Gregory's ServerI did not sign this statement, tho I agree “open” AI is not the enemy of “safe” AI
I can't endorse its premise that “openness” alone will “mitigate current+future harms from AI,” nor that it’s an antidote to concentrated power in the AI industry 1/
Meredith Whittaker
6 comments
Meredith Whittaker
As coauthors & I show (paper👇), “open” AI can, in some forms, ensure transparency+ reusability+ extensibility. This is good. But it *does not* level the playing field in the concentrated AI industry: the resources needed to create/deploy AI remain in the hands of a few firms. 3/
Meredith Whittaker
Nor does openness necessarily reduce harm, in a context where the resources required to create/deploy AI at scale remain concentrated in the hands of a few corps. Indeed, as in the case of Meta, "open" AI (like FOSS beofre it) can be leveraged to extend market dominance. 4/
Meredith Whittaker
Again, "open" AI does good things. BUT any meaningful social benefit from “open” AI will only be possible if we ALSO SIMULTANEOUSLY reduce concentrated power of the AI industry. & any proposal that leaves this unsaid risks ramifying, not alleviating, current BigTech dominance. 5/
Meredith Whittaker
Paper referenced above, here: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4543807  |
This is esp true in an ecosystem where the term “open”, in the context of AI, has no clear definition, leaving it ripe for abuse + instrumentation by firms like Meta (who signed on + are currently brandishing this same statement to promo their ersatz "open" AI offerings). 2/