It sounds like a line from science-fiction: Two of the world’s most famous technologists are butting heads over artificial intelligence.
Tesla/SpaceX CEO Elon Musk and Facebook CEO Mark Zuckerberg have very different ideas on what qualifies as “AI,” according to a recent Vanity Fair article detailing whether popular male tech figures believe in a forthcoming AI apocalypse. Their debate underlies a bigger one facing the industry: nobody can agree on what the term means.
There are generally two schools of thought when talking about whether software today (read: not an imaginary future killer robot) is a form of artificial intelligence: literalists and generalists.
Musk is a literalist. When asked about Zuckerberg’s Jarvis, a bot that mainly handles home-automation tasks like controlling the thermostat or picking music to play, Musk replied dismissively.
“I wouldn’t call it AI to have your household functions automated,” Musk told Vanity Fair. “It’s really not AI to turn the lights on, set the temperature.”
What Musk means is that one, cohesive artificial intelligence isn’t considering Zuck’s request, understanding what he means, and then turning off a light. In reality, Zuckerberg’s voice is being sent to a speech recognition algorithm, which then transcribes some text, then read by a separate algorithm to determine what the text is asking, and its decision initiates a separate piece of code that turns on a light.
It’s a bunch of algorithms working together to initiate a command, instead of one algorithm that understands voice, text, code, and an array of ideas. This is also true of Siri, Cortana, Google Assistant, and every other virtual personal assistant available today. They work by stringing together algorithms that determine what a human wants, and then either read from a script to give the answer or pull a canned response from Wikipedia. It’s a marvelous feat of design—but not what AI researchers would call an “end-to-end” model, or one algorithm that learns to complete a task entirely by itself without human tinkering.
For instance, a 2016 Nvidia paper describes end-to-end learning for self-driving cars. One neural network, a popular AI technique that learns to predict how to do actions by identifying patterns in data, looks at videos of how cars typically drive, and then from that data can steer a car by itself. This is probably closer to AI for Musk, although Teslas have dedicated vision neural networks, which would mean they’re not end-to-end, and he still calls them AI.
A lot of pure AI research (at least in the view of this reporter) focuses on the ability to learn to identify patterns in images, audio, or text, and then recognizing patterns in the real world or generating its own. Anything that hasn’t been generated by the AI, like a human-written response from Siri, is suspect.
Others go further. Some, like cognitive scientist Douglas Hofstadter, say that since machines that generate text are only statistically predicting which words would come next in a sentence, as opposed to understanding a concept and constructing a sentence to relay that information, there is no artificial intelligence yet.
Zuckerberg, on the other hand, is an AI generalist. Artificial intelligence is an umbrella term for a system that can understand commands and complete tasks. It’s a definition a lot like Facebook: user-friendly and accessible for those who don’t speak tech.
Jarvis does use some forms of artificial intelligence: The algorithms that understand speech and parse what Zuckerberg wants from the transcribed text are neural networks.
Like a lot of us, Zuckerberg uses AI as a catch-all phrase to mean software with some AI component. Generalists typically argue that as long as a program like Siri or Cortana can choose to initiate an action, whether that’s turning on a light or a bank transfer, the architecture of that system doesn’t matter when casually referenced. This argument focuses on the impact of this software rather than how it works.
No matter how Musk and Zuckerberg define the term, both believe it’s incredibly important. Both technologists fund AI research groups, OpenAI and FAIR, respectively, and stand to profit greatly from its proliferation—Zuckerberg through his social network and Musk through self-driving cars.