AI, extinction, nuclear war, pandemics ... That's expert open letter bingo
- Reference: 1685477646
- News link: https://www.theregister.co.uk/2023/05/30/ai_human_extinction/
- Source link:
More aptly a statement, the message from the Center for AI Safety (CAIS) signed by individuals like AI pioneer [1]Geoffrey Hinton , OpenAI CEO Sam Altman, encryption guru Martin Hellman, Microsoft CTO Kevin Scott, and others is a [2]single declarative sentence predicting apocalypse if it goes unheeded:
"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."
[3]
Why so brief? The goal was "to demonstrate the broad and growing coalition of AI scientists, tech leaders, and professors that are concerned by AI extinction risks. We need widespread acknowledgement of the stakes so we can have useful policy discussions," CAIS director Dan Hendrycks told The Register .
[4]
[5]
CAIS makes no mention of artificial general intelligence (AGI) in its [6]list of AI risks , we note. And current-generation models, such as ChatGPT, are not an apocalyptic threat to humanity, Hendrycks told us. The warning this week is about what may come next.
"The types of catastrophic threats this statement refers to are associated with future advanced AI systems," Hendrycks opined. He added that necessary advancements needed to reach the level of "apocalyptic threat" may be as little as two to 10 years away, not several decades. "We need to prepare now. However, AI systems that could cause catastrophic outcomes do not need to be AGIs," he said.
Because humans are perfectly peaceful anyway
One such [7]threat is weaponization, or the idea that someone could repurpose benevolent AI to be highly destructive, such as using a drug discovery bot to develop chemical or biological weapons, or using reinforcement learning for machine-based combat. Humans are already quite capable of manufacturing weapons that can take out a person, neighborhood, city, or country, mind you.
AI could also be trained to pursue its goals without regard for individual or societal values, we're warned. It could "enfeeble" humans who end up ceding skills and abilities to automated machines, causing a power imbalance between AI's controllers and those displaced by automation, or be used to spread disinformation, intentionally or [8]otherwise .
[9]
Again, none of the AI involved in that needs to be general, and it's not too much of a stretch to see the potential for current-generation AI to evolve to pose the sorts of risks CAIS is worried about. You may have you own opinions on how truly destructive or capable the software could be, and what it could really achieve.
[10]OpenAI calls for global watchdog focused on 'existential risk' posed by superintelligence
[11]US cyber chiefs warn AI will help crooks, China develop nastier cyberattacks faster
[12]Crooks don't need ChatGPT to social-engineer victims, as they're more than happy to demonstrate
[13]Will LLMs take your job? Only if you let them
It's crucial, CAIS' argument goes, to examine and address the negative impacts of AI that are already being felt, and to turn those extant impacts into foresight. "As we grapple with immediate AI risks … the AI industry and governments around the world need to also seriously confront the risk that future AIs could pose a threat to human existence," Hendrycks [14]said in a statement.
"The world has successfully cooperated to mitigate risks related to nuclear war. The same level of effort is needed to address the dangers posed by future AI systems," Hendrycks urged, with a list of corporate, academic and thought leaders backing him up.
Musk's not on board
Other signatories include Google DeepMind principal scientist Ian Goodfellow, philosophers David Chalmers and Daniel Dennett, author and blogger Sam Harris and musician/Elon Musk's ex, Grimes. Speaking of the man himself, Musk's signature is absent.
The Twitter CEO was among those who signed an [15]open letter published by the Future of Life Institute this past March urging a [16]six month pause on the training of AI systems "more powerful than GPT-4." Unsurprisingly, OpenAI CEO Altman's signature was absent from that particular letter, ostensibly because it called his company out directly.
OpenAI has since issued [17]its own warnings about the threats posed by advanced AI and called for the establishment of a global watchdog akin to the International Atomic Energy Agency to regulate use of AI.
That warning and regulatory call, in a case of historically poor timing, came the same day Altman [18]threatened to pull OpenAI, and ChatGPT with it, from the EU over the bloc's [19]AI Act . The rules he supports are one thing, but Altman told Brussels their idea of AI restriction was a regulatory bridge too far, thank you very much.
[20]
EU parliamentarians [21]responded by saying they wouldn't be dictated to by OpenAI, and that if the company can't comply with basic governance and transparency rules, "their systems aren't fit for the European market," asserted Dutch MEP Kim van Sparrentak.
