What we're reading: AI that's really people
Add Axios as your preferred source to
see more of our stories on Google.
Tech companies are billing products as AI-powered for that future-wow factor — even if artificial intelligence is only a small part of the equation. In The Guardian, Olivia Solon charts the rise of "pseudo-AI" products that have humans quietly do the dirty work.
Why it matters: Some experts warn of an impending "AI winter" if currently high expectations for the technology are disappointed and investment wanes. Revelations that AI is just underpaid people could spark just that kind of disillusionment.
Companies keep getting caught outsourcing the work their technology fails to do. Some examples from The Guardian's Solon:
- Products that promise to scan your inbox and use algorithms to tease out discounts and travel itineraries are sometimes putting private emails in front of humans, a Wall Street Journal investigation found this month.
- Expensify, the popular business-expense management system, paid people on the Amazon Mechanical Turk platform small amounts to decipher customers' receipts, but billed the system as automated, Ars Technica reported last year.
- Facebook Messenger's virtual assistant, M, used people to fulfill customer requests that the bot itself couldn't, Recode reported in 2015. M was shut down earlier this year.
- In 2016, Bloomberg profiled the overworked employees who impersonated AI chatbots and personal assistants, delivering services disguised as AI.
Reality check: Some AI experts, like Berkeley's Michael Jordan and Gary Marcus of NYU, argue that many overestimate what AI is currently capable of. For now, talking AI can only hold conversations under very constrained conditions, and small corruptions can throw off computer-vision algorithms.
Yes, but: Bullish researchers argue that a growth in three key areas — computing power, the quantity of available data, and creative new algorithms — will continue to propel AI development and stave off crisis.
Go deeper: Some computer scientists are looking past deep learning, the most popular tool in AI today, for the next breakthrough (NYT)
