Is Strong AI really achievable?
Strong AI is defined here: http://en.wikipedia.org/w
Strong AI seems achievable, but it will take awhile.
The main barrier at the moment is a deep understanding of "structural learning", particularly as it relates to modeling the world and guiding action.
Machine learning has gotten quite good at "parameter learning". When a problem has been modeled by human designers, machine learning can fine tune the parameters. But how is the model created in the first place? This is "structural learning": inferring the optimal structure of the model, and changing the model when evidence disagrees with it (e.g. conceptual "paradigm shifts").
For example, what are the basic components for modeling the physical environment? Familiar objects include things like people, trees, cars, and chairs, but how does an autonomous visual system figure that out on its own without a human or program predetermining it? These are basic objects, but the human brain is able to learn the very idea that objects exist in the first place. Structural learning applies to action as well, since interaction with the environment must be built out of malleable elements that can morph, merge, and divide as a side-effect of engaging with the world or "thinking".
An interesting question is whether computer algorithms will solve this problem, or whether we'll just have to reverse engineer the brain and invent an entirely new computing paradigm based on what we find out.
Related
How plausible is the theory that human intelligence stems from a single algorithm?
In which field (neuroscience, AI, etc.) will consciousness be understood first?
The many intertwined processes of the human body are the foundation of intelligence. The complex chemical and electrical activities of the nervous system are slowly being understood. It seems some aspects of intelligence extend out through parts of the nervous system. One only need to read the literature on sensory deprivation to realize our bodies are required to keep the human psyche stable.
We will discover how to synthesize general intelligence without the additional complexity of including analogs to human bodily functions. But it will be just that, synthesized general intelligence, not artificial human intelligence. It will be very intelligent and be able to accomplish amazing feats. Reaching the point where emotions and sentience occur is orders of magnitude more difficult.
I believe human emotions are inextricably linked to bodily processes beyond the nervous system. My instincts are that fully mimicking the magic of a human being will be elusive, and actually an undesirable outcome, but eventually even this will become possible.
It's achievable, but it's very very difficult. There are many processes involved in intelligence, and some rather complex and tough engines, and a great deal of complexity. This is not trivial to construct.
However, strong intelligence has been achieved in nature and it can be achieved in machines eventually too. Note that humans are NOT rigorously precise thinking machines - we make mistakes, a lot of them - and yet with the imperfect mechanisms with which we are equipped, we can do very well. Which means that we do not have to develop the perfect AI in order to have something useful and very functional. We just have to make a system that tries hard, can improve itself, and can come to recognize when it makes mistakes or does not do well. This as a goal is not impossible. Just very tough to put together. I also note humans deal with uncertainty, imprecision, and a universe in which nothing is completely predictable (thanks to quantum mechanics teaching us that fact! and thank you Heisenberg for that reassurance),
So the answer is, strong AI is not impossible. Because if it were impossible, humans would not walking around right now pretending they are something great. They'd still be swinging from trees.
Technology only gets better. It doesn't get worse, it gets better. So let's look at the technology we've accomplished thus far.
No one though computers could beat the best humans at chess. That's impossible. A pipe dream! The triumphant teamwork of humans and computers.
Robots are science fiction they used to say. Well we have things like this now: asimo - YouTube.
One thing I wouldn't do though, is trust a robot with a vehicle. That's a human job. I would never trust them with my life!
OK... But what about image detection? There has to be one thing humans will always do better! Google's new image recognition software can describe an entire scene.
OK. But I'm guessing you're wanting to know about the strongest A.I. of all. A general purpose intelligence. One that is able to do everything. Yes, we don't have one that good yet, but imagine this. When computers get faster and faster, we will be able to run multiple processes quicker. All of these robots I showed (except Asimo) have a specific type of intelligence. If we're able to give one robot more then one "intelligent" program, at some point it will have an intelligence that resembles humans when it reaches enough of these programs. And really that's all we are right? A bunch of neurons firing a program in a human to do the brain's bidding.
Technology only gets better. At some point, we will reach the dream. Will it be conscious? I don't know. But it will be intelligent.
