Silicon Valley is glutted with everything that the most popular form of modern AI, called deep learning, needs to be successful: massive stockpiles of private user data and a battery of high-powered servers.
The next wave of AI, however, may look quite different. A number of startups are now thinking about ways to minimize the amount of training—and thus, the amount of data and server power—needed for AI. Samsung Next, the South Korean technology company’s VC arm, is pushing this trend forward with the launch of a venture team called Q Fund. The team appears to be looking for companies that focus less on specific applications of AI (say, farming or algorithmic trading) and more on developing new ways of making decisions with computers.
“Deep learning has its own baggage,” Ajay Singh of Samsung Next told Quartz, mentioning problems with the technology like its propensity for bias and the amount of data needed. “And where is the concept of debugging?”
Singh points towards one of Q Fund’s initial investments before its official launch, a startup called Vicarious, whose big idea is giving machines “imagination,” turning to biology for inspiration on how to make machines learn to do tasks faster. Q Fund participated in a funding round of Vicarious led by Khosla Ventures last July.
Q Fund also invested in automation-robotics startup Covariant.ai, previously called Embodied Intelligence, founded by well-known AI researcher Pieter Abbeel. That startup’s mission is to use virtual reality and simulations to reduce the amount of time robots need adjusting to their jobs in the real world.
In an attempt to mitigate the risk of writing checks to companies selling false promises, Q Fund has established a deep stable of industry and academic experts from MIT, Carnegie Mellon, Princeton, and Toronto’s Vector Institute to vet potential investments.
“We’re okay if it’s unproven,” says Samsung Next’s Vin Tang. “As long as the upside is big enough.”