Виталик Бутерин предупреждает об угрозах сверхразумного ИИ из-за недостатка мирового согласования ценностей Translation: Vitalik Buterin Warns of Superintelligent AI Threats Due to Insufficient Global Value Alignment

Humanity is not adequately managing the alignment of global values to ensure that superintelligent artificial intelligence will serve the interests of all people. This statement was made by Vitalik Buterin, co-founder of Ethereum.

He commented on a post by the 3D artist known as Utah teapot.

“How can the concept of AI alignment even exist? I always ask, ‘aligned with what?’ No one ever answers anything except ‘with human values.’ But coherent, non-contradictory human values do not exist. I still don’t understand what we are even studying. It’s absurd that this is turning into a growing industry. It’s make-work,” he wrote.

Buterin emphasized that there are many things in the world that contradict ethical principles, such as murder or the imprisonment of innocent people.

“[…] We are not yet good enough at alignment to ensure that a single superintelligent AI will avoid even this,” added the developer.

Utah teapot noted that the development of aligned AI is trending towards restricting public access to its development. At the same time, large companies are selling versions of their AI systems that contribute to negative phenomena and serve as tools of war.

“I am deeply concerned that centralizing access to AI technologies allows what are essentially security issues to be imposed—like discussions around ‘psychosis’ from large language models (LLMs). This poses a risk of harm to unique or marginalized cultures and their values,” the user stated.

Buterin shares many of these concerns.

“I believe the greatest risks will come from military and other powerful structures that can exempt themselves from the security rules that apply to everyone else by default,” added the co-founder of Ethereum.

Utah teapot cited the AI startup Anthropic as an example, which develops alternative versions of “civilian” models for governments. These could be used for military or intelligence purposes.

Buterin emphasized that the likelihood of a catastrophic scenario for humanity due to AI increases if there is only one superintelligent agent-based AI with its own will and ability to act as a subject.

If the environment is pluralistic, no single system can completely control the situation. However, merely relying on market mechanisms is insufficient to create this environment. Deliberate efforts are needed, including changes to laws and incentives that may not be favorable to large corporations.

Recall that in 2022, India’s Minister of Information Technology, Rajeev Chandrasekhar, called for the development of global standards to ensure the safety of AI for humanity.