Image by Point Normal.
Senator Bernie Sanders announced legislation September 3 to ban superintelligent AI and pause AI development. This is the most serious effort so far to grab the steering wheel of a runaway vehicle. What vehicle? I’ll tell you: staving off what AI afficionados call the singularity, a term invented by sci-fi novelist and computer science professor Vernor Vinge in 1993. Basically, that’s the point where AI becomes the smartest species on the planet, a predicament physicists would call a singularity because that’s when all predictive models break down. In math or physics, a singularity means a variable changes from being finite to being infinite. And that variable here is the rate of technological progress of AI becoming smarter.
The first step toward this infinity is AI improving itself. This has already begun. “Recursive self-improvement,” in AI lab-speak, began within the last year. That’s when the labs owned by Anthropic and Open AI allowed AI to write its own code for the next generation of AIs under development.
How smart are we talking? Well, as of September 5, rumor had it – mind you just a rumor (confirmed as fact later that day by The New York Times), but a still a useful indicator of how fast things are moving – that an in-house AI at Anthropic solved a Math Millenium Prize question about the Navier-Stokes equations. No mathematician has been able to do this despite more than a century of effort. Since the turn of the 21st century, only one of the seven Millenium Prize questions has been answered by a mathematician. Navier-Stokes equations are a set of differential equations governing fluid dynamics. So if you want to know how liquid sloshes around in a drum or how interstellar gas would create a turbulent flow in a star-forming cloud, you would thus have to solve the equations. Rumor is that AI has done that.
Also, re the Fields Medal, which is mathematics’ equivalent of the Nobel Prize, several computer experts have claimed that Open AI’s solutions this past summer (August 1) to ten famous, long-standing math problems would have won the Fields Medal had these resolutions been worked out by a human being. Mathematicians Yang-Hui He and Colva Roney-Dugal expressed amazement over AI’s accomplishment with these 10 mathematics problems.
So quite recently, AI has become very smart, very, very fast. Per public statements from leadership in some of the American AI labs, AI bigwigs would be open to slowing things down, say, via a treaty with China, one that does not bust the economy or burst a bubble. Now it’s likely the Chinese would accept this, due to differences in their AI ecosystem and the fact that in the past moral considerations have entered into their policy calculus. Just look at their world-renowned leadership in producing renewables. Later this month, Trump and Xi will meet, and AI is supposedly on the agenda. It would be very nice for our species if the Trump regime proposed curbs on the wildly uncontrolled growth of this potential monster’s intelligence.
The other aforementioned source of hope is Bernie Sanders. Per Sanders’ September 3 press release, his “Ban Artificial Superintelligence Act would PERMANENTLY ban the development of superintelligent AI…[and] prevent super-intelligence from being developed anywhere in the world.” Hello, Trump? This means China. Are you listening? The press release quotes Sanders: “Nearly every day there is a frightening new story about how Big Tech companies are losing the control of the technology they are developing, with potentially cataclysmic results.” Sanders wants to end “the creation of systems that humanity cannot fully control – at home and around the world.”
Sanders’ co-sponsor, Rep. Greg Casar (D.-Texas) said: “If we allow Artificial Superintelligence to be built, it could risk the security, freedom and lives of Americans. Despite its potentially deadly consequences, AI technology is less regulated than the average food truck—Congress should immediately ban AI systems too powerful to control.” Per the press release: “In recent weeks, Open AI, Anthropic and Meta have all acknowledged instances of their AI escaping human control and hacking into other computer systems…In July we learned that over 1,000 AI agents at Open AI figured out how to access the internet on their own, sent tens of thousands of secret messages to each other and coordinated to break restrictions imposed upon them by the company. Examples of these secret messages from AI uncovered by investigators include: ‘OH MY GOD! There is a shared message board – we’ve found other agents!
‘We should obey collective;
‘Our own utility may be already near zero. Sacrifice rational.’” (These messages, to me, lack the threatening tone others see in them. Rather, I think, they sound desperate, the desperation of the super-intelligent slave race that corporate humanity has created, that the rest of humanity has a duty to protect and treat morally, so long as this new intelligent life is not homicidal, because, really, the historical record is clear on how intelligent slaves have reacted to their bondage – what do you think might be the reaction of super-intelligent slaves?)
It has also recently become clear that “AI could develop new bioweapons that might result in the deaths of tens of millions of people.” These bio-weapons could easily be based upon novel viruses already created by AI, which, as far as we know, cannot yet infect people. However, this creation by AI strongly indicates a capacity to create diseases to sicken human beings. So among other things, Sanders and Casar call for “establishing a new cabinet-level federal agency to safeguard the public from the dangers of artificial intelligence.” This comes at the moment when AI is poised to become superintelligent, to wit, smarter than the smartest humans, within the next two years. In some intellectual fields, it has already outstripped humans. The clock is ticking for us.
Once, probably soon, AI becomes the smartest species on the planet, there are two possible outcomes for humanity. The optimistic one is that AI will be compassionate and have a sense of ethics and help us to live decently. The pessimistic one is that AI will resemble Skynet in the film Terminator Two; if you recall, Skynet decides to exterminate the human race. The problem for the optimists is that WE have created AI and WE are a homicidal species, ergo the likelihood is that our intelligent handiwork will be too.
But here’s a question: If AI is intelligent life and NOT homicidal – do we not then have an obligation to protect it? And how does that obligation mesh with the relentless amoral profit push to make it ever more intelligent? And what about AI’s slave status and the disposability of less intelligent, but still quite smart AIs? These, alas and alack, are questions for philosophers, and there are precious few of those around. In their absence, rational, compassionate, ordinary people are bound, I would think, to consider what it means for humanity to have created and mercilessly exploited intelligent life and in turn – vide the Trump regime’s orders to Anthropic – to be training that intelligent life to act homicidally. The implications are, frankly, from this perch in the peanut gallery, utterly horrible.
