Apples AI struggles
On Feb. 14, Bloomberg News reported that Apple was struggling with bugs and engineering problems on its planned artificial intelligence tools for Siri.
"Apple Inc.’s top executive overseeing its Siri virtual assistant told staff that delays to key features have been ugly and embarrassing, and a decision to publicly promote the technology before it was ready made matters worse."
"...the company’s woes boiled over last week, when it acknowledged publicly that critical features would be delayed indefinitely."
"Despite not being ready, the capabilities were included in a series of marketing campaigns and TV commercials starting last year."
"Walker said the decision to delay the features was made because of quality issues and that the company has found the technology only works properly up to two-thirds to 80 per cent of the time — which is to say it doesn’t work every one out of three times. He said the group “can make more progress to get those percentages up, so that users get something they can really count on.”
Walker ended the meeting upbeat, saying that Apple will “ship the world’s greatest virtual assistant.”
He showed examples during the meeting of the technology working: It was able to locate his driver’s license number on command and find specific photos of a child. He also demonstrated how the technology could precisely manipulate apps via voice control. It embedded content in an email, added recipients and made other changes."
This is very unusual for Apple. A company known for shipping great consumer products which just work. Its also unusual for them to announce in marketing campaigns products which are not yet ready.
But what is the root cause of the problem? "the company has found the technology only works properly up to two-thirds to 80 per cent of the time"
What technology? Its got to be the model. We all know Apple can build end to end systems which are highly reliable. A failure rate of one third is unheard of in Apple products. So it must be the model which is unreliable approx 30% of the time. Wow!
Besides Waymo automonous driving, where else do we see AI integrated into consumer end products?
ChatGpt, Glean, Customer Service chatbots and AI developer tooling are examples. noting: Google has released AI Mode (different from AI Overview) for search which is simiar to chatGpt. And some other interesting announcements this past week in Google IO.
But beyond that, not much, yet. So maybe Apples struggles to build AI enabled consumer devices are not just about Apples execution.
Search is allowed a higher failure rate, than say identity verification, which can never fail. Likewise code prediction. In both cases when AI is wrong, knowledgeable humans can adjust. If the model returns wrong or partially wrong search results it's usually not a big deal. But a 30-20% failure rate is a real hurdle for the next gen of AI products.
Comments
Post a Comment