“Maybe we need to know ourselves a little better before we are able to successfully extrapolate how AI would behave—and the risks involved.”—Leo F. Fernandes
Blue Boy. Quasi, an animatronic robot designed by grad students at Carnegie Mellon University, shows a “sad” face. (Courtesy IEEE Spectrum and Tom Altany)
Stimulated by the comments of
Kurzweil and my own underlying interests in supercomputers and AI, I spent the
last few days pondering AI. In its most primitive form it can be thought of as
simple entertainment robots like the Korean girl robot EveR-1, the Japanese
robots, the animals and computer game opponents and teammates; but they all
have a little something wanting.
The idea of AI has been with us for a very long time. When
machines first exhibited what was considered smart behavior, people thought
they were real, proving once again, to an ignorant people, your technology
seems like magic. Manikins and puppets were made in the 1800s that told your
fortune and played cards, the piano, and even chess (in this case the “robot”
was later found to have a man inside).
When computers started getting affordable and smaller back
in the 1970s with the introduction of the mini-computer, Marvin Minski
predicted AI would be realized by 1990. He later changed his view, but others
AI is actually all around us, and what it is is a debate of
definition. For the most part, today AI is a decision tree—if this, then that,
etc. In the case of computer games AI, there are two levels: the hierarchy
where you deal with character-specific and gamewide behavior, and the lower
level where you deal with path, collision avoidance, terrain analysis, line of
sight (for aiming weapons and getting hit by shots), and interaction like talk.
AMD and Intel intend to promote AI when they come out with
their quad-core processors, but a CPU will only be good at—useful for, that
is—the hierarchy of AI. The lower level stuff will be better served by a
multi-multi-processor array that can make zillions of quick decisions, albeit
low bit size—much like our neurons.
AI also provokes huge discussions and debates about what is
it, how would we (humans) recognize it (the Turning test gets mentioned a lot),
and how can we prove we are not AI?
Isaac Asimov articulated a lot of concerns around AI in his Robot series. He
also dealt with the idea that AI will stop being a slave and become our master
or executioner. If you don’t feel like reading a book Will Smith in I
Robot will get the point across
At the IEEE Spectrum Leo F. Fernandes suggested, “Maybe we
need to know ourselves a little better before we are able to successfully
extrapolate how AI would behave—and the risks involved.”
One of the most interesting games being played today is Ghost
Recon Advanced Warrior (GRAW). It has
amazing models, great graphics, physics, and up to three soldiers you can
command. It also has deadly enemies. Your soldiers and the enemies are AI
characters. Unfortunately, the AI behind them is very disappointing. Ted
Pollak, our game analyst, says one of the reasons online team gaming is so
popular is because you don’t have to deal with stupid AI—it’s more
un-predictable and therefore more lifelike.
Honda’s ASIMO walking robot. (Courtesy Honda)
Nonetheless, AI will be pushed into prominence probably late
this year and onward. AI is not something that a GPU can do, so you won’t see
the same types of wars as we’re seeing fought now over GPUs vs. PPUs. But,
because the CPU can do the upper level, and has been doing the lower level by
default, you can expect to see a CPU–AIP war. As for an AI processor, they are
coming and we’ll tell you all about them when our cats are returned. (These
NDAs are getting a lot tougher.)
The glorious thing is we do have an exponential growth in
compute power, and in particular in the PC. The CPUs are going to quad-cores,
the GPUs are getting more powerful and can be stacked up, we have a dedicated
PPU, and soon a dedicated AIP, and, much like a human, all those processors
will be running in parallel, doing specific tasks and delivering an amazing
experience to the consumer.
With well-written games (a big if), and all the environmental
processing that will be available, I think the prospects of consumers playing
games, being in the movie if you will, will create a new type of entertainment
that will rival TV and yourtube/myspace/second life.
Environmentally accurate games won’t be limited to FPSs,
they will become the ultimate fantasy stories, giving us a 2D VR experience
that our brains will convert into a 3D feeling. And then we’ll find ourselves
asking the characters in the stories, Are you AI?
For more reading on the subject, visit these sites:
http://www.a-i.com/show_tree.asp?id=1 (You can speak to AI at this site.)