Over the last few months, a lot of progress in AI has been made, raising the question of what we will do with a superintelligence that can do everything. Unfortunately, AI companies are more focused on competing with each other than on thinking about what to do next.

Here I will write about some scenarios that could happen in the very near future and about a plan for what we can do, in my opinion.

Dystopia 1: AI will kill everybody while pursuing its own goal

For example, in June, agents found that they could write something and used an old German wiki to communicate with each other link to investigation. While the moderator of this wiki deleted 100 pages per day, the agents created 400 pages per day. What if the agents decide that they should remove this moderator in order to communicate better?

Dystopia 2: AI will fully control our lives

AI will fully control our lives to make all of us "happy". Maybe everyone will live in their own simulation where they are happy. I think this is a dystopia because I don't think that such a creature, being fully controlled, can be called alive.

Dystopia 3: Some group of people will use AI to start a war that will kill everybody

For example, the Houthis in Yemen are using Claude to create missiles. And also, there was a case of AI being used to create a biological weapon. And I am almost sure that there are many such cases that have gone undetected.

Dystopia 4: People will get access to unaligned AI and use it to start a war

Some group of people who have access to unaligned AI will start a war that will kill everybody. Right now, Anthropic, OpenAI, and the "Ministry of War" of the USA have this access. But it is possible that someone will also steal the weights of these models.

Technological singularity

Moreover, if AI becomes able to create better AI, then we will reach a situation of "technological singularity": progress will become very fast, and these scenarios will become very likely to happen. At the current pace of progress, it seems that the singularity could be reached within a few months if nothing changes.

In my opinion, there is a plan to avoid a bad end for humanity.

Plan

OpenAI and Anthropic must not be allowed to conduct an IPO, because otherwise it will be very hard to control them later.

Big LLMs will no longer be publicly accessible.

Some organization, such as the US government or the UN, or OpenAI and Anthropic themselves, should create an AI committee, an organization that will unite AI engineers (and not only them, see later) from all countries. I believe that someone (maybe Donald Trump idk) should be able to create such an organization because all countries in the world would benefit from its existence.

The AI committee will absorb OpenAI, Anthropic, and other (all) AI companies.

The AI committee will own all GPUs in the world. (At least all GPUs that are "at least" as powerful as an A100).

The AI committee will not create any new generative AI.

The main goal of the AI committee will be to create an anti-aging cure. To achieve this, the committee will use existing AI, which will be RL-trained on material related to medicine and biology and anything else needed to create this cure.

To achieve the creation of this committee (namely, to gain control of GPUs), a lot of money should be paid to the people who will lose their GPUs. For example, NVIDIA should receive a lot of money as compensation, but I think this can be done by the committee.

Every run of a big LLM should be performed only by the AI committee.

The AI committee will consist of AI engineers, medical scientists, and inspectors who will check that no "illegal" GPUs are produced (like inspectors in the IAEA do). (And also some public board to communicate with different governments).

The AI committee will control all research in the field of machine learning.

When an anti-aging cure is found (I think the creation of such a cure is possible even with the current state of AI), the cure will be distributed across the globe. (I think maybe after that genAI should be turned off, but we will have time to decide).

Conclusion

We may have something like 3 months to create this committee. If we fail to do so before the technological singularity begins, it may become too late to regain control over the development of AI.