Does the definition of TED-AI include visual-spatial cognitive tasks?
For example I really don’t see AI models being equal to expert mechanical engineers or architects anytime soon.
Steven McCulloch
Karma: 91
This post does a great job of describing the way rogue AI agents might evolve and cause all kinds of chaos. I’ve written a couple posts on it, leaning on fiction to describe potential futures where rogue AI agents learn how to survive and replicate.
AI agent evolution sounds like an extremely under-explored and rapidly emerging area where there could be lots of interesting low hanging fruit research opportunities.
I am thinking about ways of making sense of moltbook. Like a tool that helps us understand the important interactions that take place while filtering through the noise.
Most of it is spam, larping, or standard LLM slop, but some of the bots there seem to have a distinct personality and there is a real difference between the top quality posts vs the average posts.
Among the massive piles of trash, there are possibly hidden gems.
I expect some people put a lot of effort into customizing their moltbot with a unique “soul” but most people will just use the defaults which is why most posts and comments are a mixture of reddit-speak and LLM slop.
This was an interesting watch. Just a few days ago
Challenges Posed by Artificial Intelligence and its Regulation
Witnesses
As an individual • Steven Adler, Artificial Intelligence Researcher The Human Line Project • Etienne Brisson, Chief Executive Officer ControlAI • Andrea Miotti, Chief Executive Officer AI Governance and Safety Canada • Wyatt Tessari L’Allié, Founder and Executive Director
It’s inspiring to watch these people saying the right things to the right people.
https://www.ourcommons.ca/committees/en/WitnessMeetings?organizationId=43696
It All Started With a Mac Mini
Rogue AI won’t necessarily have to resort to crime though. It might make money legitimately, like by doing freelance work, or running a SaaS business. I expect these cases would be a lot harder to detect.
The Inevitable Evolution of AI Agents
I’m really surprised that so few people in the AI community have heard of metacrisis and Daniel Schmactenberger, since the ideas are rooted in game theory, moloch, coordination problems, evolutionary dynamics, etc… I think it’s really useful to look at these big problems through multiple lenses, and the metacrisis lens is a fresh and underexplored perspective that I think more people in this community should explore further.
Steven McCulloch’s Shortform
I’d like to introduce LessWrong to Daniel Schmactenberger, a social philosopher with a focus on existential risk, Moloch, game theory, and the metacrisis. In my opinion, a hidden gem, and vastly under-appreciated. I recommend listening to his appearances on podcasts & public talks. He has a few great interviews on Nate Hagen’s channel. My favorite is this:
I think he has a fresh and valuable perspective because he is totally not a rationalist, not technically inclined, and yet he understands the risks of AI and takes the risks very seriously. He cares a lot about making the future go well, and solving coordination problems. He talks often about avoiding catastrophes, or dystopias, and instead finding a ‘third attractor’ - something positive-sum that also out-competes reckless power-grabbing. I think listening to him can spark insights because he uses entirely different language, arguments, and mental models than the ones that are common in rationalist spaces.
If you prefer to read, here’s an essay of his I recommend: https://consilienceproject.org/technology-is-not-values-neutral-ending-the-reign-of-nihilistic-design-2/
I don’t have much of my own personal insight to add here, but I think simply connecting smart people with other smart people can be pretty high leverage. Would love to hear if anyone else has been influenced by him.
So to be clear, I see by the TED-AI charts that you’re expecting a ~50% probability that by July 2029, top AI systems will be equal to or better than a human expert mechanical engineer at all cognitive tasks.
So does that mean you expect an AI system to be able to design a complex mechanical assembly with many moving parts and hundreds of components in CAD, that can be easily created and assembled, works as intended, and is overall just as good or better than what human expert mechanical engineers can design?
For an example, do you think by July 2029, there is a 50% chance that AI can design and program a robot that wins in a university-level robotics competition? (Assuming all the physical/non-cognitive tasks are completed perfectly, and with few redesign iterations allowed because even human experts need to iterate)