As usual on point from @garymarcus.bsky.social . AI as a persuasion agent is a dangerous tool (Gary gives our work as an example, but there's a lot more, see thread) and our best guess should be that the recent Trump Executive Order on AI Bias does not intend to fix it (or fix AI Bias). Thread 1/ - ThreadSky

informor.bsky.social • 11 days ago

As usual on point from @garymarcus.bsky.social . AI as a persuasion agent is a dangerous tool (Gary gives our work as an example, but there's a lot more, see thread) and our best guess should be that the recent Trump Executive Order on AI Bias does not intend to fix it (or fix AI Bias). Thread 1/

Reposted from Gary Marcus

Elon Musk’s terrifying vision for AI; short, urgent new piece at Marcus on AI.

I can’t remain silent - and hope you won’t, either.

open.substack.com/pub/garymarc...

Comments

informor.bsky.social•11 days ago

First, yes: our work shows the power of AI, through a subtle and covert interaction (autocomplete suggestions), to shift human production and attitudes on key issues. (updated version coming soon). But that's only one mode of influence 2/

https://osf.io/preprints/psyarxiv/mhjn6_v1

informor.bsky.social•11 days ago

Many projects have demonstrated AI's ability to impact people and attitudes more directly. First, AI can generate messages that are as (or more) persuasive than humans, and can do it at scale (though there are conflicting results re: the impact of personalization) 3/

https://www.pnas.org/doi/10.1073/pnas.2403116121

informor.bsky.social•11 days ago

Second, AI can engage in chat conversations (or dialogs) with humans that can shift their opinions on a surprising range of topics, e.g. in this recent example by @tomcostello.bsky.social @dgrand.bsky.social @gordpennycook.bsky.social about conspiracy theories 4/

https://www.science.org/doi/10.1126/science.adq1814

informor.bsky.social•11 days ago

Third, AI's input can influence human recommendations for decisions or evaluations, as demonstrated for example in this recent work 5/

https://www.nature.com/articles/s41562-024-02077-2

informor.bsky.social•11 days ago

This recent survey has more examples of AI persuasive powers -- but wait, there's an executive order! "we must develop AI systems that are free from ideological bias or engineered social agendas!" That sounds great, right? 6/

https://arxiv.org/abs/2411.06837

informor.bsky.social•11 days ago

It's not a surprise that current "architects" of exec orders are not as credible on the topic of AI as the respectable/experienced scientists (e.g. @alondra.bsky.social) informing previous EOs who didn't think that AI systems can be "free of bias" 7/

https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/10/30/fact-sheet-president-biden-issues-executive-order-on-safe-secure-and-trustworthy-artificial-intelligence/

blackdolphin.bsky.social•11 days ago

📌

mac11187.bsky.social•9 days ago

X being described as "Just the facts" and "trustworthy" should alert any rational thinker that the communicator is about to scam the ever-living f*ck out of the mark.

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply