New poll: trust in AI is sinking

[bleak flint]

Trust in AI companies dropped 75 to 50%.
Trust in AI from 50 to 35%.

Overall, the Edelman respondents indicated that they look towards scientists to inform them on AI safety, providing an opportunity for the research community to step up as authorities on the subject.

Good, because scientists are scared shitless. We need to push the AIImpacts 2023 survey stats harder.

https://www.axios.com/2024/03/05/ai-trust-problem-edelman
https://au.news.yahoo.com/public-rapidly-turning-against-ai-223456655.html?guccounter=1

Podcast about this: https://omny.fm/shows/geekwire-podcast/ais-trust-problem-richard-edelman-on-the-risk-to-t

[bright basin]

Amazing.

[weary tundra]

Would be great if we could find out why distrust in AI is increasing. Seems like PauseAI is leaning mostly on "loss of control -> existential catastrophe", but maybe we'd find strong allies in people with other concerns like:

• Job loss
• Concentration of power
• Militarization of AI
• Enablement of crime (esp. cybercrime)
• Social media addiction and attendant mental health consequences
• Misinformation and undermining Democratic norms
• Copyright infringement

[bleak flint]

My completely unscientific analysis says people are mostly worried about:

1. Job loss. Your job will be taken over by a robot. You will become poor!
2. Privacy. AI reads your data and draws conclusions about your dark secrets. Other people will find out!
3. Wrongful accusations. You'll end up in jail for a crime you didn't commit. Some fake video convinced the judge!
4. People jerking off. Someone makes a deepfake of you or your daughter, yuck!
5. Hackers. Hackers hacking the cybergrid, stealing your money and identity, maybe even preventing you from your money!
6. Fake media / democracy polarization / loss of trust. The party that you dislike will gain power because of this. Bad policies will follow that make you poor!

Kinda cynical, but I do notice these trends. Most people are obsessed with personal / small scale risks and aren't that sensitive to the big catastrophic ones.

[winter lichen]

So we should talk more about these small risks?

[bleak flint]

Campaigns like these that focus on current narrow AI risks are effective in making anti-AI sentiment:

https://www.youtube.com/watch?v=zpoIrkLvEwI (someone steals your money!)
https://www.youtube.com/watch?v=Ur1ewODfCuE (someone jerks off to your daughter!)

But I don't think they will lead to the types of policy measures that we need to survive. I'm afraid they lead to national level usage bans, instead of international compute regulation and training run bans.

[weary tundra]

I'm not sure. I can see people being supportive of international compute regulations/training run bans for non-existential reasons. e.g., a US citizen's job won't be saved by pausing AI only in the US when we can just outsource the job to a Chinese AI.

In general, I'm worried that if we narrowly focus on loss of control -> human extinction, we'll miss out on a large segment of the population. I think one of the most likely ways we are to fail is not reaching a broad enough audience.

Plus, this could be a good gateway to raise awareness about loss of control. "Hey, you're worried about job loss because AI is getting super powerful? Me too! Also, because AI is getting super powerful, I'm worried we might be unable to control it, which could have other disastrous consequences for humanity."