Elon Musk and Mark Zuckerberg might not be able to agree about much when it comes to AI these days, but the pair do seem to see the same potential in Vicarious, a startup applying unsupervised learning techniques to robots. Musk and Zuckerberg were two of the early backers of Vicarious. The startup announced that it raised an additional $50 million in financing this morning (via VentureBeat), bringing its total fundraising to $138 million (according to PitchBook).
Scott Phoenix and Dileep George co-founded Vicarious back in 2010. Phoenix previously founded Frogmetrics, a customer feedback platform, while George founded Numenta, another R&D-heavy AI startup. The pair have never been explicit about Vicarious’ products, but every few months the company’s work seems to surface.
Back in 2013, Vicarious claimed that it had efficiently beaten Captchas — able to do so at high accuracy with minimal training data. Then four years later, the startup made headlines again for its Atari Breakout-playing AI.
Vicarious is searching for the Holy Grail of artificial intelligence — generalized intelligence. Google’s DeepMind and Microsoft’s Maluuba (among others) have been pouring resources into the same objective. But building generalized systems isn’t some easy upgrade. It took DeepMind almost seven years to get AlphaGo where it is today — and even so I’d call DeepMind’s work more flexible than general.
Vicarious’ approach is rooted in Schema Networks. Supposedly, by maintaining an abstract conceptual understanding of cause and effect within the task at hand, Vicarious’ Schema Networks can run effectively in different environments without retraining. This doesn’t mean that a Vicarious model could play both a first-person shooter and a game of Pong, but it could ideally play Pong with different paddle positions.
Given that Vicarious doesn’t really have a formal product, it’s difficult to judge the company’s long-term sustainability. But it’s clear that today’s lead investor, Khosla Ventures, is a believer — and that may be all that the startup needs for now.