After The AI Winter Comes The AI Spring

Blurry image of a winter forest, small snowdrifts and light snowfall – a beautiful winter-themed background wide format.

getty

By almost any account, it seems like AI is booming right now. But experts who study market history are carefully attuned to boom and bust cycles and the inevitable volatility in nearly any market. Between any big spikes in enthusiasm for neural networks and the like, there’s the specter of an AI winter. And we’ve had them before.

So what is an AI winter? Most definitions describe this kind of phenomenon as featuring a sharp decline in funding, and a herd mentality that considers the technology to be overhyped, and less popular or relevant in the future.

A resource at OpenDigital marks two AI winters, 1974-1980 and 1987-1993 or 1994.

“During these cycles, projects lose public and private support,” writes Jose Edgardo Morales Barroso. “As a result, the field enters a pause and is sometimes redefined.”

Keeping the Flame Burning

Many of us who came through that era also noticed the effects of an AI winter. For instance, there’s my colleague Steve Jurvetson, who got his MBA at Stanford in 1995, and now runs Future Ventures, a space and technology venture fund, and has a trove of space artifacts in his collection. I spoke with Jurvetson at our Stanford event this month, and he described the AI winter this way:

“I remember this vividly, because there was a long drought,” he said. “There were a few people that stayed at it, the old, famous neural network researchers in academia, but it was tough to get funding to make progress. People thought it wasn’t going to work. I had a fundamental belief that this is the architecture of the mind … and … an inevitable future vector of AI development. But it wasn’t working yet.”

He cited work on AlexNet in 2012 as the turnaround point, and talked about his early experience.

“I knew a guy who … built a home supercomputer and was doing neuronal modeling, of all things, a scientific kind of project, using them,” he said. “So I had this prepared mind, if you will, to say, ‘Wow, the computer architecture being more fine grained, being more like the brain, might actually be the future of AI.’”

130 Years of Moore’s Law

One of the high points of our talk was when Jurvetson showed off a scattershot graph representing over a century of progress under Moore’s law, which holds that the computing power for cost doubles each year. He explained how this trend held true across different kinds of analog and increasingly digital technologies, from relays and tubes to GPUs. You can see Nvidia’s contributions toward the top.

“The gray bars vertically show different technology epochs, everything from the vacuum computer that predicted Eisenhower’s win in ’56, to the relay based computer that … cracked the Nazi enigma code,” he said. “And no one knew they were on a curve back in the early days, right? This is an exponential. This is actually like a double exponential. It’s held for 130 years. That’s mind-bending … the Great Depression had no long-term impact on this. Economic recession had no impact on this.”

A Bit of Futurism

We talked automation, profound automation, and we talked space travel.

“How can we bring these neural networks into everything, every security camera, every car, every physical device?” Jurvetson asked, invoking the power of edge computing. “Imagine you could speak with a voice interface (in) your room, but it would do what you want for less cost than the plastic buttons on the interface today, right? Wouldn’t that be a better future?”

We talked about military applications, and a “huge bolus of research.” Here’s his prediction on Mars:

“Sometime in the next 10 years, you know, we’re going to be doing (voyages),” he mused. “You know, there’s some fuzziness, but it’s not hundreds of years. It’s within ten.”

He did not express interest in being on the early train to the red planet.

“I want to do the very low orbit lunar vacation, but not a settlement on Mars,” he said, calling the sister sphere “the first landing party for the humanoid robot” and noting his interest. “It’s sort of obvious now, in retrospect, that you want to prepare the environment with the greenhouses and everything else for the eventual things that will go there; the reason to go to Mars is (to create) a permanent settlement of humanity.”

Imagination in Action

Due to the name of the event, I asked Jurvetson the following question: How do you think our society needs to be imaginative right now, when we have all these tools?

“I think creativity is the fountainhead of human endeavor,” he said. “This is what creates progress. This is what allows us to move forward. If you look back at history, it wasn’t so obvious that innovation thrives in flourishing, but it’s pretty obvious today … having a creative fountainhead of new ideas is essential.”

He also came up with some

“My professional life is one of lifelong learning, of always looking for the next great thing,” he said. “We have a framework at Future Ventures, which is, we try to invest in things that are unlike anything we’ve seen before, yet adjacent to where you’ve been. So think about these as frontiers or final frontiers, and other frontiers of the unknown. That’s where we learn. That’s how we have this dynamic dynamism to do what we do.”

Inspirational stuff. What do you think will happen in the next ten years?