posted on Dec, 20 2016 @ 01:43 AM
a reply to: jadedANDcynical
Regards expert systems and the reason the interviewer skills are so important. This would be because the interviewer, in determining which questions
to ask, is the one who would be the limiting factor as to the performance of the ES. It is the interviewers job to ask the questions and develop the
circumstances under which the decisions the ES is to be responsible for. If the interviewer does not ask the right questions or determine the correct
range of responses then the ES cannot respond according to reality compared to what the interviewer encompassed. The more accurate questions/scenarios
the interviewer is able to outline, the better suited the ES will be to it's task.
Exactly. The best "interviewers" for a specific expert system are people with knowledge in that domain as they have a much better sense of which
questions to ask and how to ask them. But barring that, there are folks that are just good at this and they will start at very basic info and then
walk through a typical scenario with the expert to determine what the expert is looking for, and then what the experts response would be once they had
the info they needed. The more specific the domain the better the ES is. For instance there isn't a real "Dr. in a box" ES system out there but
there are some very good ones within specific fields of medicine like Oncology, or Nephrology to name a few.
What I'm getting out of this is that there are times an AI will reason out a solution and then remember the steps it used to produce that solution.
The next step would be for the machine to use the same steps to solve similar problem not related to the original thought processes. It may improvise
or add on to the prior reasoning solution and in that way show transfer of learning; evolution?
Yes! The best ML systems are those that remember both the solution and the path it took to get to that solution. It drastically reduces the amount
of time spent on similar problems and that "learning" often can be applied in multiple contexts. These types of systems are notoriously difficult to
build, and we often fail. The damndest thing is that we seem to be unable to do what we want the systems we build to do - learn from our successes
and our failures. When it comes to designing and building neural network systems we often don't know why system A succeeded and system B failed. And
to make things even more interesting, rebuilding system A doesn't always work as intended either. It's bloody maddening at times...
But we're working on it. And quantum computing (HW) will be what leads to the singularity. There is absolutely no doubt in my mind about that. And
we're very, very close. Closer than I can publicly talk about. Forget D-Wave (or what will be it's commercial product)...there are things light
years beyond it safely under Darpa's umbrella.
Watch this space...
edit on 12/20/2016 by Riffrafter because: (no reason given)