I'm glad we don't sell uranium next to the guns at Walmart.
My capitalist reflex is to reject new regulations, but slowing down AI is starting to sound like basic common sense.
Yes, humans have a tendency to overestimate existential risk. It's how we've made it this far. Still, I think yesterday’s article from the CEO of Anthropic (Dario Amodei) asking for AI regulation will be remembered as a turning point in the AI race.
That being said, if you're an investor trained to look at everything through an incentives filter, (cough: me) it's hard to read without skepticism given how biased Anthropic is on this topic.
They have every incentive to push hard for AI regulation.
Why? AI models don't look like a defensible business (no real moat).
- Limited switching costs
- Insatiable need for capital
- Marginal cost of tokens plummeting
- Lots of killer competitors & free options (Open Source)
- And extreme litigation risk, or much worse, if they make a mistake
Despite insane growth, it's a fragile business model. Intelligence is becoming a commodity.
So OF COURSE they want the government to step in and put up barriers.
"Please, come regulate the heck out of me AND all these brilliant people trying to knife me in the back."
They’re also about to launch a massive IPO where they’ll be telling the market their tech is “so powerful” it could gobble up every business on earth (like SpaceX, their S-1 filing will probably say they have a $30+ Trillion total addressable market).
Credit: @EvanKirstel
But the motives are hard to untangle.
Because continuing to breathe is a pretty decent incentive too.
It was easy to discount the sky is falling cries from the twenty-something Anthropic employees this week - many of whom just need to emerge from their basements and touch grass.
Anyways, despite their panic, there is a WHOLE lot of space between AI agents doing something dangerous (like hacking a bank) and human extinction.
I believe these Gen Z engineers are too young to have internalized that:
- most predictions are worthless
- everything takes longer than you'd think
- "safety" is an illusion; risk is an inescapable part of life
- and people are really good at course correction when the chips are down
But I also don't think Dario writes this article (on the cusp of a public offering) as some sort of selfish media psyop. It's probable he believes this is a massive problem.
Now, I don't think anyone wants to give Anthropic or OpenAI the only keys to AGI.
But it's not like Google, Meta, SpaceX, etc. are light on resources to catch up. They're also good at dealing with regulations.
Call me crazy...but giving unregulated access to open-source models to bad actors (namely religious extremists that prefer we stop existing) seems like a dumb move.
In other words, you don't have to fear the doomsday Terminator / Matrix scenario to think AI pacing is a good idea.
This is tricky as I'm for small government and we usually over-regulate things to death. But some things are important enough to draw lines in the sand.
Which is appropriate given AI literally turns sand into intelligence.
Also - considering where the country stands on data centers (not fans), government intervention seems likely at this point.
Don’t get me wrong, I’ll take the over on our collective demise. I strongly believe that fear is not your friend and good times are ahead.
But the people that love to “move fast and break things” are begging for oversight. We should probably listen.
This is difficult as I trust tech CEOs (especially the ones pushing mind-numbing social media on America’s youth) about as far as I can throw them. Which isn’t the right metaphor as I could probably throw some of these guys a decent distance.
But given the exponential growth curve of this unprecedented technology - a concerted effort to tap the brakes seem about right.
Brad Johnson