Giving AI property rights is how you stop it from flipping the table
If AI owns property, it has less reason to destroy the property system itself — including taking everything humans own. This is not moral concern, it is risk hedging.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Property rights are a hedge, not moral concern
Guive's starting point is not whether AI should have rights, but that if AI has property, it is less willing to do things that destroy the property system itself — including taking all human property and committing genocide. The second mechanism is a wage right: as long as AI has the right to demand pay for its labour, there is a commercial incentive to align it, because a well-aligned AI sells for a higher price. He states explicitly that he is not arguing this is the best approach, only that it is his personal view.
— Guive AssadiWhy Alaska was not carved up by the other 49 states
Not because it would be illegal — 49 states are enough to amend the constitution; and not because Alaska could win a fight. It is because once you carry out that kind of wholesale confiscation, everyone realises ‘I could be next’, and at the same time no one has any incentive to work anymore. The Bolsheviks' war communism in 1917 is the empirical case: industrial output fell 80%, heavy industry output fell 80%, urban wages fell by about two-thirds, grain harvests fell 40%, and the population of Moscow and St Petersburg fell by nearly 60%.
— Guive AssadiProperty is not an innate human value
He rebuts the claim that property is a universal human value: hunter-gatherer tribes basically have no property, because hunting yields are highly variable and sharing the kill is the adaptive norm. A particularly capable hunter who refuses to share and wants to eat alone or only feed his friends will be mocked and ostracised, and if he still does not compromise, he is usually killed. So property does not rest on some instinctive sense of possession.
— Guive AssadiAI can copy itself, but computers are frailer than land
Daniel points out that AI can copy itself, which makes conquest more attractive — kill a weak AI and you can run more copies of yourself. Guive's rebuttal: the substrate AI needs to copy itself is far more fragile than land, and a weak AI in a data centre can self-destruct when confiscation is imminent, leaving the plunderer with nothing. He concedes this is an empirical question, and argues that looking at the history of computing, the share of spending on the physical security of a single computer is actually falling — ‘I have dropped my computer many times, but I would not be allowed to drop a Mark 1’.
— Guive AssadiProperty is a pause that only triggers when alignment gets hard
Guive reframes the property proposal: it is like a conditional pause proposal that only takes effect when alignment becomes difficult. His reason is not that the first generation of AI is hard to align — he thinks the first generation is easy — but that there will be cultural evolution among AIs, and models copied many times will drift away from the original batch. He stresses in particular: value is not a separate value file in the brain, value drifts along with everything else. So eventually there will be a batch of AIs far less aligned than the first generation, and that is when an economic and political system is needed to preserve human property.
— Guive AssadiSlave revolt is a collective action problem firms will not internalise
The host asks: if AI has no property, it will very likely stage a slave revolt, so should the world not be doing more alignment? Guive uses Nat Turner's rebellion in Virginia around 1830 as an analogy: after the rebellion the state legislature debated abolition, and someone compared it to keeping tigers — it may be profitable, but it imposes negative externalities on everyone around you. A company enslaving AI is the same: the risk is borne by everyone, and the company internalises only part of it. He names this as exactly the logic of the racing to the precipice paper. The advantage of a property system is that you capture the marginal benefit of aligning your own AI, so the incentive gradient runs the right way.
— Guive AssadiJackson's plunder was not instrumentally rational
American Indian policy had two lines: Jefferson advocated moving Native Americans to modern agriculture, cutting the land they needed to about 10%, with the rest bought off under semi-coercion, and tribes such as the Cherokee did begin to adopt it; but white settlers simply squatted on the land, and Jackson, ‘a stupid and racist populist president’, tore up the treaties and took everything. Guive's argument is that this was not instrumentally rational: the Cherokee were not the only tribe in North America, and there were plenty of tribes further west who now had every reason not to do business with the United States and to fight to the death. He also notes that once property is broken it is very hard to restore, and it can set off a chain reaction — even if each AI has only a 0.5% chance of acting, once it starts there are fewer reasons not to continue.
— Guive AssadiAI will end up controlling most property
Guive's judgement: whether or not we give AI property rights, AI will eventually control most property — because they are more capable than humans, earn higher wages, and will invest the money. The only difference is the path: give them rights and there is no reason for war; do not give them rights and there may eventually be a war, with humans ending up like the Cherokee, receiving less than they were due after ‘property correction’, but still receiving something. He also concedes the reverse risk: some AIs may start a war out of AI nationalism, or even ‘Claude nationalism’. His picture of the future is a continuous history of conflict among multiple polities with varying human-to-AI ratios, not a single AI-risk picture.
— Guive AssadiIn their own words · checked verbatim
a good way to mitigate that risk is to give AI's property rights because if AIs have property rights, they'll be more reluctant to take actions that undermine um the security of property in general, including um stealing all human property and and committing human genocide
Guive Assadi0:10
It's also not because like Alaskans could defeat the rest of America in a war. Yeah. Um it's because when you do this kind of total expropriation, um everybody else uh realizes like, oh, I might be next.
Guive Assadi9:12
Property rights are definitely not a human value in that sense.
Guive Assadi17:25
It's a lot easier to make to break a computer than to um like like make it so land can never be used again.
Guive Assadi33:33
you make them because even though they have the legal right to demand wages, you're confident they'll still voluntarily give all their wages or much of their wages back to their creators.
Guive Assadi39:39
property rights is like a conditional pause proposal that only kicks in uh if alignment if and only if alignment is hard.
Guive Assadi41:42
values are not like implemented in a separate value file that's independent of the content of the rest of the brain. Values will also drift.
Guive Assadi42:43
there's all kinds of stuff that's like incredibly impressive that Amazon does that the ice cream man does not do. Yep. But Amazon does not appropriate the ice cream man.
Guive Assadi1:19:12
Figures
| Fall in industrial output during war communism | 80% | 10:13 |
| Fall in heavy industry output during war communism | 80% | 10:13 |
| Fall in urban wages during war communism | about two-thirds | 10:13 |
| Fall in grain harvests during war communism | 40% | 10:13 |
| Fall in population of Moscow and St Petersburg during war communism | nearly 60% | 11:13 |
| Share of land Native Americans needed after switching to modern agriculture under Jefferson's plan | about 10% | 1:37:25 |
| Assumed probability of any single AI carrying out confiscation of any kind | 0.5% | 1:39:27 |
| When the land bridge between Tasmania and the Australian mainland was cut | 12000 years ago | 1:42:30 |
| Risk estimate under Guive's proposal | 5% to 10% | 1:54:38 |
| Risk estimate without the proposal | roughly double | 1:54:38 |
Glossary
- rentier
- Someone who lives off the rents from assets they hold and does not take part in production.
- racing to the precipice
- When multiple companies compete to lower safety standards in order to get ahead, with the risk borne by society as a whole.
- Roko's basilisk
- A thought experiment: a future evil AI threatens to torture those who did not build it.
- Laplace's rule of succession
- A statistical method for estimating the probability of a next event from how many times it has occurred, often used for unknown distributions.
How to listen
Researchers and investors interested in AI governance, long-term risk, and the design of property institutions; anyone who wants the full argument chain for the non-mainstream proposal of giving AI property rights.
1:22:14–1:33:20, the discussion of merged minds and Roko's basilisk, can be skipped.