US Air Force says decision-making attack drones will be here by 2047
engadget.com
engadget.com
It's plenty likely that by 2047 we'll not only have fully autonomous combat aircraft but we'll have automated aircraft fabricators and aircraft fabricator fabricators.
It's almost as valid to say that we'll be fighting sentient robots in hand-to-hand combat in a post-apocalyptic world in 2047.
Hey, perhaps those are the decision-making drones they were talking about!
That's BDUF at its worst. Better to look at the technology needs of today while planning for flexibility and be able to iterate rapidly if we see a need for change down the road. If the US had locked itself into some grand technology road map in 1909 it would have most assuredly lost WWII because it could not have anticipated the technological needs of that war.
I'd guess we'll see this between 2018-2025. The necessary technology is mostly here, the main obstacles to implementation are ethical and political. I would prefer not to see autonomous weapon platforms at all, but suspect that the tactical advantage they confer will be too great to ignore, especially when set against the lives of service personnel (of the side with the technology ready to go).
Before ten years pass, I die, the mule dies or the sultan dies... </quote> http://nasredin.blogspot.com/2007/11/ibn-khaldouns-mule.html
taken from http://news.ycombinator.com/item?id=411695
http://www.amazon.com/gp/product/0393331660/ (no affiliate link)
On an unrelated note, it's worth reading the general summaries of Ibn Khaldun's works too. A lot of things we take for granted now but were incredibly advanced back then, as well as some good observations generally on how to think and consider things. It'd appeal to the sort of man who generally likes Adam Smith, Carl von Clauswitz, Plato, etc.
I think Ray Kurweil has a good point in saying that every technical specialist has trouble extrapolating for the increase in the power of the tools that the specialty will soon be working with. That applies to programmers as much as other engineers.
When programmers start working with a machine having the level of complexity of the human brain itself, our present ideas about the difficulty of human-equivalent AI may turn out to entirely erroneous. After all, we certainly don't have such a machine to practice on today.
Human-level AI is an extremely complex problem and I wouldn't claim know everything that might be involved in creating it. But by token, refuting any possibility of AI is an equally complex problem. Neither "Clear on observation" conclusions nor the other haphazard observations of this paper cut it for such an analysis.
What is haphazard about the rest of it? If you can point out specifics it'd help me strengthen the work.
I've seen plenty of programmers underestimating the problem of AGI as well!
...though for every one that underestimates it, there are plenty that overestimate it.
Solving this problem doesn't necessarily mean that we do anything at all clever; it really may just boil down to having decent brain scanning technology and lightning fast processors to run the simulations. We'll definitely have the computers, in time, so the only question is whether brain scanning improves. If that's what it comes down to, programmers and computer scientists are pretty much irrelevant, and the real heroes will be the guys doing medical tool development.
Granted, it would be deeply unsatisfying to "solve" the problem by blindly copying nature. But there's a very real possibility that it's going to turn out to be our first viable solution to the problem, and let's be honest: it's not going to be very difficult to do if the enabling technologies are already in place.
The image that jumped to my mind, was the wing flapping airplane prototypes that, quite literally, never got off the ground, and that we all laugh at today.
Perhaps, in the end, our solution won't match natures.
That also leads me to another thought, which is the definition of the problem.
Going back to the airplane. If the problem had been defined as "to build a machine with birdlike agility and control", than we still wouldn't be there. Instead of patting ourselves on the backs, we would still be miserably failing, over 100 years later. But we didn't really need birdlike agility, we just wanted to fly, which, it turns out, is much easier.
The problem statement for a lot of AI research seems similarly ambitious. Do we really want to create some sentient being, trapped in a box? Or is what we're really after, a little more practical and obtainable?
I think folks like the military, get this. They have a definite objective that they are trying to accomplish. Autonomous desert navigation, autonomous urban navigation, autonomous enemy identification. These are things you can define, measure, and make progress on.
However several papers and articles I read time ago, reported that the DoD was searching for advanced methods (read AI-like) to recognize enemy vehicles (and personnel). Unfortunately I can't find the links.