The Authoritarian Mind: Gun Control ‘Logic’ Fails Just as Badly When It’s Applied to Artificial Intelligence

As a student in an emergency management program (I’m cramming my two-year master’s degree into like 14 years), I’m finding discussions of artificial intelligence safety to be pretty interesting. Some people have great ideas, while other people don’t. But, one thing I didn’t expect to see was a group trying to apply anti-gun logic to AI.

After looking over their position and thinking about the wider situation, I found something that readers here would probably be interested to see. In short, anti-gun logic fails when applied to other situations too, and for pretty much the same reasons.

Before I can discuss why their logic predictably fails, I need to make sure you’re speed on the AI safety debate in as few words as possible. The fear is that a future (perhaps a very near future) AI program will be more intelligent than all people at every task. Such a superintelligence could cause humanity serious problems if its goals aren’t aligned with ours. For example, if we told that kind of AI to maximize the number of paper clips made, it could, for instance, go out of control and try to make paper clips out of the iron in our bloodstreams when it runs out of other sources.

In other words, AI doesn’t have to be evil to be a problem. Poorly thought out goals that our existence runs afoul of could be enough to get us all killed. At least theoretically.

Some people who are afraid of AI going full Terminator on us are asking for government regulation that imposes AI safety standards on companies that are developing these programs. But one group I came across recently is going all-in on government power and is calling for a complete ban on artificial superintelligence.

Specifically, Control/AI (a more technical and cool way of saying “AI control”?) is calling for the following policies:

• Banning the deliberate development of superintelligence

• Prohibiting dangerous AI capabilities and superintelligence precursors such as automated AI research and hacking

• Requiring companies to demonstrate that an AI will not use forbidden capabilities before they run it

• Creation of a licensing system for advanced AI development

As I’ve explained elsewhere, government regulation of AI companies isn’t a fix-all for this kind of risk. Such an approach might work to slow down or end AI development for a short time because the massive data centers needed to create and run AI programs are physically very big and use a lot of electricity. Running a rogue AI operation would be about as difficult as running a large illegal gun factory or a big marijuana growhouse. It’s just hard to hide something so big and power-hungry for very long.

But, just as 3D-printed firearms have already proven, going after the big players can’t keep a technology under wraps forever. The cost of computing power and the amount of electricity it requires has dropped precipitously over the last 100 years. The first calculator was as big as half of a city block, while today it’s an app in your phone that’s infinitely more powerful, takes up virtually no space and uses almost no electricity.

Banning or heavily regulating advanced AI development won’t be possible once it becomes feasible to do that kind of research in mom’s basement. The technology will be developed by somebody. It’s just a matter of how much time we have to prepare.

Just as with guns, the fact that bans and heavy-handed regulation are ineffective doesn’t mean we have no other options for protecting the public from irresponsible actors. There are plenty of ways to limit the risks of rogue AI. Options include:

• Prohibiting the use of AI programs to control critical infrastructure (dams, power plants, refineries, pipelines, etc.)
• Limiting the amount of resources an AI program can have access to
• Prohibiting AI from controlling dangerous chemicals, explosives, and weapons in most situations
• Increasing preparedness for the misuse of AI (hackers, DIY builders of killer robots, etc.)

One thing is for sure. Failed anti-gun logic, with its outright bans, rigged permitting and licensing schemes, and other ludicrous control strategies that rely on government power, doesn’t work any better when it’s applied to things other than guns. Flawed authoritarian approaches will be flawed no matter where it’s applied.