lukifer

My friend's grade-school kid gets exuberantly excited by CRTs and VCRs, and I know a few Gen-Z's with record collections. Maybe there's truth to the Dennis Duffy joke on 30 Rock: "technology is cyclical" :)

If that was the move, you'd expect they'd be actively fear-mongering against open models specifically: not just their training, but their usage and deployment. They'd want to paint open models as dangerous weapons for terrorists, etc, and that only the big firms are capable of responsible engineering and "guardrails".

Instead, we've seen the opposite: all the big scary stuff has been from their own models (in a few notable cases, also demonstrating that they're actually quite irresponsible in their monitoring and security).

"We need to outlaw single-prop planes, they're a risk to the public: only Boeing has the ability to produce planes safely. How do we know single-prop planes are dangerous? Because Boeing planes keep falling out the sky."

If that were the intent, from the OAI/Anthropic perspective, you'd think we would have seen at least one of the scary incidents (either real or simulated) being performed by Qwen or DeepSeek, rather than from their own models.

If anything, their public messaging has been to entirely pretend that open models don't exist, that they're mere toys or curiosities, that only the big dogs with eleven zeros of capex have capabilities which could be considered dangerous.

You could certainly argue that the various "regulate AIs like airplanes" bills might shut down some open model development (domestically; they're certainly not calling for international weapons treaties). But the big firms haven't exactly been jumping backflips in support of Sanders' Ban Artificial Superintelligence Act.

Flight is a testable characteristic (at least within certain bounds: a chicken can only fly a little, but it certainly flies more than a worm or a penguin).

What test would you propose where an LLM (or an RL agent containing an LLM) would reliably fail, but where an average human would reliably succeed?

I haven't dug into these claims in detail; I take seriously a possible relationship between quantum mechanics and the brain. But "free will" is hardly a settled question either (I'm closer to the "compatibilist" position, but Sapolsky's "Determined" makes a robust case).

And even if quantum physics are essential to human brains, it's not clear that introducing dice transcends determinism into free will, as opposed to simply adding unpredictability. ("The physics made me do it" -> "the dice made me do it") Let's not forget, there's a significant amount of randomness in AIs as well ("temperature"). And sure, it's simulated algorithmic randomness; but does that imply, if we somehow wired every `rand()` call into radioactive atom decay, that means the AI "wakes up"?

And all that is besides the point: I don't see any reason to assume "intelligence" requires "free will". I'm entirely capable of conceiving, in the abstract, a being which is both intelligent, and deterministic. You still haven't given me a definition (or better yet, a test), to distinguish a "real intelligence" from unintelligent mechanism.

Run the same thought experiment the other direction: at a low level, the human brain is just physics and chemical mechanisms, which in principle could be perfectly imitated by transistors calculating the same output. (Yes, such a thing isn't realistic, but neither is billions of humans doing perfect matrix math by hand.)

I still haven't seen a robust definition of "intelligence" that would allow me to tell the difference. (Bear in mind, this is a definitional struggle even in biology: if we were walk back the evolutionary chain from a human, to a protozoan, it's not clear that you could pick a single point where a non-intelligent creature gave birth to an intelligent one. It's a Paradox of the Heap.)

There's a sense in which humans were created... by other humans. :) Until you go far back enough in the evolutionary chain, when the creators were our primate ancestors.

Douglas Adams had a yarn about evolution, about a puddle that wakes up, and declares that the hole in which it sits must have been perfectly designed for it by its Creator. But of course for a puddle to exist, it must perfectly mirror its environment. It makes no sense for a puddle to not fit its hole. Emergent complexity has the same characteristic: it's inseparable from the environmental pressures which led to it. Two sides of one coin.

It's a deep rabbit hole, but there is also a sense in which we co-evolved with memeplexes, biological and informational life forms, each shaping and adapting to the other. To the extent our nervous systems act as a substrate for memetic evolution, perhaps LLMs offer memetic "life" a new evolutionary environment.

I've yet to encounter a robust definition of intelligence which would rule in every human, while ruling out every form of existing AI (noting that we're talking about agents and "reasoning models", where the LLM itself is a component in a larger system). If you can offer such a definition, I'd be eager to hear it.

I personally prefer a practical, behaviorist definition: a feedback loop capable of prediction, modeling, and steering, towards arbitrary goal states. That makes it clear that we're merely talking about degrees of sophistication and capability, rather than a magical leap where mindless mechanism stops, and "real intelligence" begins.

Now taking Polymarket bets on when we'll first see an AI agent run for office. Voters have every right to be cynical at the moment; if the AI adopts a charismatic enough video persona, I could see some of the public going for it merely because it's some kind of shake-up to status quo.

Given that such a thing makes no sense legally (as of now), it would probably be done with the centaur model: a human meat proxy who pledges to follow the AI's governance advice.

Humans seem to struggle conceptually with the liminal space between selection-pressure automata (an RNA virus; an insect which evolved camouflage), and an evolved complexity with agency, capable of "understanding" its actions (certainly humans; arguably many animals). And this makes a certain sense: we evolved to treat agentic things as being categorically different from non-agentic things. If we see sudden movement, it's vital to rapidly assess the difference between a tree branch in the wind, versus a predator.

The evolved complexity of the corporation seems to fit within that middle space: more sophisticated than a stick-bug (which exists merely from non-stick bugs being eaten), but not quite to the point where OpenAI/Anthrophic/Google/etc can "understand" its actions. And yet those quasi-intelligent feedback loops, evolving from iterated selection pressures of markets and ROI, seem to already be sufficient to domesticate us in their own interests, piggybacking on the nervous systems of employees, investors, customers, and citizens.

One of the dilemmas of trying to communicate the full spectrum of AI Risk, is trying not to insult the intelligence of the human animal in the process. And don't get me wrong: what human wetware can accomplish with 20 watts is the most miraculous thing in the known universe. And yet how many of us can have our cognitive sovereignty one-shotted by engagement algos, Skinner boxes, gameplay loops, propaganda, advertising, flattery, social conformity, bias, fantasy, charismatic demagoguery, or straight-up bullshit?

If we grant that we are on track to make something smarter than humans (I think so): it's almost a face-saving white lie to spin yarns about a Skynet nuclear apocalypse, or a 7D chess move to mass-assemble a nanovirus with 100% lethality without anybody noticing. I do think those scenarios are worth taking seriously; but what's harder to communicate, is just how effectively a superhuman AI (or a diverse ecology of agent swarms) might be able to manipulate human behavior. It's something few of us are able or willing to truly process (not least because how many of us live in denial of how much our nervous systems are already hacked by technomodernity).

The appropriate analogy for what's to come may look less like the anthill carelessly demolished to make room for a highway, than the domesticated worker ants from Tchaikovsky's "Children of Time".