Ok, we've all heard it. AI will do everything. We'll be living in a world of AI assistants and agents, and it's already begun.
Although there's been an abundance of AI safety reports, especially from non-profit-cum-for-profit OpenAI—as well as a slew of organizations, committees and conferences knee-deep in evaluating the social, economic and ethical implications—the perspective of futurist Anders Sandberg warrants consideration:
"Thinking AI safety is about stopping AI development is like thinking climate change mitigation is about stopping modern industrial society. Both are unworkable strategies and not really what serious actors in the respective fields are considering and doing."
Eric Schmidt, undoubtedly one of today's most serious actors in AI, recently spoke at Princeton University. During Q&A, Schmidt, former CEO and Chairman of Google, responded to a student saying:
"This is going to happen very fast regardless of what you and I agree to because there's too much money, too much scale, too many competitors."
When asked by another student about the best AI career path Schmidt responded:
"If somebody could come along with the next big thing after Transformers that would be like a big deal. How do [agents] go to the equivalent of tree search to look at all their different choices to figure out what they're going to do? That's really important and a really hard problem, I would work there."
It could be said that 2024 was the rise of the AI agent economy.
In August, Skyfire.xyz released its blockchain payment network that allows AI agents to spend and receive money. Stripe released its AI agent toolkit, and Coinbase introduced "Based AI Agents" allowing users to set up AI agents with crypto wallets and onchain access. Then there's Truth Terminal, the crypto trading bot that became a crypto millionaire for its stake in meme token Fartcoin, which although sounds laughable, is reported to have an $8 million market cap according to media outlet, The Block.
The idea that AI agents will pay our bills, cook for us and wash for us, as Schmidt explains in the trailer for his book, Genesis: Artificial Intelligence, Hope and the Human Spirit, gets more complicated when "there are millions of agents that are written by different people, and they start working together and speaking in a language that we don't understand."
So how can we control AIs? At Day 2 of the Beneficial AGI Summit, science fiction author David Brin suggests that anchoring high-level AIs to the physical world with a pingable ID could be an answer, and/or rewarding “good” AIs to catch malign AIs. The rewards? Things like autonomous memory space that AIs need to reproduce.
But if AIs do get out of hand, Eric Schmidt has the solution: "Unplug them. Unplug them immediately."
Sources:
https://www.reddit.com/r/ChatGPT/comments/1fov6mt/i_think_im_working_for_an_ai/?rdt=53387
https://www.npr.org/2024/11/22/nx-s1-5199766/eric-schmidt-interview-the-justice-department-google-ai
https://www.youtube.com/watch?v=OT_S4g5G5N4
https://techcrunch.com/2024/08/21/skyfire-lets-ai-agents-spend-your-money/
https://docs.cdp.coinbase.com/learn/docs/based-ai-agents
https://docs.stripe.com/agents
https://www.youtube.com/watch?v=xFz80aaGeQs
https://www.youtube.com/watch?v=YEbkMOq2jPw
https://www.youtube.com/watch?v=OT_S4g5G5N4
https://informedi.org/2024/05/09/ai-is-helping-automate-one-of-the-worlds-most-gruesome-jobs/
https://techcrunch.com/2024/08/21/skyfire-lets-ai-agents-spend-your-money/
https://docs.cdp.coinbase.com/learn/docs/based-ai-agents
https://docs.stripe.com/agents
https://www.ccn.com/education/crypto/what-is-truth-terminal/