[visarga]

> a huge underclass with nothing, nothing to do, and no access to that automation

Why do you think the poor won't have AGI in their pockets and robots as well? If we look at how LLMs are evolving, they become easier to run on edge, and there are so many of them being pushed out every day. Costs for robotic hardware are also going down fast. Intelligence is free to have, unlike UBI. And robots can build robots, or make it cheaper to build robots.

For example instead of AI, take web search - if you are rich or poor, you get the same search space and tools. The rich are not searching better. They don't have radically better operating systems, social networks or phones. Same thing with content - we all have the same massive pool of content to watch. The rich don't have their own private movies that are 10x better.

[strbean]

Because AGI is probably going to be incredibly expensive to develop, require mega-scale access to data and compute to train, etc.

And unless the models get leaked or stolen, whoever gets there first likely won't give out the secret sauce.

Then you have a single actor with access to, basically, limitless productivity. Would they share it? Given any use case for it, the owners of the AGI technology could trivially outcompete anyone else. Why would they let anyone else use it at all?

[insane_dreamer]

If you don't have a job that allows you to pay the rent and food, then access to AGI tools -- even free -- won't help you very much.

[fragmede]

I don't know why you'd say that. Makes applying for EBT and food stamps easier. And finding the local soup kitchens, the best places to buy a tent, what the laws against vagrancy and camping "overnight" in your car. Which brands of cat food are least unhealthy for human consumption. (you'll have to jailbreak the model to have it answer that last one.)

Those are probably going to be quite a lot of help!

[insane_dreamer]

"Sorry we created this AI that took away your job and made you homeless, but in exchange you can use our cool app for free that will help you escape starvation and maybe even find the optimal homeless shelter! And think of it -- your sacrifice has made it possible for devs earning 6-figure salaries to be more productive, and doctors to write better emails, and the USA now has a higher GDP than ever! Large corporations profits are up! Society (for the rest of us) is better because of it."

Reminds me of Walmart putting small business owners in a small town out of business and then offering them a minimum wage job at their store.

I recommend watching Nomadland if you haven't already. (Update: Unless you were being sarcastic, in which case, carry on.)

[fragmede]

Poe's law strikes again! The cat food bit wasn't over the top enough?

[Nevermark]

If someone doesn’t have a job that pays, how do they get those things?

I.e. destitute and homeless people do manage to creatively scrounge. But now imagine millions or billions of destitute people with no hope of a paying job. What are they going to scrounge in that context?

Jobs are a critical ingredient to people buying those common things with the rich.

[visarga]

This jump from "AI assistance" to "millions or billions of destitute people" is unjustified. Yes, AI can assist, but it doesn't replace human presence as it is now, in any field. Instead it creates demand for improvements across the board and creates more work for us.

When we went from horses to cars, we increased the volume of shipments and percentually more people work in transportation now than in 1910.

Another example - programming has been automating itself for 60 years. Each new language, framework or library, each project on Github makes future work more efficient. Yet we have seen an increase, not a decrease, in development jobs, and good salaries.

I would say humans are the critical ingredient AI needs to be effective, at least for now. And in the far future where AI can work without assistance, then it just empowers everyone to not need to work. We can use AI directly for our own benefit, automating self reliance for people. Unlike UBI, you can copy an AI and give it to everyone for free.

In both scenarios: weak AI making room for jobs, or strong AI making work not necessary, it turns out ok. But have more faith in our insatiable greed, we won't run out of work before we run out of desires.

[Nevermark]

> When we went from horses to cars, …

In this situation, we are the horses

[strbean]

This is all speculation based on the development of AGI. If we achieve AGI, yes, it would enable humanity to have a Star Trek-style fully automated space communism utopia. However, that hinges entirely on access to AGI and the fruits of it's productivity being distributed across the population, instead of hoarded by a small group.

If AGI is developed and kept closely guarded, whoever has it will have essentially limitless productivity, and quickly concentrate all wealth and power. They wouldn't even need to engage with markets, they could simply build overwhelming autonomous military power.

[visarga]

Here you are making a mistake - AGI by its nature should know all there is to know, and yet need to make progress by searching for new approaches and discoveries. That doesn't happen all at once, it works field by field, and discoveries actually come from experimental validation. There is no "secretly developing AGI" to "quickly concentrate all wealth and power".

Like bitcoin, you are basically saying someone could outcompute humanity and own the ledger. But in reality the combined research power of humanity, which is necessary for AGI to advance further, is much deeper than any one entity could achieve in isolation. Research is a social process.

[Nevermark]

> the combined research power of humanity, which is necessary for AGI to advance further, is much deeper than any one entity

AGI won't need people to advance further. That is pretty much the functional definition.

But the bar is even lower. Just as today's rich keep increasing the economic distance between themselves and the poor, with the middle and creative classes already feeling that gravity, so will the AGI's - even if they stalled out only as smart as we are, but cheaper in inference mode. (A limitation that is highly unlikely.)

No doubt humans will facilitate AGI activity in our economy and real world for practical reasons for a little while. But at some point, they won't need us physically or socially either.

[visarga]

> AGI won't need people to advance further. That is pretty much the functional definition.

No, AI can ideate as well as any human, but that is not sufficient. Scientists ideate and test. Engineers design and test. There's always a validation stage, where ideas meet the bottleneck of reality.

Are you saying scientists without labs and tools to run experiments on could do science? It's all in the brain or GPU? That is so naive.