Researcher Yann LeCun of Facebook Artificial Intelligence Research Lab (FAIR) is on a humanitarian quest to develop a face recognition AI that could intervene if you drunkenly attempt to upload selfies to Facebook. Wired reports on this development, which builds on the already-existing technology that distinguishes your face when tagging photos, as well as the strides made in voice-recognition technology. The grade of “deep learning” requisite to distinguish your sober self from your drunk self represents LeCun’s ambitions for AI capabilities.
FAIR has, to date, developed algorithms that curate your Facebook feed (based on what you’d be most likely to click on) and hashtag suggestions. One of FAIR’s research papers for the Conference on Computer Vision and Pattern Recognition claims that their face-recognition technology approximates “human-level performance.” This, of course, can be disconcerting; recognition depends on access to a wealth of personal data including habits, expressions, the company we keep when we’re sober versus the company we keep when we’re drunk, etc. LeCun insists that, where other technologies may have compromised user agency (selling information to advertisers, for example), this AI will offer users more agency, specifically when alcohol has impaired your decision making skills.
The efforts of LeCun’s team represents a growing trend in the “arms race for deep learning talent.” And it’s just another reason to forget about Facebook when breaking out the wine bottles. [Wired] [Image: Flickr/Eugenia Loli]