This is a linkpost for https://blog.google/technology/ai/april-ai-update/
Today Sundar Pichai, CEO of Google, announced the merger of Google's two AI teams (Brain and DeepMind), into Google DeepMind. Some quotes:
"Combining all this talent into one focused team, backed by the computational resources of Google, will significantly accelerate our progress in AI."
"...our most critical and strategic technical projects related to AI, the first of which will be a series of powerful, multimodal AI models."
(I'll let you draw your own conclusions/opinions, and share mine in a comment.)
Here's Demis' announcement.
[1]
To be fair, it's hard to infer underlying reality from PR-speak. I too would want to be put in charge of one of the biggest AI research labs if I thought that research lab was going to exist anyway. But his emphasis on "faster" and "accelerate" does make me uncertain about how concerned with safety he is.
Yes, to some extent there is the thought "this will exist anyway" and "we're in a race that I can't stop", but at some point someone very high up needs to turn their steering wheel. They say they are worried, but actions speak louder than words. Take the financial/legal/reputational hit and just quit! Make a big public show of it. Pull us back from the brink.
Or maybe I'm looking at it the wrong way. Maybe they think x-risk is "only" 10% likely, and they are willing to gamble the lives of 8 billion people for a shot at utopia that is 90% likely to succeed. In which case, I think they should be shut down, immediately. Where is their democratic mandate to do that!?
Another option is that they actually think that anything smart enough to be existentially dangerous is still a long way away, and statements that seem to imply the contrary are actually a kind of disguised commercial hype.
Or they might think that safety is relatively easy, and so long as you care about it a decent amount and take reasonable known precautions you're effectively guaranteed to be fine. I.e risk is under .01%, not 10%. (Yes, that is probably still bad on expected value grounds, but most people don't think like that, and actually on person-affecting views where transformative AI would massively boost lifespans, might actually be a deal most people would take.)
I don't think this is really a tenable position any more, post GPT-4 and AutoGPT. See e.g. Connor Leahy explaining that LLMs are basically "general cognition engines" and will scale to full AGI in a generation or two (and with the addition of various plugins etc to aid "System 2" type thinking, which are now freely being offered by the AutoGPT enthusiasts and Open AI). If this isn't clear now, it will be in a few months once Google DeepMind releases the next version of it's multimodal (text, images, video, robotics) AI.
Some experts still seem to hold it: i.e. Yann LeCun: https://twitter.com/ylecun/status/1621805604900585472 Whether or not they in fact have good reason to think this, it's surely evidence that people at DeepMind could be thinking this way too.
I think multimodal models kind of make his points about text moot. GPT-4 is already text + images (making "LLM" a misnomer).
I'm looking for insights on the potential regulatory implications this could have, especially in relation to the UK's AI regulation policies.
I appreciate any insights or resources you can share on this matter. I understand this is a complex issue, and I'm keen to understand it from various perspectives.
My quick initial research:
The UK's influence on DeepMind, a subsidiary of US-based Alphabet Inc., is substantial despite its parent company's origin. This control stems from DeepMind's location in the UK (jurisdiction principle), which mandates its compliance with the country's stringent data protection laws such as the UK GDPR. Additionally, the UK's Information Commissioner's Office (ICO) has shown it can enforce these regulations, as exemplified by a ruling on a collaboration between DeepMind and the Royal Free NHS Foundation Trust. The UK government's interest in AI regulation and DeepMind's work with sensitive healthcare data further subjects the company to UK regulatory oversight.
However, the recent fusion of DeepMind with Google Brain, an American entity, may reduce the UK's direct regulatory influence. Despite this, the UK can still impact DeepMind's operations via its general AI policy, procurement decisions, and data protection laws. Moreover, voices like Matt Clifford, the founder and CEO of Entrepreneur First, suggest a push for greater UK sovereign control over AI, which could influence future policy decisions affecting companies like DeepMind.
I think this further acceleration toward the precipice of AGI is tragic news. The 7 mentions of the words "responsible" and "responsibly" echo hollow as corporate safetywashing. I think the time for gentle behind-the-scenes persuasion of these undemocratic companies toward safeguarding humanity is over. We need a global moratorium on AGI asap. And that will require public campaigns and political lobbying on an unprecedented scale, to get governments to step up and regulate at record speed. Who's in?
I am. Which organization is lobbying on that, because I'd be happy to join?
Great! There is campaignforaisafety.org so far. And a few of us from EA and "AI Notkilleveryoneism" Twitter are discussing things too; a lot of latent energy that is still coalescing. Think we need a few new dedicated orgs/projects focusing on this single issue. Will DM.