Demis Hassabis Outlines a Plan to Harness AGI Safely
Demis Hassabis has a plan to harness AI safely
I believe we are standing at a pivotal moment in human history where Artificial General Intelligence is only a few years away. This new system will exhibit all the cognitive capabilities of the human brain. To navigate this transition, I am proposing a comprehensive framework designed to ensure frontier AI is developed safely and benefits all of humanity.
Artificial General Intelligence, a system that exhibits all the cognitive capabilities the brain has, is probably only a few short years away.
- noelwelsh
The premise is "Artificial General Intelligence (AGI), a system that exhibits all the cognitive capabilities the brain has, is probably only a few short years away".
If this is true, establishing an institution to ensure things like "publishing model cards with technical details, maintaining strong internal cybersecurity, vetting key personnel, and providing sufficient resourcing for safety and security research" is really mostly irrelevant.
TFA does talk about what really needs to be done, but punts this into future work: "Even if we solve these hard technical challenges, there will be further complex economic and philosophical questions to tackle: what sorts of new economic models will be needed to help everyone thrive in a post-scarcity world? What values do we want to live by, what will meaning and purpose be, and how might even the human condition itself change?"
There's also a need to consider the rights that this new intelligence should have.
- thegrim33
Spoiler: The plan is .. add massive regulation, but only to the US, don't affect other countries developing it in any way other than "setting a good standard that'll hopefully influence them". Seems like an airtight plan.
- khurs
>This is a pivotal moment in human history. Artificial General Intelligence (AGI), a system that exhibits all the cognitive capabilities the brain has, is probably only a few short years away.
There is a heatwave in London, perhaps Demis needs to stay out of the sun and drink more water.
Or perhaps he is seeking more funding/a fight to maintain his divisions AGI research budget.
- segmondy
Not surprised, seems these labs start calling for regulation once they are losing or have competition. OpenAI started calling it for it once Anthropic got better, Anthropic started calling for it once the Chinese models got good, Google is now calling it for it because they are falling far behind.
- minraws
Demis has only 1 plan, how to dodge releasing a new model at this point, jokes aside I value thinking about AI safety, but are we really so close to AGI? It doesn't feel like it, LLMs still diagnose my headache as a chronic illness or a brain tumor from time to time... honestly stressful.
- pshirshov
Blah-blah-singularity, so let's cripple the models so much they refuse to talk about React, because who knows if you are not cooking chemical weapons or meth in your browser's DOM, right?
- gruez
The proposal:
>The American government, he says, should develop a system for testing the safety of new AI models before they are released. “It’s important that it’s not just an industry body,” he adds. But a regular government agency wouldn’t do either. “It would not be able to move fast enough, or have the right resources.” Instead, Sir Demis suggests taking inspiration from FINRA, the Financial Industry Regulatory Authority, a private agency in America that regulates brokers and stock markets.
- bloppe
> Earlier proposals from America and the European Union looked to the amount of computing power used to train a model as a rough guide for when oversight was required. Sir Demis instead suggests designating AI models as “frontier” if they meet certain thresholds on a selected set of benchmarks. The creators of those models would then be designated as “frontier labs” with extra responsibilities. Sir Demis is proud of the “elegant” way that approach sidesteps the question of whether academic or open-source models should be included or not.
Goodhart's law applies in reverse as well. Once the chosen benchmarks are known, model makers will aim to come in just under the threshold on those specific benchmarks, while maximizing their scores on other benchmarks.