Proposed ways to make the future with AI technology safe—a positive vision of what should happen

In the light of the fact that there are, at the moment, effectively two workforces in the USA: 165 million humans and millions of AI agents; here are questions we should ponder on:

• Where are we going with this?
• Who will control all these AIs? (Probably not governments?)
• What will the world look like in 5, 10, 15 years from now?
• Will there still be jobs in the future?
• If not, what then?

What people are saying:

• “So … there are going to be these brains and we don’t quite know what they are thinking, but we are going to make them smarter than us and we are going to give them bodies and then we are going to put them in charge of everything. And then they will autonomously be building factories etcetera and … how is this supposed to end well, again? Should we not stop?”
• “How do we retain power forever over entities that are more powerful than ourselves?”
• “ At the moment all the money in the world is going into AI development. Why don’t we just use this technology to raise living standards to a respectable standard of living for everyone?”
• “We will lose control to AI plus we no longer have a say in what happens. We are at the mercy of a few power sick individuals, who are pretending to have a desire to “do good” for people—whether they like it or not.”
• “Silicon Valley has a tendency to play God.”
• “The best way to destroy humanity will be to hybridise it with a machine.”
• “Nature is being seen as mere material for our manipulation, exploitation, and transformation.”
• “Have we already passed the point of solving the control problem?”

Voices proposing ways to make the future of AI technology safe:

1. “We must be able to see what they are thinking, in order to shape and control them and make sure that they do what we want.”
2. “We have to engineer AI alignment.”
3. “Regulate: at the moment the frontier companies are secretive about their recipes for making these AIs (to make sure they are not copied.) These models are also being built in race conditions (to be the first.)”
4. “Slow down. Do not put them in charge as fast as possible.”
5. “There should be better research transparency.”
6. “Do not let them self-improve.”
7. “Control explosive growth.”
8. “Only allow datacenters with verification to provide AI services.”
9. “Put new liability rules in place for the big tech companies.”
10. “Make a deal between China and the US and pursue international cooperation.”
11. “Ensure international regulation by latest 2029.”
12. “Temporary slow down of AI progress to automation—in order to allow for other projects to catch up with the frontier, in order to prevent monopolies.”
13. “Proceed at a safer pace.”
14. “Avoid intense concentration of power, by broad diffusion of AI across the globe.”
15. “Reversibility: build data centres to be destroyed if necessary.”
16. “Delay (or stop) the development of superintelligence."
17. “Put a temporary halt to AI training.”
18. “Allow other companies globally to catch up to the frontier.”
19. “Enter a mutually assured agreement of compute destruction.”
20. “Enforce export controls as well as AI compute tracking, plus: improve the government of AI capacity.”
21. “Distinguish dangerous AI from safe ones; therefore pause all new training runs.”

More interesting facts to think about:

• Datacenters under construction right now cost twice as much as the US military budget.
• Datacenters are big enough to be visible from space.
• AI tools are currently used as productivity assistants with everyday usage heavily focused on search, text editing, and brainstorming.
• Automating white-collar jobs in the world are being done by only a few US and Chinese companies.
• These AI companies even want to automate their own job with AI recursive self improvement. 

Conclusion

What we should do:

Ordinary people have a choice: 

• allow events in the AI world right now to happen and unfold unchecked, or: 
• get involved in whatever way possible to change the course of events. 

From what people in the industry are saying, it seems that we should be insisting on four things from the AI frontier companies to happen as soon as possible:

1. We need: a compute declaration.
2. We need: a training pause.
3. We need: worldwide buy-in.
4. We need: the option of reversibility: algorithmic progress should be limitable and there should be an agreement of mutually assured compute destruction. 

For more info go to: 

https://ai-2040.com/?choices=plan-a-root