Tech leaders urge pause in 'out-of-control' AI race

Sounds like two camps among the signee

1- Tool will be abused for the worsening of society very quickly
2- AI will wake up and destroy us all

After GPT-4, tech leaders urge a 6-month pause in the AI race : NPR

Are tech companies moving too fast in rolling out powerful artificial intelligence technology that could one day outsmart humans?

That's the conclusion of a group of prominent computer scientists and other tech industry notables such as Elon Musk and Apple co-founder Steve Wozniak who are calling for a 6-month pause to consider the risks.

Their petition published Wednesday is a response to San Francisco startup OpenAI's recent release of GPT-4, a more advanced successor to its widely used AI chatbot ChatGPT that helped spark a race among tech giants Microsoft and Google to unveil similar applications.

What do they say?
The letter warns that AI systems with "human-competitive intelligence can pose profound risks to society and humanity" — from flooding the internet with disinformation and automating away jobs to more catastrophic future risks out of the realms of science fiction.

It says "recent months have seen AI labs locked in an out-of-control race to develop and deploy ever more powerful digital minds that no one – not even their creators – can understand, predict, or reliably control."

"We call on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4," the letter says. "This pause should be public and verifiable, and include all key actors. If such a pause cannot be enacted quickly, governments should step in and institute a moratorium."

A number of governments are already working to regulate high-risk AI tools. The United Kingdom released a paper Wednesday outlining its approach, which it said "will avoid heavy-handed legislation which could stifle innovation." Lawmakers in the 27-nation European Union have been negotiating passage of sweeping AI rules.

Who signed it?
The petition was organized by the nonprofit Future of Life Institute, which says confirmed signatories include the Turing Award-winning AI pioneer Yoshua Bengio and other leading AI researchers such as Stuart Russell and Gary Marcus. Others who joined include Wozniak, former U.S. presidential candidate Andrew Yang and Rachel Bronson, president of the Bulletin of the Atomic Scientists, a science-oriented advocacy group known for its warnings against humanity-ending nuclear war.

Musk, who runs Tesla, Twitter and SpaceX and was an OpenAI co-founder and early investor, has long expressed concerns about AI's existential risks. A more surprising inclusion is Emad Mostaque, CEO of Stability AI, maker of the AI image generator Stable Diffusion that partners with Amazon and competes with OpenAI's similar generator known as DALL-E.

What's the response?
OpenAI, Microsoft and Google didn't respond to requests for comment Wednesday, but the letter already has plenty of skeptics.

"A pause is a good idea, but the letter is vague and doesn't take the regulatory problems seriously," says James Grimmelmann, a Cornell University professor of digital and information law. "It is also deeply hypocritical for Elon Musk to sign on given how hard Tesla has fought against accountability for the defective AI in its self-driving cars."

Is this AI hysteria?
While the letter raises the specter of nefarious AI far more intelligent than what actually exists, it's not "superhuman" AI that some who signed on are worried about. While impressive, a tool such as ChatGPT is simply a text generator that makes predictions about what words would answer the prompt it was given based on what it's learned from ingesting huge troves of written works.

Gary Marcus, a New York University professor emeritus who signed the letter, said in a blog post that he disagrees with others who are worried about the near-term prospect of intelligent machines so smart they can self-improve themselves beyond humanity's control. What he's more worried about is "mediocre AI" that's widely deployed, including by criminals or terrorists to trick people or spread dangerous misinformation.

"Current technology already poses enormous risks that we are ill-prepared for," Marcus wrote. "With future technology, things could well get worse."

 

Isn't it pretty arrogant to think that humans can control something they don't understand? The very experts who came up with the latest AI design and architecture have said for years now that they do not understand it. It's literally a black box to them. They stack transformer on top of transformer, and out comes some magic (magic because we don't know how it happens). That's already happening today. If we don't understand it, how can we ever predict what it will do? The chance of it simply destroying humans (for whatever reason) cannot be ruled out.

 

No, that's not it. "Bad information" is already baked in. Bad actors will use their own version of ChatGPT4 and more advanced versions to do bad things. That's pretty much a given. This is an area that could be mitigated through self and state regulation and standardized ethics. Still, there is a long way to go to get there, but it's relatively easy to accomplish. There are AI ethics teams at all these companies, and some governments have already started working on regulations. We can also react fairly quickly once we realize how damaging it can be as a tool used by "bad humans".

Animals aren't smarter than us, so we aren't that concerned about them. The risk is with an entity that is much smarter than us that cannot be controlled. You can imagine it as an alien (just that it is created by humans). When an alien of superior intelligence shows up on Earth, can you guarantee it's nice to humans? Then why would you assume an AI of higher than human intelligence will be nice to humans? This isn't a new concept - AI scientists have talked about this since the 70s. But no one has taken it that seriously, probably simply because they think AI with that type of intelligence isn't going to be here for at least another 50 years, if ever. Now, it's not so clear.