There is much talk in the tech scene of a ** permanent underclass**.
The idea is that a general superintelligence can automate all work, and hence drive the value of labour to zero. Because it is not possible to build wealth through labour anymore, whoever is wealthy, in particular in terms of ownership of 1) the AI value chain 2) non-scalable goods such as land, stays wealthy, and everyone else is screwed. Hence a frantic race is developing to make as much money as possible in the next few years while we still can. I do think that this scenario is possible, and that incentives to save money are substantially higher than if the AI revolution were not happening. But the freezing in of wealth scenario rests on several (load-bearing, as Claude would say) assumptions, each of which may turn out false:
ASI (artificial superintelligence) arrives and does in fact largely eliminate the value of human labour
Redistribution is limited. This probably requires a non-democratic outcome: Part of the argument for property rights and limited redistribution is that they incentivize value creation through labour, and the combination of labour and investment. If the value of labour, including that of entrepreneurs and investors, drops to zero, a population may turn towards socialism, or at least increased redistribution
The ASI or whoever controls it still really cares about property rights rather than just dispossessing shareholders, landowners, etc.
I do think such an outcome is possible, if someone from the “democracy is incompatible with freedom” camp around e.g. Peter Thiel controls AI, and they stick to their stated belief in property rights. To stick to property rights, the post-ASI elite needs to either be large and interdependent enough to require property rights to coordinate, or believe strongly enough in property rights as a value that they protect it even when it does not serve them anymore. ASI elites would probably be smaller than previous ones because of the limited need for human labour. I also don’t strongly believe in the latter, because parts of the tech richt support threats to annex Greenland, which seems like a violation of property rights to me. Overall I believe the wealth freeze-in scenario is possible, but with a likelihood well below 50%. When planning for the future, both as individuals and as a collective, we should consider it, but we should consider also what happens if we relax one or more of the assumptions underlying it.
AI as a a bubble
If we strongly relax the first assumption, we are in the AI as a bubble scenario. I do not believe this is the case beyond a relatively short-term dip (e.g. like the dot-com crash), given how much I and many others already use LLMs every day. But if this scenario were true, we’d probably be best served by not thinking about AI too much, and just getting on with our lives.
Ai as a normal technology
A softer relaxation of the first assumption is the AI as a normal technology scenario. In this scenario AI is similar to, say, the industrial revolution: Much work gets automated and there is a lot of disruption, but there are still plenty of opportunities for humans to add value in a way that can be monetized. This almost certainly is what happens in the short term, and some people argue this is also how things will stay long term. In this scenario I see a good (but not at all certain) chance of democratic capitalism being maintained, just as it ultimately was after previous technological revolutions (despite experiments with socialism): Both entrepreneurship and investment judgement (whose importance strengthens capitalism and property rights) and labour more generally (whose importance strengthens democracy, or at least an autocracy caring about its citizens) retain value. The best strategy to deal with the AI as a normal technology scenario is to shift our skills to the ones that are more likely to be complementary with AI. In its implications, it is somewhat similar to the wealth freeze-in scenario in that there is still an incentive to work hard to create wealth: Technological revolutions are great opportunity windows to do so. But there is no reason here for “last chopper out of Saigon” anxiety.
Egalitarianism
If we relax the assumption that redistribution is limited, we may end up with some sort of egalitarian outcome. This would come about because now that labor, including investment judgement and entrepreneurship, has little value, the population no longer accepts arguments that wealth disparities are acceptable because they incentivize value creation. Alternatively, a sort of benevolent dictator ASI or person controlling the AI could have egalitarian values. This scenario could actually be disconcerting for those who are well off at the moment: They may lose some of their non-scalable goods, like real estate. Wealth as a source of status could disappear.
I can see two broad sub-scenarios:
A kind of worldwide egalitarianism, e.g. implemented by an AI that has those values, or in a multipolar scenario. In that case we can be quite relaxed, because what we do now does not have that much impact on our economic outcomes. Once it arrives we
need to replace work as a source as a source of status and meaningA more national one, e.g. frontier lab profits might primarily be distributed to the US population. In that case, citizenship of the country that achieves ASI would be the most useful asset
There may also be an in-between scenario, where there is a substantial UBI (universal basic income) for most people, but some people with wealth or rare skills in remaining professions, e.g. entertainers, retain an increased income. This could be an interim state, as UBI gradually increases and becomes a larger and larger share of overall income.
Intelligence curse
Property rights could also be broken in a way that does not involve democracy:
A government autocrat may dispossess shareholders in the AI value chain
A CEO of a frontier lab could operate with fewer and fewer staff until he alone controls the AI
Once these people control a drone army, any of these interventions would be difficult to resist. *Operational *control of the AI is crucial, not theoretical ownership. We see this in the example of international law having limited consequence if there isn’t someone powerful to enforce it.
Such centralized power without much requirement for people could lead to an intelligence curse scenario where a small group of people rule as they see fit, seeing little need to take care of the broader population, such as in many resource-rich countries like DR Congo. How would one prepare for such a scenario?
One could try to become part of the elite, through direct operational involvement and strong connections. But this elite in the long run may be even smaller than the elite of resource-cursed countries: those states needed soldiers and engineers, but now AI can do those tasks. Making some money from a wrapper startup exit is certainly unlikely to be enough. If one is powerful enough to become part of the intelligence curse elite, one may also be powerful enough to influence outcomes towards another scenario, depending on one’s values
If one is not part of the elite, general apocalypse prepping may be helpful, including in particular having many strong trusted relationships to face the coming hardships together
Misalignment
If a misaligned ASI that cares little about humanity’s welfare takes over, we have a number of dark scenarios including: Some hellscape resembling factory farming
What those share is that the agency of even elite humans in these scenarios is very low; any preparation we do as individuals is unlikely to have much impact. All we can do today is:
Try to prevent such catastrophic scenarios Enjoy the time we have left
Putting it all together
What are the implications of this? To maximize one’s individual outcome:
Building wealth certainly still is a good idea: In the AI as a normal technology scenario, which is likely at least in the medium term, and for the modest probability that the permanent wealth freeze does in fact happen. Building capabilities relevant in the AI as a normal technology world naturally goes along with building wealth
Obtaining US citizenship could be helpful in scenarios where an ASI ends up in democratic US control, with some sort of
America firstoutcome not unlikely. In the meantime, the US is one of the best places to build wealth. A more diversified setup having also e.g. UK citizenship improves our chances even more. Chinese citizenship could also be helpful, but is more or less impossible to obtain: even Chinese green cards are very difficultHaving strong relationships is helpful in a lot of scenarios, and helps buffer the emotional impact of rapid change. This includes, where possible, relationships with powerful stakeholders in the AI and political scenes, but also everyday relationships with friends and family
At the same time we can work to contribute to achieving a better outcome collectively: such a crucial time is an exceptional opportunity to draw meaning from one’s work. Fatalism is not required, and if things do turn out bad, it sure would feel better to go down swinging. For example through donating to organizations such as METR or through work.
And perhaps most important of all, we should try to relax and approach whatever we are doing with a sense of ease: the only life we know for sure we have is the one right now.