for apps, and isn't it ~95/5 in Android?
But yeah, I fully agree with your video game point too, I actually don't worry about any "violent young man" problem in the aggregate. Between porn and video games, we've got them pacified - and those are both part of my "Infinite Jest style VR heavens" for sure.
I think the future probably evolves into something like "Manna compounds" (pace Marshall Brain's short story here: https://marshallbrain.com/manna1) for a big chunk of the populace, where they camp out in dormitory compounds built on cheap land and get free food, lodging, and energy, and the video games / Infinite VR heavens.
And for the remaining ~20% (or whatever portion remains), I think the competition and race dynamics will actually get worse.
So to your point, there are gigantic multipliers on capability for relatively cheap in the near future. BUT I can envision a future where that also means is that there's a lot more headroom, AND each incremental hour is more valuable, which also means each incremental hour has a higher opportunity cost. If we're assuming people's own minds and time are still relevant (ie not fully dominated by AGI), they'll become an ever-more-valuable scarce resource.
So on the headroom point, much like during the Data Science revolution, there's going to be a lot of alpha laying around that can be picked up with extra effort and smarts. So sure, you can get an "as good as Ivy education today" AI tutor for your kids for $100-200 a month. But the people at the $2k a month tier are going to get hyper-Ivy quality that's like 50% - 200% better than today, and they'll be incomprehensibly farther ahead and are the only ones that can now get into the Ivies and Goldmans and FAANGs of that future. Educational Red Queen's Race has intensified.
Similarly, right now people have a ton of free time - the *median* person spends 7-9 hours a day on screens (in Zennials it's all phones, but in older gens, it's phones + TV's and computers). BUT in a future where AI has raised the headroom everywhere, each hour you spend on recreation has a much higher opportunity cost - now somebody who decides to work an extra hour drives 2-10x the productivity, revenue, or other value as before. Anyone who cares about promotion or growing their company is correspondingly incentivized to put a lot of formerly recreational time into working, because the returns are so visible and so high. Career Red Queen's Race intensified.
>Anyone who cares about promotion or growing their company is correspondingly incentivized to put a lot of formerly recreational time into working, because the returns are so visible and so high.
This is such a great point, it's almost worth its own post. I'd never really considered it before. Part of the reason videogames are so addictive is that they provide reinforcement at _just_ the right time. Right as you're exhausted and about to give up, you defeat some boss and lightning shoots out of your crotch. But in life (especially with high-effort/high-payoff jobs), you have to work in many cases years before getting any real positive feedback. To the extent that productivity gains also make the returns more visible . . . that would be incredible! I do think there are plenty of people who probably could be more motivated but who just don't have the patience to work on something for five years with basically zero feedback until they break through.
Another interesting thought experiment is whether diminishing returns will hold for LLMs. With most of technology, the performance band heavily compresses over time. The cheapest Android phone isn't really that much worse than the top-end iPhone or the Pixel Fold measured by productivity. You can still visit the same websites, use the same apps, etc. It took a bit for that to happen; back in the 90s, lots of people had older (like "purchased-just-two-years-ago-older") PCs that literally couldn't run significant software packages.
If price and performance scale linearly (or worse), then yeah no question these things will provide massive benefits to people with assets to begin with. But if they asymptote out like tech historically has, then I think it ends up being very egalitarian (at least measured on money/class; it'll make the returns on intellectual curiosity and conscientiousness even more exponential).
Of course, I was one of the naïfs who thought that stuff like Khan Academy would be a massive boon to high IQ/low SES kids, and that really never quite panned out.
This post resonates a lot with me!
It resonates particularly because I am one of these PhDs and have worked for one of the big online advertising companies in the past. Of course, there were many good reasons for that: I find their products provide value; there were many learning opportunities; I worked on internal tools and was far removed from anything related to monetization... but all things considered, I certainly contributed to the dynamic described in this post.
I made a switch in 2020 and have worked for effective nonprofits in the global health space in the last five years. This has made my life a lot more meaningful. The salary cuts have been more than outweighed by increased happiness.
If anyone reads this and becomes curious, I'd be happy to answer questions or provide mentoring.
> This has made my life a lot more meaningful. The salary cuts have been more than outweighed by increased happiness.
Nice! Congrats on making that transition successfully, I think it really makes a difference in terms of quality of life. The time we spend working is such a large part of our days, to be spending it towards higher ends that you feel are intrinsically worthwhile really means a lot.
I admire your offer of mentorship, too - kudos. A great way to pay it forward.
An "O" for your SWOT analysis:
This Thursday I'll be reviewing the first clinical trial on a gen-AI therapy chatbot, which was published last week (10.1056/AIoa2400802) and yielded happy findings.
The authors acknowledge financial interests in the success of their bot, but even if they end up running one of those trillion dollar enterprises, mental health treatment is arguably shielded from the usual monetized AI influences, because the clinicians and clinical practices that inform the bots are grounded in countervailing principles.
For example, every therapist knows that feeling better momentarily is not necessarily a sign that you're healing in any meaningful sense. A moment of relief desirable, but it's not reliable evidence that the therapy is working.
So - in theory - gen-AI therapists won't be designed to prioritize little dopamine hits for users. Maybe the non-gen AI crap that's cheap or free will end up being ultraprocessed food for the soul, but the better products might be genuinely helpful. (In which case, some people will end up seeking relief from gen-AI for the miseries it creates.)
Depends on the monetary model. If therapy bots are one-payment fix the problem type things the incentive is to fix the problems as quickly and efficiently as possible. Maybe mental health insurance could help solve this? If everyone is always paying the same theres no reason to maximize therapy time.
If the model is subscription or frequent returns then the incentive is to fix the problem just enough to keep the consumer coming back. This is the conflict of interest that therapists are always needing to fight against.
This makes the authoritarian regimes of the 20th century look unrefined and primitive in their systems of control. People are being convinced to put on their shackles
At some point, won't the only people having kids be those who have detached themselves from the system you describe?
"Life finds a way"?