There's a new report making the rounds, from Georgetown's Center for Emerging Technology, on AI and disinformation.
Essentially the takeaway is that human-machine teams can generate fairly effective disinformation right now, BUT it's computationally & financially expensive...
Essentially the takeaway is that human-machine teams can generate fairly effective disinformation right now, BUT it's computationally & financially expensive...
So *for now* all we have to worry about is state adversaries using GPT-3 or something similar to spread disinfo.
Or very wealthy people who can just blow $200K on creating ONE system, or tens of millions per year to generate content equal to, say, 1% of all activity on Twitter.
Or very wealthy people who can just blow $200K on creating ONE system, or tens of millions per year to generate content equal to, say, 1% of all activity on Twitter.
Except... I think the report is wrong about it being confined to state actors for now. *Political parties* already spend a lot of money on their digital operations. Why not more for this?
Maybe they don't think the cost-benefit is there *yet*, but remember:
Maybe they don't think the cost-benefit is there *yet*, but remember:
computing power gets cheaper all the time, so at some point it WILL cross that threshold.
And eventually, one presumes, it will be cheap enough for individuals to do it on a whim. 🙃
Hopefully by then we'll have technical solutions to filter out GPT-3-generated content, but:😬
And eventually, one presumes, it will be cheap enough for individuals to do it on a whim. 🙃
Hopefully by then we'll have technical solutions to filter out GPT-3-generated content, but:😬
Oh, right! I should link to it. Summary is here, full PDF is at the bottom.
cset.georgetown.edu/publication/tr…
cset.georgetown.edu/publication/tr…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
