Comment on "The Human Costs of AI"

By Lary Walker

Comment on the post on this page:

I think the above two paragraphs are generally true. Personally, I have strong reservations about the fact that AI seems to have moved to a focus on 'machine learning' -- which is driven by algorithms. My concern is that NO ONE knows why the computer reaches the conclusions it does by analyzing vast amounts of data.

My company did Expert Systems which focused on capturing the detailed knowledge of experts and leading less expert users to follow that detailed process. My best programmer was convinced that if the expert had a process, we could capture it and lead others through it.

The AI world abandoned expert systems because they could not figure out how to extract this detailed knowledge from the experts. My company did NOT have this problem. We invented 'reflective knowledge engineering' which interacted on a continuous basis with the expert -- essentially using the expert to on a weekly basis to guide us to a path that reflected what he or she actually did.

Our experience with this process was that the experts got excited about it -- because -- for the first time, they could see their own expertise flow in action!!

Machine learning relies on data -- not expertise and experience. I am concerned that by analyzing data, it may be difficult to extract the process that created the data. Also, in our work, we discovered the issue mentioned in your paragraphs, i.e. how to cross from one domain of focus to another. Our expert systems were single purpose solutions.

We have thought about how to 'translate' terms/data to work across differing domains, but certainly do not feel we know how to do that yet.