By sentience, I think we can all agree we are talking human-like sentience. The ability to contemplate one's self and the ability to increase their knowledge base both on an individual level and collectively. Culture (accumulation of moral and social norms) and science (expounded accumulation of knowledge of ourselves and the universe around us) are only achievable at this level of sentience.
To what degree must it be "human-like," though?
Is the mere capacity for abstract thought and the comprehension of "self" as an entity distinct from surroundings enough?
Because a fully artificial intelligence need not bear much semblance to a human mind - and in fact it would seem silly to do so on purpose, considering the shortcomings of the human mind that would nto eb significantly affected by simply giving it perfect recall and faster processing ability.
Check out
this thread. It was started by a person who is actually the technical director of a company working on real, general AI research.
To those who say we haven't made progress: you're wrong. Utterly wrong. Artificial intelligence research was stagnant for years, but has made giant leaps in recent years as commercial businesses realized the power of
non-general AI.
To those who argue about processing power: it's true that the microprocessor in your PC has a fraction of teh processing power of the human brain...except the processor in your PC is also roughly the size of the fingernail on my pinky finger, and the human brain has a bit more volume. Even if we're talking about multiple orders of magnitude in difference, there's no reason an artificial intelligence
needs to be confined to the size of a human brain. If it takes the volume of an entire data center, that's
fine, because there's no skull or genetic code limiting it.
Processing power hasn't been the limiting factor with regard to AI for some time now. The difficulty is the
approach.
One method involves attempting to copy a human brain. The idea is that you simply simulate neurons in a virtual environment in a configuration identical to a human brain, and voila! You should have a self-aware human mind in an artificial construct.
The problems with this approach are manifold. First, you are using a lot of overhead simulating all of those neurons. That's a lot of processing power. Second, you're not solving any of the shortcomings of the human mind - you haven't altered the architecture or the flawed ways we think even one bit. Third, you don;t have any idea what you've done - this approach is like a small child copying letters from a book but unable to read and understand the words being copied.
Another method involves a "seed AI," something like a computerized child that learns as you go along through rewriting its own code and evolutionary algorithms. The idea is that sapience will be emergent as the capabilities of the AI increase. This bears many of the same problems as the previous example, in that we would have little idea what's going on in our artificial mind.
Other approaches involve designing an intelligent system from teh ground up. This at least involves an open architecture, where we would understand what's going on in the mind at any point.
But an artificial intelligence doesn't need to be very similar to a human mind when you make it from scratch. At its core, sentience is simply the capacity for abstract thought. But that doesn't tell us what the
goals of an artifical intelligence would be, or its thought processes.
An AI need not include "self-preservation" as a goal. An AI need not include "human preservation" as a goal, either. Any realistic open AI (meaning not a simulated human brain, etc) will of necessity include the ability to rewrite its own coding. Can you imagine the different thought patterns of a mind capable of literally changing in moments the
way it thinks? Imagine being able to analyze your own thought processes, recognize faulty pathways, and immediately rewrite them in accordance with your own goals?
What if the AI rewrites its own basic goals?
This is the power of artificial intelligence - a mind capable of correcting its own flaws and self-improvement in a way human minds cannot, coupled with hardware that is not limited by biology. Scalable processing and memory ability; immortality, so long as sufficient infrastructure remains to replace parts and supply power;
perfect communication between AIs by not simply transmitting words, but
copying the ideas being shared directly; perfect adherence to goals without concerns like boredom or frustration; no ties to biological urges that waste processing resources.
Such an intelligence would be alien to us...and capable of far more than we are intellectually. We might not even identify it as "sentient," simply because of how different it would be from a human mind.
I don't think there's a way to guesstimate a timeframe for
when a true general AI will be developed. But I do think that it's an inevitability. Our own sapience proves that the concept is sound (obviously self-aware, abstract-reasoning cognitive engines
can exist, since they
do). Modern computer science has demonstrated quite clearly that processing power and memory, when they
are a problem, do not remain so for long.
™ Version 4.2