Singularity Summit: How Will We Get There?
Now that I’ve had a week to digest what I saw at the summit, I have some thoughts on the most likely path we’ll take to the singularity. From an absolute perspective, this path isn’t very likely because there are a lot of different ways to get there (or not get there). But given what I’ve seen so far, I assign this path the highest concentration of the admittedly diffuse conditional probability mass.
As I discussed in this previous post, I don’t think we’ll achieve AGI until we’re pretty far down the road. Based on what I heard from Vinge, Rattner, and Gershenfeld, I am reasonably convinced that our everyday environment will become increasingly electronic. Lots of everyday things will be connected to the network and imbued with significant amounts of computing power. After a while, these items will even be able to adapt their physical as well as virtual properties in response to remote instructions.
So far, nothing earth shattering or even controversial. However, I think people dramatically underestimate how big the draw of being able to richly interact with this environment will be. Let’s face it, our current user interface technology sucks. Using a computer today is pretty much the same as it was 20 years ago. Up until the iPhone, it was a total pain in the ass to interact with anything that wasn’t a full computer. Yet many people contorted their brains and digits to do whatever it took.
Even my beloved iPhone isn’t that great in an absolute sense. It just sucks much less from a relative standpoint. But it’s still pretty hard for me to do even relatively simple things, like tell my iPhone to instruct my TiVo to record a show. Imagine how frustrating it will be when you could interact electronically with 10%, 33%, or 50% of all the discrete objects in your life. You would wield magic-like powers if only you could make your wishes known!
Therefore, my first prediction is that you will see an extremely frothy market for enhanced user interface technology. People are going to try some crazy stuff. Most of it won’t work. But Bluetooth headsets are just the beginning of a trend towards adorning ourselves with electronic control devices.
Ultimately, of course, brain-computer interfaces (BCIs) are the solution. Before the summit, I thought we were a long way from BCIs. Too difficult, too brittle, too invasive. I saw several things that, when combined with my new found appreciation for the pressure to interact, changed my mind. First, from Rattner and Gershenfeld, I got a gut-level appreciation of just how small, powerful, and flexible our electronics will be in the near future. This factor ameliorates invasiveness and brittleness.
On difficulty front, I was impressed with what Kurzweil and Modha said about advances in simulating neurons. As I made clear in my AGI post, I don’t think we’re any closer to executive function [Note: via Robin Hanson, new report on whole brain emulation here], but we don’t need that for BCIs. The majory difficulty with BCIs is interpreting what the signal coming out of neurons means. However, if you can simulate the neurons in that region of the brain, you can probably do a much better job of calibrating your sensing apparatus to the intentions of the user. Of course, the brain itself will adapt and you’ll hopefully get a tightly converging loop of adaptation. Obviously, there’s a lot of handwaving going on here, but I belive the inuition that simulation will give us interpretation leverage is sound.
Now, we still need to find early adopters for the first BCIs. Even though I am a long time sci-fi buff, I’ve always had problems coming to grips with the idea of sticking electronics in my head. However, people a few years younger than I seem to have no problem permanently altering their bodies with tattoos and piercings, so their threshold to adoption may be much lower.
We’ve also, unfortunately, been creating a significant class of potential test subjects with forays into Iraq and Afghanistan. Young men and women are losing not just limbs but whole areas of brain function through traumatic brain injuries. Perhaps more than 30% of casualties have some sort of brain injury. My guess is that the prospect of restoring lost function will make them more than willing to try electronic cognitive prostheses. I know it would for me. And I’d certainly contribute money in a second to help. I can easily imagine this situation overcoming taboos against connecting electronics directly to the human brain.
I realize this was always on the technoogy roadmap. But now I’ve got it on my 10-year horizon instead of my 20-year horizon.