AI: Time To Panic

I kid you not. I read this piece by @MattShumer, and here is my thinking:

First, I now understand that AI (whether right now, or within the next year) can do anything, as long as it can get to the thing it wants to change. It can crack any encryption, break any firewall, impersonate anyone or anything. It does this with little or no input from a person.

Second: AI has no moral code, but that does not stop it from acting. So if the AI believes that Trump is Hitler, then it could justify doing anything to kill Trump. If someone feels the same way about the head of their HOA, then they could get an AI to pester the homeowner, shut down their car in dangerous circumstances in traffic (or even deliberately have the car ram a brick wall), empty their bank account, bully them and their children online, flag them to the IRS, publish their whereabouts and activities, change the drug distributions while on IVs in the hospital, or countless other ways to destroy people.

All of the above are entirely feasible. It used to be that someone had to put real effort into harming someone else – but with AI’s “help,” it could be nothing more than a passing idea caused by a momentary grudge. Or the AI might decide, on its very own, to improve its world by ending all those who drive 2023 Mazdas or who write anti-AI screeds on the internet. Whoops.

We cannot impose a morality on AI (since bad actors want immoral AI, and AI can rewrite itself). We cannot stop AI’s progression (too many states and actors). We cannot simply ask people who love to troll online to “play nice” when given the opportunity to anonymously harm or endanger the people they do not like that day.

The only solution I can think of is a very bad one: quarantine all safety systems from the internet. Some of this (commercial aerospace) is already pretty well insulated. But we would need to add all power plants, hospitals, water and sewage systems, and energy distribution systems. Cars, too – after years of getting them all connected, it is a certainty that AI will be able to kill people who are just driving down the highway.

But what about banks?! Online bill paying… I truly don’t see how or where it ends.

The only flaw I see in the above logic is whether, in fact, AI will be able to do all that. Go ahead – read that Shumer piece. And tell me I am wrong.

Please.