NY Magazine’s Herrman: Conservatives Claim AI Chatbots Are Biased Because They Don’t Like the Answers

BRZEZINSKI That was Meta ░░░░░░░░ Mark Zuckerberg s online ░░░░░░░░ from January announcing Facebook ░░░░░░░░ removal of its fact ░░░░░░░░ system describing it as ░░░░░░░░ return to free expression ░░░░░░░░ Zuckerberg and Meta are ░░░░░░░░ about liberal bias in ░░░░░░░░ intelligence platforms namely Meta ░░░░░░░░ own large language model ░░░░░░░░ Llama According to our ░░░░░░░░ guest Meta is making ░░░░░░░░ argument that suggests the ░░░░░░░░ of training data available ░░░░░░░░ the internet leads to ░░░░░░░░ chat bots answering questions ░░░░░░░░ topics in a way ░░░░░░░░ has historically leaned left ░░░░░░░░ technology columnist for New ░░░░░░░░ magazine John Herrman joins ░░░░░░░░ now All right tell ░░░░░░░░ exactly how this is ░░░░░░░░ or they claim this ░░░░░░░░ happening HERRMAN Well What ░░░░░░░░ Zuckerberg is doing here ░░░░░░░░ what Meta is doing ░░░░░░░░ making the case that ░░░░░░░░ s bias in AI ░░░░░░░░ all these new chatbots ░░░░░░░░ been trained on basically ░░░░░░░░ whole internet They ve ░░░░░░░░ all these articles all ░░░░░░░░ academic literature all these ░░░░░░░░ and everything And what ░░░░░░░░ out of that is ░░░░░░░░ AI model that has ░░░░░░░░ all this stuff including ░░░░░░░░ that they say are ░░░░░░░░ leaning So you ve ░░░░░░░░ people asking chat bots ░░░░░░░░ controversial social issues about ░░░░░░░░ related stuff and they ░░░░░░░░ getting answers that at ░░░░░░░░ some of them are ░░░░░░░░ with And conservatives love ░░░░░░░░ attention to this You ░░░░░░░░ it about Covid you ░░░░░░░░ ask it about vaccines ░░░░░░░░ ask it about you ░░░░░░░░ the outcome of the ░░░░░░░░ election and you re ░░░░░░░░ to get consensus sort ░░░░░░░░ supported general stuff like ░░░░░░░░ Joe Biden won that ░░░░░░░░ or you know like ░░░░░░░░ vaccines tend to help ░░░░░░░░ these things And this ░░░░░░░░ become a hot button ░░░░░░░░ in the way that ░░░░░░░░ many decades media bias ░░░░░░░░ been you know a ░░░░░░░░ area of concern but ░░░░░░░░ politically useful the way ░░░░░░░░ social media bias as ░░░░░░░░ Zuckerberg is talking about ░░░░░░░░ become this really powerful ░░░░░░░░ you know as an ░░░░░░░░ for people like Donald ░░░░░░░░ to say Facebook is ░░░░░░░░ Twitter is censoring us ░░░░░░░░ like that Next up ░░░░░░░░ s AI The next ░░░░░░░░ thing is the next ░░░░░░░░ argument about bias BRZEZINSKI ░░░░░░░░ Molly Jong Fast So ░░░░░░░░ s a sort of ░░░░░░░░ adage that facts have ░░░░░░░░ liberal bias right Yeah ░░░░░░░░ to us about that ░░░░░░░░ that seems to be ░░░░░░░░ s happening here HERRMAN ░░░░░░░░ tools are almost like ░░░░░░░░ parody of that argument ░░░░░░░░ they the way these ░░░░░░░░ work is they pull ░░░░░░░░ truly anything these companies ░░░░░░░░ get their hands on ░░░░░░░░ of information just scraped ░░░░░░░░ the internet from other ░░░░░░░░ sources And yeah if ░░░░░░░░ scrape a bunch of ░░░░░░░░ quality content a lot ░░░░░░░░ that stuff is going ░░░░░░░░ be produced by you ░░░░░░░░ experts who are under ░░░░░░░░ these days This isn ░░░░░░░░ to say that AI ░░░░░░░░ isn t a real ░░░░░░░░ People have been studying ░░░░░░░░ for decades but they ░░░░░░░░ usually focusing in good ░░░░░░░░ on questions like If ░░░░░░░░ use AI to screen ░░░░░░░░ or if we use ░░░░░░░░ for you know looking ░░░░░░░░ apartment applications is this ░░░░░░░░ to be discriminatory Are ░░░░░░░░ gaps in the data ░░░░░░░░ you know we should ░░░░░░░░ What this is doing ░░░░░░░░ sort of flipping that ░░░░░░░░ into this really familiar ░░░░░░░░ actually the problem is ░░░░░░░░ against conservatives Whatever else ░░░░░░░░ were talking about now ░░░░░░░░ s about this.”