The reality of AI is less scary than the movies
I was talking to someone at work recently and mentioned the Palm Pilot. He never heard of it. Some of us remember it being released in 1996 before the smartphone and social media, and in the early days of the internet. It drove the creation of the smartphone, though the people at Intel at the time didn't see how a portable, hand-held device like this could become common. One of the founders and inventors was Jeff Hawkins who also founded Handspring and worked on the Treo that evolved into a very early smartphone with a camera, which this brings us to today's topic, artificial intelligence.
- After the Treo, Hawkins moved out of the digital world into neuroscience and for the past few decades has been working on understanding the neocortex and how intelligence works in humans. In his latest book A Thousand Brains: A New Theory Of Intelligence, he explains the discoveries made by his team on how the neocortex makes models of the world, how these may work and how they are linked together. He also distinguishes between the old brain and the neocortex as the intelligence centre. Then the book moves onto how these discoveries might be applied to artificial intelligence.
- I have written about that subject a few times in the past but as I was reading the first section of the book, I was already thinking how this could map to modern AI. Most current AI systems do one thing well. They can detect cancer, beat the world's best Go player and, for some parts, drive a car. The Go playing AI for example can't drive a car at all.
- As a human we continually make new models in real time, as we are exposed to new sensory inputs. The current AIs are trained and after that are run. The number of new inputs is small and primarily designed to refine the existing models, not build new ones. Humans have the old brain components and the neocortex integrating in a myriad of ways, current AIs don't model this kind of interaction though there have been some attempts to do so. Next to the neocortex is the limbic system where the emotional responses are processed, versus the neocortex where intelligence sits. When we experience pain and jerk our hand away from a hot stove the neocortex is the last to know about it because you need that old response to kick in before registering it intellectually.
- If you think about the above you can begin to understand how difficult it would be to make an artificial human brain with its old animal and newer intelligence components. The primary focus of true AI is on the intelligence aspects. This is why the concept of, say, the angry sci-fi AI robot doesn't make much sense in the real world. Apart from ego and aesthetics a human form AI doesn't make a lot of sense either. Why build eyes and ears when you can have sensory detectors over the whole surface and provide a far superior model of the external world that way?
- If you are in the field of AI and have not yet read Hawkins' book I urge you to do so, if for nothing more than to provide a different perspective and approach on the subject. We have been trying to build an AI since the 1950s with more downs than ups. This could be because we have been looking at it the wrong way all along. It might also mitigate some of that inner fear about the emergence of something like Terminator's Skynet.
- After spending a lot of time playing games over the years, the two worst games for me are RAID Shadow Legends and Ingress because unless you are super wealthy you will not advance beyond a certain point, especially for the events. They also have the most expensive items for purchase I have ever seen in a game, by far. I'm most of the way through a year of not advancing at all and elements like the Team Arena are just for masochists. The worst is Ingress by Niantic. Their latest version has been out for a long time now and it still freezes on top-of-the-line smartphones and the game stops displaying energy often reducing playability since it's a real position-based game. Both games have other issues.
- In other news this week, there is finally a push to start making computer chips in places other than China and Taiwan with a number of companies talking about moving fabrication to places like the US. Micron for one has pledged US$40 billion towards this effort. I wonder if this has anything to do with the US car industry leading the world in production cuts due to chip shortages. The Google Cloud is now moving in Thailand, New Zealand and Malaysia. Alibaba Cloud already operates in Thailand and Malaysia. I saw this headline on theregister.com website indicating that self-driving cars are still a way off: "Tesla full self-driving fails to notice child-sized objects in testing."
James Hein is an IT professional of over 30 years' standing. You can contact him at email@example.com.
- artificial intelligence