How they did it: The crew requested language fashions the place they stand on varied subjects, comparable to feminism and democracy. They used the solutions to plot them on a political compass, then examined whether or not retraining fashions on much more politically biased coaching information modified their conduct and skill to detect hate speech and misinformation (it did).
Why it issues: As AI language fashions are rolled out into services utilized by hundreds of thousands, understanding their underlying political assumptions couldn’t be extra necessary. That’s as a result of they’ve the potential to trigger actual hurt. A chatbot providing health-care recommendation may refuse to supply recommendation on abortion or contraception, for instance. Read the complete story.
—Melissa Heikkilä
Read subsequent: AI language fashions have not too long ago develop into blended up within the US tradition wars, with some calling for builders to create unbiased, purely fact-based AI chatbots. In her weekly publication all about AI, The Algorithm, Melissa delves into why it’s a pleasant idea—however technically not possible to construct. Read it to seek out out extra, and if you happen to don’t already, signal as much as obtain it in your inbox each Monday.
The must-reads
I’ve combed the web to seek out you at the moment’s most enjoyable/necessary/scary/fascinating tales about expertise.
1 A girl was wrongfully arrested after a false face recognition match
It’s notable that each particular person we all know this has occurred to has been Black. (NYT $)
+ The motion to restrict face recognition tech may lastly get a win. (MIT Technology Review)
2 AI startups are preventing soiled 😈
We’re speaking faux names, rivals posing as prospects, and even bombing Zoom calls. (NYT $)
+ It’s all beginning to look rather a lot like a bubble. (WP $)
3 A vote in San Francisco might change the way forward for driverless vehicles
All eyes are on whether or not the state board will approve an enormous enlargement of autonomous taxis on Thursday. (NBC)
+ Big tech firms are struggling to win over native residents and public officers. (WSJ $)
4 Is Texas’ electrical energy grid going to have the ability to deal with electrical autos?
There are causes to be optimistic, not only for that state however the US as a complete. (The Atlantic $)
5 Criminals are enthusiastic early adopters of AI instruments
On the darkish internet, they declare to have created two giant language fashions that may help with unlawful actions. (Wired $)
+ Criminals are additionally utilizing AI-generated books to rip-off individuals. (NYT $)
+ We are hurtling towards a glitchy, spammy, scammy, AI-powered web. (MIT Technology Review)
6 The period of plentiful low cost stuff could also be coming to an finish
Maybe that’s not a completely dangerous factor, frankly, for the sake of the planet. (WSJ $)
7 People are eager to recreate Black Twitter elsewhere
There’s been a large exodus from the positioning. But the place ought to people go? (WP $)
8 Big cities want to vary
To thrive, they should reinvent themselves to be extra than simply locations the place individuals work. (Vox)
+ What cities want now. (MIT Technology Review)
9 WhatsApp is engaged on 32-person voice chats
Sounds like pure chaos! (The Verge)
10 Even Zoom is making staff return into the workplace
Ironic, maybe. But not that shocking. (Quartz $)
Quote of the day