Visions of the Singualrity 3
Tuesday, March 23, 2010 at 5:49PM We have now covered what it means to be human, and some ways that humans could become exponentially smarter or hive-minded. But what about the possibility of computers or other machines becoming intelligent all on their own?
As we have advanced artificial intelligence over the years, and understood our own intellect in the process, we have further and further relegated our own gifts to the mundane, as universal intelligence has become the only factor that makes us the real smarties in the room. Most, if not all, of our core intelligences have been conquered separately by AI:
The list could be endless, but the ending truth is that all of these things are limited only by current processing speeds. Even the fact that no one AI entity can do everything we can is ultimately a question of speed. Eventually, even if no more ground is broken into how our brain works, machines will be able to brute-force simulate us.
Is simulation the same as true Machine Intelligence? Strictly, yes. But this path leads back to the same things we talked about last time. More interestingly is the fact that AI should be able to out-performs us even quicker than we can simulate ourselves.
Humans ultimately do a lot of things the hard way. Our brains are not set up to crunch numbers quickly, but computers are. A well-constructed universal AI would draw from all disciplines of the Artificial Intelligence community. From Perceptrons to Genetic Programming to Finite State Machines and Expert Systems; and an ideal system would grow and change and find new ways to solve old problems. It would compile tried and true operation for speed and employ reduction routines to weed out magic. (By magic, I mean aspects of code that don't actually make the process better, they were simply there because they were when the process was first evolved. Like prayer in the creation of a Katana.)
How does this all come together?
There are a few clear paths to human enhancement; which is not true of machine intelligence. It isn't even clear what kinds of AI will be the backbone of the first universally intelligence machines. However, I see a couple more likely avenues.
One thrilling option is the creation of competing entities in a physical or virtual space. Through some genetic process, such as Genetic Programming, these entities would each be given random code to control their various senses and manipulators, the best survivors would be mixed and mutated to create new entities in a new world and over time intelligence may arise in them. Given access to various additional schemes, such as facial recognition engines, natural language dictation and decoding, and other tools; this AI could learn to recognize objects by name and shape, even individual people, and converse with them. I like this path because it doesn't really require any new kinds of AI or even enhancements to our existing techniques. It simply requires horsepower.
Another option is through Perceptrons, which can become more numerous and speedy in their simulation of the actions of neurons. There are holes here, however, as the learning and usage phases of perceptrons are strictly separated. Someone must initiate a training process for these systems to be able to learn a task, and then that task can be performed. Imagination Engines begin to show more promise, but they also have lengthy and tightly controlled training phases. To be a viable avenue to real universal AI, it will need to be able to self regulate this training. It would need to be able to self-guide.
Undoubtedly there are more paths to AI enlightenment, it is only a matter of time before one is born. Once a single intelligent machine exists, more will quickly follow. They will become smarter at an exponetial rate and we will either have to join them, or be left behind in the evolutionary arms race.
One wonders if they would tolerate us? This is the scary stuff of The Terminator, or Blade Runner or The Matrix; but does it have to be that way? Could we co-exist?
Reader Comments