This post was also published on TechCrunch.
On a 10- to 20-year horizon, large-scale technological innovation is going to center around machine intelligence, robotics and sensors. Each of these fields requires gargantuan amounts of capital and a lot of patience, a combination well beyond the scope of even the most progressive venture capital firm.
As Google has demonstrated with its self-driving car, the combination of machine intelligence, robotics and sensors can already perform better than a human at a complex task such as driving a car, something that 10 years ago was unthinkable to most people.
No doubt, Tesla has built an amazing car and after much trial and tribulation, brought it to market. However, General Motors had already shipped a production electric car years before. Tesla took advantage of the innovator’s dilemma, where legacy car companies are virtually incapable of embracing electric-only cars and integrating modern electronics.
Tesla’s roadmap includes “autopilot” and eventually “autonomous” features. Perhaps Tesla also will deliver these features slightly before legacy car manufacturers do, including Mercedes and Lexus, which are aggressively adding similar features. But the winner in this game is Google, which has a multi-year technology lead and can extract enormous licensing fees.
The amount of raw computing horsepower necessary for cognitive computing is massive. Integrating next-generation sensors such as LIDAR is extremely complicated. The regulatory environment for introducing smart machines is extremely unpredictable, and the required experimentation and unpredictable timelines of this type of work puts it squarely in the “research” side of research and development. And venture capitalists inherently hate research – they like development.
Even the biggest venture-backed play in machine intelligence, DeepMind, with $50 million in very patient funding and 75 top researchers on staff, was recently acquired by Google for $500 million. Google has been on a massive buying spree lately, snapping up the building blocks of future technology with robotics acquisitions, other AI acquisitions such as DNNresearch, and key hires such as Ray Kurzweil, who is considered by many to be the godfather of commercial cognitive computing. Google is among the only customers of D-Wave, a much-maligned quantum-computing company.
Companies like Google, IBM and Microsoft have been building out machine-learning teams that can leverage their investments in vast networks of computers built around the globe. The amount of transistors needed to match the number of neurons in a human brain is a tremendous 100 billion, and it will take us until around 2025 to replicate on a computer chip, according to Kurzweil, although this might be aggressive due to the physics around increasingly smaller transistors in microchips.
The market uptake of machine intelligence is going to take a while. IBM has shown that a computer can outperform humans at chess and Jeopardy, and is transitioning its cognitive computing work into fields such as medicine. Even at the scale of an IBM, shareholders are complaining about the cost of this transition and the head of the Watson group was recently replaced. If IBM’s Watson were a startup, its investors would have long ago forced it to sell so that they could put their capital into more efficient short-term and mid-term investments.
Other companies with large computing grids are starting to get into the game. Facebook recently kicked off an artificial intelligence lab with the hire of Yann LeCun from NYU, and also acquired speech recognition company Mobile Technologies. EBay hired Hassan Sawaf from SAIC to spearhead its machine-intelligence efforts. Yahoo! has spun a deal with Carnegie Mellon to access their researchers. Apple, with its relentless focus on the client side of computing, is struggling to keep up with making its email systems scalable, let alone keep its Siri acquisition on par with Google’s relentless drive towards machine intelligence.
The upside is huge. Every sector of the economy will soon enough have its own version of a self-driving car, even fields as advanced as medicine. As anyone with an undiagnosed or somewhat diagnosed medical condition can tell you, the amount of guesswork and over specialization of medical professionals is maddening. A computer can take a holistic approach and quickly narrow down what a problem could be, and iterate with exclusionary tests. Given the fundamental shift required, it will take quite a while for this transition to happen in fields such as medicine.
Despite repeated fears, startups are definitely taking on far bigger challenges than just photo and chat apps. But will startups be able to compete with these giants in areas such as machine intelligence? Perhaps Google, IBM and Amazon will offer Cognition-as-a-Service that would usher in a wave of new companies, much like Amazon’s Infrastructure-as-a-Service offering reignited the web.
IBM has created a $100 million investment fund for Watson-based companies and just made its first investment. Amazon could kickstart a cognition service by acquiring nascent cognition service companies such as Wise.io, Expect Labs, and BigML and offering them at scale. Now that would unleash a generation of “smart” startups.