Musk's Artificial Intelligence Warnings Are Hype
As we await a detailed blog post from Elon Musk about the "existential threat" posed by artificial intelligence, perhaps pondering all his previous warnings on the subject, it's useful to consider two relevant factors: the actual state of AI research and the Tesla founder's extraordinary capacity for self-promotion.
Musk claims his fear that machines will take over the world comes from his familiarity with the work of companies such as U.K.-based DeepMind, in which he was an investor before last January, when Google bought it for $400 million. Late last month, DeepMind reported what it, and some commentators, have called a breakthrough.
This was to successfully add external memory to a neural network -- that is, to create something like a human's short-term memory to go along with a digital imitation of neuron chains. Here's how Technology Review explained it:
This allows the neural network to store variables in its memory and come back to them later to use in a calculation. This is similar to the way an ordinary computer might put the number 3 and the number 4 inside registers and later add them to make 7. The difference is that the neural network might store more complex patterns of variables representing, for example, the word “Mary.”
Someday, DeepMind's Neural Turing Machine will be able to understand a sentence such as "Mary spoke to John," or even produce it from inputs that feed the working memory. One might have thought computers could already do this, but OK, I'm willing to accept the novelty of DeepMind's approach: It is indeed making some extremely early progress toward replicating the human brain.
If Musk is to be believed, however, in five to 10 years machines like DeepMind's will be able to take over the world.
There's an obvious gap between what AI can actually do and Musk's apocalyptic predictions. Today's artificially intelligent machines aren't even as smart as very young children. Last year, when researchers at the University of Illinois tested the best available technology for IQ, they got, for one system, the score of a 4-year-old child, and only because it scored high on vocabulary and the ability to recognize similarities -- qualities we have come to expect from search engines. These are not skills that make people smarter than computers; they're skills we're rapidly losing as we delegate them to machines.
In March, Musk and other tech luminaries invested in another artificial intelligence start-up, Vicarious FPC. Its goal is to reproduce the neocortex, the part of the brain responsible for vision, coordination, language understanding and math ability. It has gotten only as far as recognizing CAPTCHA sequences, used to tell if a visitor to the site is human or a robot.
In a blog post, Yann LeCun, one of the world's foremost AI experts (who now works for Facebook), denigrated reports of this achievement as "AI hype of the worst kind," insisting that "hype killed AI four times in the last five decades." LeCun, however, is a scientist. Musk, in contrast, is an unconventional marketer of the highest caliber.
Consider the Hyperloop media extravaganza from August 2013. It looked as if people would be riding super-fast, solar-powered intercity trains tomorrow. While that may in fact never happen, the performance forged a strong link between Musk and tomorrow's transportation technology. That's useful for selling expensive futuristic cars, especially when a company as mundane as Toyota has more promising technology in the field.
Or take Musk's announced intention to die on Mars: a plug for SpaceX, the entrepreneur's rocket company that needs government orders and subsidies to stay afloat.
Musk is good at hype, and Vicarious constantly needs more funding. The latest big investment, $12 million, came earlier this month from the Swiss electrical engineering company ABB.
Vicarious may at some point get quite good at recognizing physical objects, but it's a big leap of faith to imagine its system telling a biker in 2019, "I need your clothes, your boots and your motorcycle."
It's much easier to imagine it as an expensive acquisition target, more lucrative for investors even than DeepMind.
That's why my admiration for Musk as a pitchman so outweighs any fear that artificial intelligence will soon take away my job.
(Corrects dollar amount in 11th paragraph.)
This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.
To contact the author on this story:
Leonid Bershidsky at firstname.lastname@example.org
To contact the editor on this story:
Mary Duenwald at email@example.com