Where is Frisco?
ORIGINAL: pompack
ORIGINAL: dinsdale
Do you think it's been easy to bring each new generation of graphics breakthroughs? Even before getting to software, do you think it's been easy moulding this generation of hardware?
Easy isn't relevant, if it were then there would only be "simple" discoveries. The problem isn't the difficulty, it's the lack of resources spent in R&D.
precisely but....
The problem is that real AI still requires a LOT of research, and not just applied research. A few years ago I was cognizant of (but no I did NOT do any of the real work) a number of large and well funded research projects into AI applications.
There are any number of extremely important tasks that require the judgment of highly trained human beings. Some of these tasks are also repetitive. Some of the tasks are also stressful. (just for example, although it was not one of the tasks, consider an air traffic controller.) The difficulty of finding highly intelligent people who are willing to be trained to do repetitive work that requires critical judgment is extreme. Since such people are also expensive to train and have a tendency to quit. This makes it even more difficult. The task of developing AI software to do some of this class of job was recognized to require some fundamental research and a LOT of expensive applications work. However the payoff was deemed to be worth it and this was the government, so several were funded (and funded into the tens of millions).
Not one of those projects ran to completion. The issue was that if you have a task important enough to warrant spending that much money to automate, it has to be RELIABLE. And it wasn’t. Things would work well for a while and then some utterly irrational result would pop up. Some fundamental theoretical breakthroughs were required, and the funding agencies were just not interesting in an open-ended research project. Instead any further work that took place was aimed at providing tools for the human to increase his productivity.
The point is that AI has been one of those great concepts that has just not gone anywhere very fast. If your task is relatively forgiving like producing a game opponent, it’s not worth spending the money it would take to develop it. If the task is to automate some critical human-judgment function, it’s not reliable without some basic research breakthroughs that can’t get major funding. Such research is taking place, in academia, but at a very slow pace.
If any of those big projects panned out, the fallout would surely have made game AI more practical. Conversely if anyone just sat down and spent the money to produce a true AI for a forgiving environment, they might go bankrupt but the rest of us would certainly benefit from it.
