Vitalik Buterin, the creator of Ethereum, questions the race to develop increasingly autonomous artificial intelligence systems and rejects the idea that accelerating their development is inevitable. He also criticizes those who work at companies pushing advanced AI on the argument that they will be able to make it safer from within.
Buterin believes that those concerned about the risks of AI should not contribute to the development of frontier autonomous systems and questions the concentration of power justified in the name of this technology's future safety. "Don't work for a company that is making frontier fully autonomous AI capabilities go even faster. Don't live in the San Francisco Bay Area," he says.
Vitalik, AI and effective altruism
The statements are part of a reflection on artificial intelligence and certain strands of effective altruism linked to AI safety. Buterin acknowledges the influence effective altruism has had on his intellectual formation and maintains his defense of values such as concern for others, cosmopolitanism and the rigorous use of resources. However, he questions some of the strategies that have emerged around AI.
Vitalik argues that in effective altruism circles related to AI there are influential people whose strategy consists of gaining as much power as possible in order to be in a position to act when a crucial moment arrives. That reasoning, he warns, can serve to justify any concentration of power. Give me power so I can do X is an equally convincing argument, regardless of what X is, he states. Until that hypothetical decisive moment arrives, the actions of a person who accumulates power for altruistic reasons can be exactly the same as those of someone who does so out of self-interest.
Vitalik does not believe the AI race is inevitable
Buterin also questions the argument that, since advanced artificial intelligence will end up being developed anyway. He defines this reasoning as inevitabilism. According to Buterin, AI is not a market in which there are infinite actors capable of immediately replacing one another. Much of the progress is concentrated in a few people and companies. For this reason, he argues that if any of those actors decides to stop or slow down, the development of the technology can be slowed.
He also rejects the idea that the only possibilities are accelerating toward full automation or abandoning the development of artificial intelligence. In this sense, he proposes developing forms of partial automation that enhance human capabilities instead of fully replacing them.
This model, he argues, would make it possible to extend the period in which humans and AI working together can outperform exclusively automated systems and would offer more room to manage an eventual transition toward superintelligence. Buterin's criticism does not amount to a general rejection of effective altruism. What he questions is that certain ideas associated with this movement can be used to justify present-day decisions by appealing to enormous but hard-to-verify benefits in the distant future.
Results are impossible to verify
His warning is that, if a story about enormous long-term positive consequences can be used to justify practically anything, it stops being useful for deciding what to do. In the case of AI, Buterin thus questions whether hypothetical future benefits can be used to justify decisions that today contribute to accelerating the development of increasingly autonomous systems.
According to Vitalik, promises about the enormous future benefits of AI can become a justification for present-day decisions whose results are impossible to verify. The more distant the promised benefit, the greater the risk that an argument only needs to seem convincing, rather than demonstrate that it actually works, he argues