SaveSave story

Experts say AI isn't quite fire or the wheel — yet

Illustration: Caresse Haaser, Rebecca Zisser / Axios

Some researchers and business leaders are putting artificial intelligence — in its current and aspirational forms — on the same pedestal of human invention and innovation as fire, electricity and the light bulb. But other experts say we will not know for a long time whether AI will ever merit such lofty imagery.

"It could be. If AI really leads to the birth of intelligences greater than humans', it will arguably be the most important event in the history of life on Earth since, well, humans. But that's a very big if, of course. In the meantime, AI's impact is far smaller than electricity or fire's (and in fact, you could say that AI is part of electricity's impact, since it wouldn't exist without it)."
— Pedro Domingos, professor of computer science, University of Washington

If this were the Stone Age, where we’re at with AI is “we know what wheels are but not how to build them,” Gary Marcus, a psychology professor at New York University and founder of Geometric Intelligence, an AI company acquired by Uber in 2016, told Axios. "But wheels are a whole lot easier to build than AI."

Consider this: Fire arguably made humanity. Taming it more than a million years ago brought our ancestors safety from predators and allowed us to leave the trees. When early humans learned to use it to cook food, it provided more calories that “jump started a brain that was getting larger already,” says Jeremy DeSilva, an associate professor of anthropology at Dartmouth College. (And making fire when it doesn't happen incidentally is a completely different animal altogether, he says. That probably didn't happen until 30,000 years ago.) Like toolmaking before it and farming after, reining in fire was a cultural innovation that spawned a biological change to our species.

“Without the control of fire, I don’t think there is a Homo sapiens,” says DeSilva.

Where it's at: Machines employing neural networks — one AI method — have advanced at recognizing images and translating language. They classify data fed into the system and learn to recognize patterns from which they can predict what is coming next in sequence or sentence. Twenty years ago, language processors got one out of 3 words wrong and weren't usable, says Thomas Dietterich, an emeritus professor of computer science at Oregon State University. "It is astonishing that we now have Skype translator."

These advances have led Elon Musk and others to worry about superintelligent machines.

But these AIs do not make inferences from limited information (i.e. when you put something in a box, the AI doesn't understand that it will stay there until it is removed) and they lack the ability to have a conversation.

“I don’t think we’ve come close to fulfilling that promise yet,” says Marcus, who recently published a much-discussed paper about the limits of the AI technique deep learning. Speaking this week to the Financial Times, Marcus said hyping of AI could lead to a third "AI winter," a new period of arrested development in the field.

Profound possibilities

As AI is developed and deployed in various industries and disciplines, it's realistic to expect it will succeed in helping in some ways and be frustrating in others. “I think [AI] has potential in the long term to be really profound in the same way something like genome editing has potential to be profound,” says Marcus. "It is certainly the case that it could be but the jury is still out."

Biological: Whether and how AI and technology more broadly will have a biological impact on our species — like fire or farming — is unknown. "For all the other events, there is a feedback loop. We can see the biological impact with the other transitions but with the technological transition we don’t know," says DeSilva. We aren't at the point of determining whether screen time, for example, will change our species' vision or how high-functioning neural prosthetic devices might affect us.

Philosophical: "Are there things that can be known by a computer that we cannot ever understand? I don't know," says Dietterich. Pointing to top Go players who are studying DeepMind's AlphaZero — which learns to play chess and Go without a training set — to improve their own games, he says, "Now we can use computers to understand things that we don’t know and then try to understand them. And that makes us better."

SaveSave story

Some self-driving car companies hit brakes on tests after fatality

A nuTonomy test vehicle in Boston.
A nuTonomy test vehicle in Boston. Photo: nuTonomy

Following the death Sunday of a pedestrian struck by an Uber self-driving car, some companies are halting their own autonomous vehicle testing, Among those who is halting testing is nuTonomy, which says it did so at the request of the city of Boston. Uber also halted its testing, as has Toyota, per Bloomberg.

Why it matters: The first death caused by a self-driving car was sure to shake up the industry, despite arguments that self-driving cars are significantly safer than humans behind the wheel.

Zachary Basu 7 hours ago
SaveSave story

GM still plans to launch self-driving cars next year

GM autonomous
Mary Barra, Chairman and CEO of General Motors, speaks during a press conference about the company's autonomous vehicle program. Photo by Rachel Woolf / Getty Images

General Motors said it will not delay its plans to field a fleet of self-driving cars next year in the wake of yesterday's fatal accident involving an autonomously-driven Uber vehicle.

"Protecting the safety of the people in the communities in which we live and work, as well as our employees, is at the heart of who we are as a company. Our plans to commercially launch in dense urban environments in 2019 remain unchanged but, as we’ve said from the start, we will not launch until we are satisfied that it is safe to do so.”
— GM spokesperson

The state of play: This follows news that Uber and Toyota have halted testing of their autonomous vehicle programs after yesterday's accident.