This is especially true when it comes to algorithms like ChatGPT, or any number of other programs that can write sentences and even whole essays that sound perfectly reasonable-until you realize that a lot of what they return is completely false. Pete Ryan AI Nightmare #2: All Hail the Stochastic ParrotĬomputer scientist Amy Zhang, head of the University of Washington’s Social Futures Lab, tells me that what worries her most is AI’s propensity for nonsense. “These systems are generating personal data that nobody gave them, and it becomes more real than the truth,” Whittaker says. Many databases are centralized, so it’s easy for a data double to follow you around for years or even a lifetime. Because so much of our data is repackaged and sold to third parties, a data double created by artificial intelligence might disappear from your insurance office, but show up again in your bank records, your school admissions paperwork, or a workplace background check. Meredith Whittaker, CEO of the private-messaging app Signal and president of the ethics organization AI Now, calls the skewed online versions of real people that MiDAS created “data doubles.” They function a lot like stolen identities, and once established, they’re nearly impossible to delete. When the truth came out about the flawed algorithm, a group of the victims brought a class action suit against the state, which was settled in late 2022. Perhaps worst of all, none of them could understand why they were suddenly being treated like criminals. Some of the accused were unable to get jobs, and others could no longer receive their unemployment payments. AI Has Successfully Imitated Human EvolutionĪs if that weren’t bad enough, once MiDAS incorrectly identified thousands of innocent people as fraudsters in a state database, other institutions that used the database to do background checks repeated that fake information.Over the next two years, MiDAS falsely accused more than 40,000 people of fraud, in many cases forcing them to pay egregious fines. Though the agency has never revealed where the algorithm harvested data, subsequent lawsuits suggest that it may have collected inaccurate details from unemployment data sets furnished by employers. Working with three different software development companies, the agency unleashed the Michigan Integrated Data Automated System, or MiDAS, in 2013. In 2011, the governor of Michigan authorized funds for the Michigan Unemployment Insurance Agency to develop a new algorithm that was supposed to detect fraud. AI Nightmare #1: Curse of the Data Doubles Here, we’ve collected a few scenarios that keep AI experts (and the rest of us) up at night. Play icon The triangle icon that indicates to playĪccording to researchers, the algorithms, models, and programs we call “AI” are about to make our lives a lot worse before they make them better-just (mostly) not in a way that will require cage fights or surplus military rations. More dire research into the ways armies will incorporate machine learning suggests that innocent people might die when the AI we develop to help us make smarter decisions instead makes unforeseen mistakes. Instead of human batteries and skull-crushing robots, we will get unrelenting traffic jams, screwy credit reports, and an exacerbation of inequalities that already exist. The experts’ mostly hopeful projections foresee a future where AI causes insidious and annoying problems. We asked specialists in the field how it might all go horribly wrong-from AI takeovers of autonomous weapons to unscrupulous chatbots. An entire field of AI research is focused on understanding its dangers and-let’s all hope-preventing them. Over the past few months, AI technology has advanced in enormous, headline-making leaps, making many machine learning experts wonder whether similar horrors lurk in our near future. Then there’s the popular thought experiment from University of Oxford philosopher Nick Bostrom, in which he imagines a superpowered AI that’s been trained to optimize a single function-so it methodically diverts all of the planet’s resources into making paper clips. The machines in The Matrix convert humans into batteries whose brains are permanently plugged into a simulation of the year 1999. Skynet nukes Earth and then unleashes armies of evil robots in the Terminator movies. I n science fiction, artificial intelligence is always causing epic disasters.
0 Comments
Leave a Reply. |