Google is shutting down a rather obscure project: Google Duplex on the Web. “Duplex” is Google’s branding for AI that does “simple but familiar tasks that saves you time.” The branding exists on two products: this “web” feature and Google’s human-impersonating voice AI, the latter of which is still running as far as we know. This version of Duplex—Duplex on the Web—was a Google Assistant feature that could autonomously navigate websites on your behalf and do things like buy items and check in to a flight. The feature couldn’t have been very popular, and TechCrunch spotted a support page update that says Duplex on the Web will be dead by the end of the month.
Duplex on the Web launched in late 2019 and was announced earlier that year at Google I/O. The normal checkout process for an item involves a lot of navigating and pasting of saved data. You’ll need to find the item and possibly the time slot you want if it’s a reservation, enter in your billing info, and mash “next” a lot, and Duplex on the Web was supposed to be able to do all that autonomously. While it would probably be faster and more reliable if companies just made a voice API, Duplex on the Web was a hack. The Assistant would pop up its own web browser and individually click through checkout screens while you watched. Google’s automated mouse clicker theoretically would have scaled well because it could bring voice support to a website without needing any work from the website owner.
Now, though, it’s dead. Google’s support page says that “Duplex on the Web is deprecated and will no longer be supported as of later this month. Any automation features enabled by Duplex on the Web will no longer be supported after this date.” Google told TechCrunch, “By the end of this year, we’ll turn down Duplex on the Web and fully focus on making AI advancements to the Duplex voice technology that helps people most every day.”
We’ll take a wild guess and say the reason Duplex on the Web is dying is due to a lack of usage. One of the (many) problems with voice assistants is that they are basically command-line interfaces. There’s no UI or buttons that clue people in to what functionality is available, so users have to just know what commands are worth saying. Most people can probably guess “what’s the weather tomorrow?” is a valuable command, but very few people probably knew that the Assistant could autonomously navigate a website to buy a movie ticket or check in to a flight on your behalf. At least command-line interfaces have a “help” command that shows a big list of commands. With no comprehensive list of accepted commands for Google Assistant, it’s not clear how anyone is supposed to learn about these features.
Besides the eternal problem of discoverability, it’s not clear this feature ever actually solved a problem. It’s not exactly hard to buy something on the Internet or check in to a flight because companies try to make those things as simple as possible already.