When Androids Dream Of Electric Sheep: Dreaming And Artificial Intelligence

When Androids Dream Of Electric Sheep: Dreaming And Artificial Intelligence

We already know that dreaming is a cross-species phenomenon. Dogs, cats, and even duck-billed platypuses dream. Renowned science fiction author Philip K. Dick once asked, Do Androids Dream of Electric Sheep? (His book with this title was the inspiration for Blade Runner, by the way). Turns out—in a way—they do.

In the interest of not jumping the gun, it’s important to say that researchers have not yet achieved anything like "true" artificial intelligence—that is, something that we can say is intelligent in the same way that humans (or dogs, cats, and duck-billed platypuses) are. Increasingly advanced examples of robots that can simulate intelligence abound, but we’re not yet living in the age of Bicentennial Man.

That being said, thanks to advances in a special type of computer processing known as neural networks, AI researchers have been able to create machines that actually learn new things with minimal supervision. Recent breakthroughs in fields like speech and image recognition are largely due to advances in so-called deep learning technology.

Sleep is an essential part of learning in human beings—far from being a passive process, sleep is when our brains consolidate, organize, and even develop insight into information we’ve acquired during the day. The extent to which dreaming plays a role in learning is up for debate.

Some researchers believe that incorporating sleep and dream-like processes into deep learning for AI is a crucial step. A Paris-based research initiative aptly named Deferred Restructuring of Experiences in Autonomous Machines, or Project Dream, is working on exactly this. Researchers involved in this project span multiple major European universities. They claim that allowing robots to review what they’ve learned in scheduled "off-time" helps them undergo the sort of deep learning that’s leading to major breakthroughs in AI. Google is doing a similar thing with their DeepMind AI.

The key similarity between said off-time and human or animal dreaming is "unstructured learning. The technical details behind this process are beyond the scope of this blog, but if you’re interested, check out this video for an example of what an actual AI dream looks like. It’s safe to say that the future is here.

Previous article A Life Well Slept