Speaking from the sidelines of the JPMorgan Global China Summit, the former Google DeepMind researcher outlined the massive capital requirements needed to prevent a machine from identifying wooden furniture as a flock of geese.
SHANGHAI — Outlining the ambitious growth trajectory for his new visual reasoning startup, Elorian co-founder and CEO Andrew Dai announced Tuesday that the firm is seeking $50 million in seed funding to hire enough machine learning specialists to successfully teach a computer what a chair is.
Dai, a former researcher at Google DeepMind, told Bloomberg's Haslinda Amin that the massive influx of capital is strictly necessary to overcome the monumental technical hurdles of getting software to look at a four-legged piece of living room furniture without classifying it as a commercial airliner.
According to Dai's funding pitch, the $50 million will primarily be used to poach specialized engineers who can meticulously label millions of images of chairs, couches, and ottomans, thereby preventing the artificial intelligence from hallucinating that it is trapped in a sentient forest.
To the layman, seeing a chair is a passive, instantaneous event, but for a machine, it requires processing billions of parameters just to rule out that it isn't looking at a horse.
Dai noted that while his tenure at DeepMind provided him with near-infinite computational power, a startup must be highly targeted with its resources. He explained that Elorian's flagship model currently analyzes a photograph of a standard dining room set and confidently outputs "mild weather," a glitch that he estimates will take at least 18 months and $30 million in compute costs to resolve.
At press time, Dai was reportedly assuring a group of skeptical venture capitalists that if they doubled the funding round to $100 million, the startup could likely teach the computer to recognize a table by 2027.