Posted by
michael
on from the dreaming-the-future dept.
Anonymous Coward writes: "Ray Kurzweil and other digerati discuss when popular sci-fi concepts will manifest in the real world. See part I or
part II."
> Concept: The ability for artificially intelligent devices to feel emotions.
It's not at all obvious -- to me at least -- that we should want AIs to feel emotions. Who wants a warehouse full of smart bombs with hurt feelings?
Emotions can very clearly lead to inappropriate behavior. Granted, there may be times when emotions lead to positive behavior. But do they ever lead to positive behavior that couldn't be programmed into an AI without emotions? Unless that's the case then emotions are something known to be dangerous and not known to be useful, and therefore should be avoided as a life-threatening bug.
Granted, it may be a fact of nature that "intelligence" (whatever that is) is impossible without emotions. But unless/until that has been demonstrated, let's keep emotions off our wish list.
Now back to Part I:
> Concept: The idea of a computer becoming so complex it can understand, reason, listen, speak and interact in the same way as a human, including using deception and self-deception.
> Now we have: Machines that learn, software that breeds/replicates. 'Narrow AI,' i.e. computers that can perform 'narrow' tasks that previously could only be accomplished by human intelligence, such as playing games (e.g. chess) at master levels, diagnosing electrocardiograms and blood cell images, making financial investment decisions, landing jet planes, guiding cruise missiles, solving mathematical problems and so on. Currently exponential progress curve showing no sign of slowing down.
First, as with emotions, I dispute the desirability of AI agents that can knowingly deceive themselves and others.
Second, I'm not convinced that much of the laundry list in the second paragraph qualifies as "intelligence" instead of merely "appropriate algorithms". (Are we going to have to call MATLAB an intelligent agent because it's good at certain kinds of math problems?)
Third, I am amazed that they would say that we're making "exponential progress" in anything that might reasonably be called "AI". My games don't seem to ship with AIs that are "exponentially" smarter than the ones that shipped five years ago. Dish up some facts, please!
That said, here's a link to a paper [160K PDF] that someone turned me on to recently. It's from a talk some AI researchers gave at a conference last year. They start by asking where is all the cool movie-style AI, and answer with the observation that no one is working on it. Their proposal to remedy that situation is that AI researchers should get involved in game AI, because many modern games require agents that are more "intelligent" than the common solve-one-problem stuff that has been coming out of the AI community for the last few... decades.
I think the authors of that paper overstate their case by calling game AI agents "human level" AI, but at least it's a step in the right direction. It's a bit of a light-weight article, but it's an easy read. And it would be way nice if 2/3 of the world's academic AI researchers started working on gaming applications!
Slashdot Mirror
First, from Part II:
> Concept: The ability for artificially intelligent devices to feel emotions.
It's not at all obvious -- to me at least -- that we should want AIs to feel emotions. Who wants a warehouse full of smart bombs with hurt feelings?
Emotions can very clearly lead to inappropriate behavior. Granted, there may be times when emotions lead to positive behavior. But do they ever lead to positive behavior that couldn't be programmed into an AI without emotions? Unless that's the case then emotions are something known to be dangerous and not known to be useful, and therefore should be avoided as a life-threatening bug.
Granted, it may be a fact of nature that "intelligence" (whatever that is) is impossible without emotions. But unless/until that has been demonstrated, let's keep emotions off our wish list.
Now back to Part I:
> Concept: The idea of a computer becoming so complex it can understand, reason, listen, speak and interact in the same way as a human, including using deception and self-deception.
> Now we have: Machines that learn, software that breeds/replicates. 'Narrow AI,' i.e. computers that can perform 'narrow' tasks that previously could only be accomplished by human intelligence, such as playing games (e.g. chess) at master levels, diagnosing electrocardiograms and blood cell images, making financial investment decisions, landing jet planes, guiding cruise missiles, solving mathematical problems and so on. Currently exponential progress curve showing no sign of slowing down.
First, as with emotions, I dispute the desirability of AI agents that can knowingly deceive themselves and others.
Second, I'm not convinced that much of the laundry list in the second paragraph qualifies as "intelligence" instead of merely "appropriate algorithms". (Are we going to have to call MATLAB an intelligent agent because it's good at certain kinds of math problems?)
Third, I am amazed that they would say that we're making "exponential progress" in anything that might reasonably be called "AI". My games don't seem to ship with AIs that are "exponentially" smarter than the ones that shipped five years ago. Dish up some facts, please!
That said, here's a link to a paper [160K PDF] that someone turned me on to recently. It's from a talk some AI researchers gave at a conference last year. They start by asking where is all the cool movie-style AI, and answer with the observation that no one is working on it. Their proposal to remedy that situation is that AI researchers should get involved in game AI, because many modern games require agents that are more "intelligent" than the common solve-one-problem stuff that has been coming out of the AI community for the last few... decades.
I think the authors of that paper overstate their case by calling game AI agents "human level" AI, but at least it's a step in the right direction. It's a bit of a light-weight article, but it's an easy read. And it would be way nice if 2/3 of the world's academic AI researchers started working on gaming applications!
Sheesh, evil *and* a jerk. -- Jade