Indeed. It doesn't "learn" in the sense humans do. It has to be told it is wrong, and in what way, then programmed not to be wrong next time.
I'm not aware of it having any ability to be told it's wrong and then figuring out why and correcting itself. I remember reading that, for any question it has to answer, it builds a list of likely answers, arranged by a confidence number. If it's told one of them is wrong, the confidence number of that answer would become 0 and it would never use that answer again. But the other answers could also all be wrong!
Robert Maxwell wrote:
Flying spaceships is actually pretty easy for computers, because it involves fairly straightforward physics calculations and a limited number of decisions to make (e.g. which engines to fire at which time to reach point B.) A spacecraft autopilot is quite a bit less complicated than a system like Watson, so it's a bad example.
1. Expert systems can learn more information based on the rules and associations they already have. Depending on how fuzzy its logic is, it could have relations like the following:
* All mammals have hair.
* All mammals are animals.
* All dogs are mammals.
* Yorkshire terrier is a breed of dog.
* A breed is a subcategory/type.
* All Yorkshire terriers are dogs, mammals, animals, and have hair.
That's a very simple example, but it's relations like these that make up the knowledge of a system like Watson.
2. That's a poor question because computers interact with humans in all kinds of ways. If you mean carrying on a conversation with humans, well, programs like ELIZA remain perennially popular despite not being all that advanced. They are just human enough
to fake it. Alice
is a more recent example that's good at conversing with humans, albeit in text form. Speech recognition is pretty good these days, so it wouldn't be hard to make a talkie version.
Computational speed is less relevant than memory access strategies. Watson's uniqueness is in the vast amount of information it can sort through quickly, not how fast its CPU is. It is far more data-bound. So, that means very fast (and ample) memory, with fast CPUs being of secondary importance for its purposes.
Thanks Robert Maxwell
for the information. In the docu-fiction Alien Planet
the probes investigating the planet are claimed to have the 'intelligence,' of a 4 year old. Does Watson and systems like it possess that level of intelligence?
And could a system like Watson - or even Watson itself - with enough programming successfully investigate a planet with little or no human intervention?
In otherwords - do we possess the technology today to send a probe to an alien planet that would run mostly autonomous of human intervention?
Today, NASA sends semi-autonomous probes to other planets. The Mars rovers cannot be controlled in real-time, for instance. Instead, NASA sends them instructions to be executed--like "go to this location and sample the soil"--and the rover figures out how to get there and deals with any obstacles along the way. It makes its own decisions about how to best achieve its objectives, given its equipment and circumstances.
But you have to realize that this does not constitute "intelligence," in the sense of having a will and being able to make totally independent decisions based on accumulated experiential knowledge.