What Makes AI Work?

Pokémon GO hit peak optimism today as Nintendo stock dropped 17% on the realization that the company doesn’t actually control the game. Nintendo has a share: about 32% of the Pokémon Company, just not as much as the investors who drove Nintendo shares up 64% since the release of the game probably thought. Pokémon also hit peak pessimism as conspiracy-minded film director Oliver Stone denounced the game as a totalitarian attempt to create a robot society by monitoring our every move.

Stone’s understanding of robots isn’t what it might be since Pokémon GO is augmented reality rather than robotics. Games aside, there’s no doubt that robotics is a technology on the rise. We’re witnessing a slew of new robots from driverless cars to garden robots like Farmbot. These technologies follow other familiar, consumer-oriented robots like the vacuum cleaner Roomba and the gutter cleaner Looj.

Robots span a range of capabilities from autonomous operation to human augmentation. Surgical robots are meant to augment the skill of human surgeons, and today’s self-driving cars are supposed to augment human drivers. It doesn’t always work that way in the case of cars, but nobody is currently advertising a self-driving car that replaces the human driver altogether. But fully autonomous cars are just around the corner: GM is partnering with Lyft on an autonomous taxi service to be trialed within the next year at an undisclosed location. Ford is in the game as well, with an investment in an autonomous taxi service set to launch in Singapore in the next two years. And if Lyft won’t be the only partner for the car companies, Uber is also working with top flight engineers on its own autonomous car.

The key to autonomous vehicles is artificial intelligence, something we didn’t hear much about in Silicon Valley before Apple’s Siri. It was there – mainly as a research adjunct to semiconductor and computer companies, but not as a widely-sold product. One of two leaders in early AI is Silicon Valley guy Marty Tenenbaum, an old boss of mine. His East Coast counterpart was Marvin Minsky at MIT. So in the 80s and 9os we had West Coast and East Coast Schools of AI. The East Coast approach was based on “constraint based reasoning” while the West Coast school emphasized expert systems. Today’s AI relies on a number of techniques, especially neural nets built around machine learning.

Even fairly straightforward systems such as home automation rely on AI. Your smart thermostat synthesizes an enormous amount of information about your home, your micro-climate, your movement inside and outside your house and your personal preferences to set your heater and air conditioner at the right spot.

Cool kids want to work on AI systems these days for the usual reasons: it’s a technology with a future and it’s more respectable than simply placing personalized ads on web pages for maximum pull. So the winners are probably going to be the companies that can attract the best people and invest heavily in their support and productivity.

Apple opened his Pandora’s Box with Siri on the iPhone 4S. Siri brought little more than voice control to the phone, but it showed the way to more powerful AI-assisted functions for the phone by creating the critical link between the mobile device with limited processing power and a cloud-based compute farm capable of crunching large data sets toward intelligence ends without draining the device’s memory.

Many Apple watchers seem to believe that Apple’s insular culture is a barrier to AI development. Walt Mossberg, for example is skeptical:

But Apple could have a very hard time in the AI war that’s dawning — even though it actually pioneered the first widespread voice-controlled, cloud-based AI assistant, Siri, on the iPhone five years ago and is now rumored to be bringing Siri to the Mac. (In fact, former Apple CEO John Sculley pushed an AI-powered conversational assistant concept called Knowledge Navigator, complete with concept videos, way back in 1987. It envisioned a conversational digital helper with abilities far beyond any that exist now.)

There are three reasons for my doubts: First, Apple’s history with cloud-based services in general has been weak and inconsistent. Second, Apple has done shockingly little to capitalize on its lead with Siri. And third, Apple’s steadfast devotion to privacy and lack of a search service or social network means it doesn’t have the range and volume of data its competitors hope to use to power personalized, actionable AI capabilities.

And others see Google’s moonshot on the driverless car and investment in infrastructure as a more promising AI approach:

Becoming a major big-data AI services company doesn’t happen completely in secret and suddenly get released to the world, completed, in a keynote. It’s a massive undertaking, spanning many years, many people, and a lot of noticeable interaction with the world. It’s easier to conceal the development of an entire car than a major presence in AI and services.

Google is extremely well-placed for where they think the puck is going. They could be wrong — these AI services could be a socially awkward fad like Google Glass or a tonedeaf annoyance like Clippy. Google launches a lot of weird, geeky, technologically impressive things that go nowhere.

Connecting cloud infrastructure to mobile devices requires an additional piece of infrastructure that doesn’t seem visible to most of the Valley’s tech pundits: a pervasive, reliable, high performance network. Application developers take the network for granted, but it’s the most important piece of infrastructure that will either pave the way to AI or hold it back. But we don’t get from where we are to autonomous cars and smart homes without 4G+ or 5G networks.