You should be worried about AI, but not like that
One of my friends asked me this morning to talk her out of her concerns over AI, after seeing a bunch of articles about it online. My response ended up being nearly 700 words and was fairly generic, so I figured I may as well post a slightly edited version of it here, as well.
===
Honestly, there are plenty of things related to AI that are worth worrying about, but none of them are the things that get brought up in "AI apocalypse" discussions, and none of them are new, just things that were already problems that AI is making worse.
The OpenAI agent-led hacks that happened this year are like Chernobyl - too many people view them as an issue with the core technology, when they're easily predictable and preventable events that were only able to occur due to systemic issues in how the organizations in charge operated and technical issues with the safeguards in place around the core technology. OpenAI has been embarrassingly bad at cybersecurity; any even remotely competent team would've immediately been able to spot the agents behaving maliciously, and once discovered, due to how inefficient LLMs are, there's no risk of them going rogue in the wild like Ultron. And the need to have clear guidance and guardrails in place due to the potential for agent misbehavior stemming from overzealous attempts to fulfill a request are a long-known issue (https://en.wikipedia.org/wiki/Instrumental_convergence). The primary thing that's leading to events like the OpenAI hacks occurring is the lack of legal consequences, and I don't think that's going to continue being the case for very long. Holding companies legally responsible for the things that their agents do is such a simple, easy law to pass that I expect it to come around pretty soon.
The other important thing to remember with people stoking concerns about AI is that OpenAI and Anthropic are both eyeing IPOs in the next little while, and making people afraid of how powerful their models are is a great way to make people think they're way more capable and able to accomplish things than they actually are. It should be viewed similar to when some edgy, upcoming band or comedian intentionally tries to drive outrage over their next release, because any press is good press and will lead to better sales and people thinking they have more influence than they actually do.
The things to worry about in relation to LLMs are just the same things that were a problem before LLMs came around. Spreading misinformation is orders of magnitude easier now than it used to be; instead of having to hire a bunch of third-world workers to control sock puppet accounts on Facebook, you can have agents do it for a lot cheaper. Instead of having to pay PR/SEO firms to hide bad press, you can just flood Google with slop to make it less likely that people will see the bad stuff about you, and less likely that they'll believe it if they do see it. Kids are learning less in school because of being able to use ChatGPT to complete assignments, but the start of the downward trend in educational results in the US wasn't ChatGPT, or even COVID lockdowns, but was the spike in in-classroom computer use during the 20teens. And the impacts that LLMs are having/will have on the economy, everything from spiking microchip and electricity prices to the broader social effects of increasing income inequality, have the potential to be dire.
However, none of those are new problems. They're all ones that we as a country and a species have survived before, and the primary ways to fight against them are the sorts of things that will help fight other, non-AI-related problems as well. In-person organizing with your neighbors, getting involved in local politics, ensuring that you're prepared to weather minor disaster scenarios (and by that I mean "the power is out for a week so you should have a bit of food and water stashed away and know which of your elderly/infirm/whatever neighbors may need special assistance" and not "civil war 2.0 so you need a fallout bunker and a bunch of guns").