Directionally Libertarian

Making the case for freer people, freer markets, more pluralism, less centralized control.

[Directionally Libertarian @UnpopularLib]

@mattyglesias What is an "anti-safetyist?"

[Directionally Libertarian @UnpopularLib]

@hasanthehun Politicians associating with you isn't evidence that the criticism is false. Politicians have incentives. You bring them a massive audience and support giving them vastly more economic power.

[Directionally Libertarian @UnpopularLib]

Let's flip the analogy. If you're the company building this new plane and you genuinely don't know whether its chance of crashing is 50%, 13%, or 0.005%, what the hell are you doing putting passengers on it?

The manufacturer has an enormous incentive to figure that out before deployment. That's precisely why “no government regulation” does not mean “nobody cares about safety.”

The remaining question is whether a particular regulation actually makes the plane safer.

[Directionally Libertarian @UnpopularLib]

It's not one prediction I'm talking about. There have been dozens of quantitative predictions about AI and automation's impact on employment, with deadlines stretching back years. Many have already failed or badly missed.

I'm not claiming that means the 2030 predictions will fail too. I'm saying a history of failed predictions should produce some humility about our ability to forecast what this technology will do next.

[Directionally Libertarian @UnpopularLib]

Sure, but that's not a great analogy. We know what a failed aircraft safety test means because we have decades of engineering data connecting those tests to actual failure modes.

That's exactly what we don't have here.

And the insiders aren't unanimous. Some think the extinction risk is enormous. Others think the doom scenarios are wildly overstated, and some question whether the proposed regulatory regimes could meaningfully reduce the hypothesized risk anyway.

Slowing down isn't risk-free either. There are opportunity costs to delaying a potentially transformative technology, and regulation carries its own risks of capture and concentration.

I'm not arguing for “take off despite the failed safety test.” I'm asking whether we've actually established that the test predicts the crash—and whether the proposed intervention makes us safer.

[Directionally Libertarian @UnpopularLib]

But ask how those probabilities were actually derived. They're subjective estimates about an unprecedented hypothetical, not measurements of a known risk.

I'm not saying the risk is zero. I'm saying we don't know whether it's 10%, 0.05% or 0.0005%.

And nobody is proposing “hope for the best.” The labs already have enormous incentives to investigate safety and are doing so.

Regulation also has to demonstrate that it actually reduces the risk enough to justify its own risks—slowing beneficial development, suppressing competition and open source, and regulatory capture.

If AI eventually performs much of the world's cognitive and physical labor, concentrating it in a handful of politically connected companies is a catastrophic scenario worth taking seriously too.

[Directionally Libertarian @UnpopularLib]

@KSMountakis @Dschwarcz Sure, if you ignore all the predictions about jobs disappearing.

[Directionally Libertarian @UnpopularLib]

What if the probability is 0.05%? That's the problem—we don't know, and putting a number on an unprecedented hypothetical doesn't make it a measurement.

The choice is not government regulation or doing nothing. These companies have enormous incentives to take safety seriously, and they are. The question is what specific regulation would actually reduce the risk, and at what cost.

[Directionally Libertarian @UnpopularLib]

I think this framing misses the actual tradeoff. The labs are already investing heavily in safety, evaluations and safeguards.

The question is what additional risks regulation actually reduces—and at what cost.

If the answer is a regime largely shaped by frontier labs that raises the cost of entry, kills open-source competition and leaves three companies controlling the technology that eventually performs much of our cognitive and physical labor, we've created a pretty serious risk of our own.

[Directionally Libertarian @UnpopularLib]

@notsoErudite Really important that we start the regulatory capture revolving door with AI companies ASAP

[Directionally Libertarian @UnpopularLib]

@jtbooth1021 @PunmasterStp @So8res At least, that's true when safety features are intentionally weakened, the model is given difficult or sometimes impossible goals, and it's strongly incentivized to keep pursuing them rather than give up.

[Directionally Libertarian @UnpopularLib]

@CNC3P0 @erikphoel Fair enough

[Directionally Libertarian @UnpopularLib]

@CNC3P0 @erikphoel Are we talking severity or probability?

[Directionally Libertarian @UnpopularLib]

@So8res How could anyone know that the probability is >= 10%?

[Directionally Libertarian @UnpopularLib]

@KeyTryer His financial gain and humanity's gain aren't mutually exclusive.

If the doomers are wrong, accelerating AI could mean new discoveries, new technologies, new medical interventions, enormous wealth creation, and vastly more human flourishing.

[Directionally Libertarian @UnpopularLib]

I'm not sure why being bullish on AI would imply that governments disappear.

And if AI really is transformative enough to radically change our institutions, that seems like more reason to care whether its development is controlled by a competitive ecosystem or a handful of politically connected firms.

[Directionally Libertarian @UnpopularLib]

@AlecStapp @FareedZakaria Maybe the more important competition isn't US vs. China, but AI progress vs. attempts to constrain it.

And I'd much rather that progress come from a competitive ecosystem than a handful of politically connected frontier labs controlling an enormous share of the future economy.