I had some interesting conversations at AAAS on the topic of Artificial Intelligence. In particular around the question: “Can we replicate the functionality of the human brain?”
Everyone I ran into who does actual scientific research on the human brain just shook their heads at the idea of creating an artificial human brain. Their arguments were twofold: (1) We still, after all this time, have no idea whatsoever how to model the brain, and (2) From what we know, the hardware complexity required to replicate just the low level neural activity in a single brain is vastly beyond the combined power of all of the world’s CPUs, even if it turns out that what the brain does is Turing computable in any practical sense.
Furthermore, they don’t think what the brain does is Turing computable in any practical sense. And don’t even get them started on Ray Kurzweil.
On the other hand, pretty much everyone else I spoke with — people who don’t know much about the subject — seemed firmly convinced that we will have an artificial human brain within the next ten years (except for a skeptical few, who thought it might take as much as twenty years).
These non-neuroscientists, generally quite intelligent and informed people, responded to any suggestion that replicating the functionality of the human brain might be out of reach by simply rolling their eyes, while saying things like “Hey, they once thought human flight was impossible.”
Somewhere in here is an interesting story about the extreme disparity of opinion between (1) those who have spent years studying the brain and (2) everyone else.
I’m just not quite sure yet what that story is.
I went to a small neuromorphic computing workshop in Telluride, CO. It was all very interesting, but I learned that the way they are making progress is by drilling a hole in a monkey skull, inserting a probe, and recording the action of a single neuron as they perform experiments. At that rate of learning, it will be many, many years before we understand what all the neurons in a monkey brain are doing, let alone finding humans who want to volunteer to be the monkey.
I read about this billion dollar projects in the US and EU and shake my head. If their goals are anything like how they’re portrayed in the popular science press, I suspect they will be badly disappointed.
I think you’ve summed up the vast gulf between neuro people and everyone else quite nicely!
Experts (of non TED-talky variety) tend to be more circumspect about their fields than enthsiastic outsiders. I guess each needs the other though! Accelerator and breaks!
What I can’t figure out is, how do they talk the monkey into signing the consent form?
I typed in about five different replies to this post. I am deleting all of them and just substituting an ambiguous “Ha ha”. I guess it depends on what you mean by the intelligence of the human brain, now doesn’t it.
Perhaps artificial intelligence is the wrong approach. Instead of trying to replicate the behaviour of neurons from a micro scale, maybe we should be spending more time investigating the interconnectivity of the human brain on a macro scale.
Have you ever heard of the Human Connectome Project? They have developed a scanner accurate enough to be able to image every neuron in the entire brain, and image processing algorithms to turn those images into a three-dimensional map of a brain’s neural structure (it’s surprisingly regular!)
Now what if someone were to take that map… and attempt to simulate neural activity?
One difficulty with starting from the brain’s neural structure is that there is a huge leap from the physical structure to its neural activity, which is much more complex than the physical structure itself. In a very rough analogy, it would be like having just a computer in hardware with no software, and with no real idea how to program, or even how a computer program might work. What’s worse, these “programs” are a lot more complex than the programs inside a computer.