We've asked OpenAI for clarification on Altman's position(s) and will update this story if we hear back. ®
Get our [22]Tech Resources
[1] https://www.theregister.com/2023/05/01/ai_geoffrey_hinton_resigns/
[2] https://www.safe.ai/statement-on-ai-risk
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZHZx9ftzi0zfvUXUG3aqzwAAAMk&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZHZx9ftzi0zfvUXUG3aqzwAAAMk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZHZx9ftzi0zfvUXUG3aqzwAAAMk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.safe.ai/ai-risk
[7] https://www.safe.ai/ai-risk
[8] https://www.theregister.com/2023/03/21/stanford_ai_alpaca_taken_offline/
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZHZx9ftzi0zfvUXUG3aqzwAAAMk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[10] https://www.theregister.com/2023/05/24/openai_superintelligence_global_agency/
[11] https://www.theregister.com/2023/04/12/us_chatgpt_threat/
[12] https://www.theregister.com/2023/04/29/kaspersky_social_engineering/
[13] https://www.theregister.com/2023/05/15/will_llms_take_your_job/
[14] https://www.safe.ai/press-release
[15] https://www.theregister.com/2023/03/29/uk_seeks_lighttouch_ai_legislation/
[16] https://www.theregister.com/2023/03/29/uk_seeks_lighttouch_ai_legislation/
[17] https://www.theregister.com/2023/05/24/openai_superintelligence_global_agency/
[18] https://www.reuters.com/technology/openai-may-leave-eu-if-regulations-bite-ceo-2023-05-24/
[19] https://www.theregister.com/2023/04/18/eu_lawmakers_want_ai_regulation/
[20] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZHZx9ftzi0zfvUXUG3aqzwAAAMk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[21] https://www.reuters.com/technology/openai-ceos-threat-quit-eu-draws-lawmaker-backlash-2023-05-25/
[22] https://whitepapers.theregister.com/
"There's nothing specifically "AI" about it."
Absolutely. There's nothing in this list of alleged risks that couldn't be accomplished by a dedicated team of nutters (and / or government funded "research"). We seem to be managing quite well in the chemical / biological weapons department, and misinformation is rife, no need for AI.
Who exactly are the CAIS, or whatever they're called? How long have they existed, how are they funded, what are their affiliations, and what are the agendas of those they are affiliated with / funded by? It's all very well telling us who's supporting the statement, but a little more information about the organisation would be helpful. I may well just be a cynical old git, but my Spidey sense is tingling a bit here.
Should be...
"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."
But as with whatever other global priorities we *should* have, we'll keep pushing "AI" it if makes us a dime, screw the risks. We could redesign our LLMs to be based on legally obtained, sanely curated, inspectable data, with audit trails, guard rails and watermarking, but we'd looooooooooose money if we did that. So we'll just keep making empty pronouncements and jamming "AI" into toasters.
Call me cynical, but firms who have already developed functioning AI products are calling for AI development to be regulated? Is this a case of pulling the ladder up behind them to ensure no one else can compete?
A 'bang' or a 'whimper'?
"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."
According to T. S. Eliot, the world will end not with a bang, but a whimper*. My concern is that however much we may prepare for an AI created global catastrophe, it is far more likely to be accidental destruction than deliberate or even foreseen. When the Curies discovered radioactivity there was a rush to develop and sell radioactive 'medicines' so that you could ingest these newly discovered elements. No-one seems to have considered that they could possibly be harmful. X-rays were once used to remove unwanted body hair. I suspect that an AI controlling something humans do not realise is critical could easily occur and some failure or mistake made that endangers a large part of humanity or the habitable planet.
The problem is that AI is turning out to be really useful for some things we cannot get rid of it.
Of course the signatories will doubtless be criticised for having a short list of risks (pandemics and nuclear war), but if every global threat were to be included it would make for quite an unwieldy sentence. (I bet people will say they should have included man-made global warming, plastics pollution, anti-biotic resistance, over-population, and genetically modified crops.)
*https://interestingliterature.com/2021/02/eliot-this-way-world-ends-not-bang-whimper-meaning-analysis/#:~:text=%27This%20is%20the%20way%20the%20world%20ends%27%2C%20T.%20S.,encountered%20%27The%20Hollow%20Men%27.
All this panic
of AIs coming to do your job is because of who is being targetted for replacement now.
A small history lesson follows, when weaving was automated, weavers rebelled but in the end their jobs were automated and simplified , but the great and the good didnt care(that much) then as they were the ones making more money from weaving machines, roll the scroll of history forward to my lifetime, CNC/robotics were coming in, first in mass production resulting in the great manufacturing decline of the 1980's(not helped by manglement/unions/thatcher), but scroll forward to now... we have 7 cells for mass production utilising 8 people, go back 40 years and that work would have needed 25-30 people, but again beyond a few mumbled sympathies and platitudes for those losing their jobs , not much happened from the great and good.
Now AI is here and its aimed squarely at the high paying jobs that the great and good have always kept to themselves and refused to automate for whatever reason.
And now they are squealing like a stick piglet how unfair it is that their jobs are now under threat from automation when nearly everyone else across the economy has always been under threat of having their job automated away from them.
But there is a glimmer of hope.... when everyone is out of a job and trying to get by on income support.... who exactly are going to be buying the products of our 99% automated economy?
Yawn.
"However, AI systems that could cause catastrophic outcomes do not need to be AGIs", followed by some examples of using AI to develop something dangerous.
I find this rather lame. You could say the same about "science" or "mathematics". We've faced these kinds of existential threat for ten thousand years and, er, we're still here. This is because none of these risks are self-propelled. They need a sufficiently large number of idiots to make them happen and, so far, we've avoided assembling a critical mass of such idiots. Yes, we do need to keep on doing that, but we've been working on that problem for a long time. There's nothing specifically "AI" about it.