Posted by alexandra_k on June 15, 2005, at 16:05:06
In reply to Re: step in the right direction, posted by Dr. Bob on June 15, 2005, at 0:39:59
> > I have yet found no basis for a standard I can describe in terms that can be compiled into machine language.
> That's the sticky wicket...You could bypass the programming issue if you had yourself a big enough (PDP) / connectionist network. You feed in a post - and request a civility determination. The net delivers its verdict. If it delivers the wrong verdict then there are statistical algorithms for adjusting the weightings on the network so it should deliver the correct result. If it delivers the correct verdict then you move onto the next post / sentence (?). Feeding in the posts and adjusting the weightings counts as 'training up' the network.
It works quite well as a rock / mine detector :-)
And I think people are having a lot of success with face recognition.It would probably be a mammoth project just to get it correctly determining grossly uncivil posts. The success of such a machine would be partly dependent on the trainer who decides whether the machine is delivering the correct result or not. Over time it might get quite good.
It might handle all the cases that we consider 'clear' so well that we are willing to go along with it on cases that are less clear.But I don't see how that process would help us understand the civility rules. I mean, we could make a PDP net that delivered the same 'answers' as a person would. But that doesn't help us get any clearer on *how* a person is able to deliver the answer. Except that the brain is rather like a PDP net. But we knew that already...
Alternatively...
I was serious with my suggestion to see whether there are different kinds of uncivil behaviour. I can't find the post at the moment...
But the notion is that if you can figure that out then it helps clarify the concept of civility. You need a clear concept / theory BEFORE you are going to have any success with programming. And a nice measure of clarity of concept / theory is whether it is programable.The hard bits would be in getting the program to recognise new instances of a type of uncivil behaviour. Neural nets have more success with recognition than traditional programs.
Another difficulty is that such a machine wouldn't be able to give an explanation of its verdict. Dr Bob fares a little better in this respect ;-).
If there were a standard explanation as to why each type of uncivil behaviour is considered uncivil then the machine could deliver this as a reason.
It would be useful to have a brief summary as to how the instance was interpreted to be an instance of that particular kind of uncivil behaviour as well...
That might be trickier. Especially when the machine doesn't really know what it is doing...
Hmm
I sort of thought that coming up with different kinds of incivility could give certain people something useful to do with respect to searching the archives and helping everyone better understand the civility rules...
This might fare better than frequent requests for determination???
???
poster:alexandra_k
thread:511186
URL: http://www.dr-bob.org/babble/admin/20050614/msgs/513223.html