A.I. IN THE NEWS
Fix AI lending with...more AI: The finance industry has struggled to use AI in assessing who is eligible for loans—too often systems are built on data suffused with historic racism. An HBR author proposes building training models “not merely on the loans or mortgages issued in the past, but instead on how the money should have been lent in a more equitable world.”
Show us the code: AI is suffering from a replication crisis. Academics are fed up with companies like Google publishing flashy research without sharing the data that underlies it. “It’s more an advertisement for cool technology” than science, says one researcher who, along with 30 others, called on the journal Nature to require contributors to disclose the source of their findings.
Stand down, slaughterbot: Fears of warfare involving autonomous “slaughterbot” machines are likely overstated, says an Axios report on military use of AI. Even though drones proved decisive in the recent Armenia-Azerbaijan conflict, the biggest role for AI in the future is likely to be speeding human communication and decision-making rather than making Terminator-style machines.
Everything in moderation: Facebook provided more details about how it's using AI to police its troubled platform. The tech giant says machine learning now forms a bigger part of its moderation efforts, but acknowledges there are limits. “The system is about marrying AI and human reviewers to make less total mistakes. The AI is never going to be perfect,” says a Facebook engineer.
FORTUNE ON A.I.
A.I. recruiting startup wins competition to help military veterans find jobs—By Jeremy Kahn
The rise of the MOOCs: How Coursera is retraining the American workforce for a post-COVID economy—By Beth Kowitt
Apple’s user tracking prompts privacy complaint from Facebook nemesis Max Schrems—By David Meyer
He’s worried A.I. may destroy humanity. Just don’t confuse him with Elon Musk—By Jeremy Kahn
Catch me if you can: Remember the Pokémon craze of 2016, when half the world was stumbling down sidewalks and back alleys trying to capture the digital creatures on their phone? That cooled off, but Pokémons are still with us—and, thanks to AI, now number more than ever.
Programmer Matthew Rayfield makes digital toys, and recently used his own code and OpenAI's Generative Pretrained Transformer 2 (GPT-2) to create 3000 new little beasties, some of which could pass as close to real Pokémon.
From a Vice article explaining the project:
“The result was 100,000 lines worth of sprites. He used that text to re-train GPT-2, which output a random text-based sprite, and then reverse-engineer the line version into a colored-in image. What comes out are garbled little pixel creatures that, if you use your imagination, are pretty close to Pokémon.”
It’s a cool project, but I’m waiting to see if Rayfield can figure out to how make legions of new Transformers.