The future of artificial intelligence
Vinod Khosla wrote an interesting series of articles on artificial intelligence and its impacts in healthcare (Do we need doctors or algorithms?) and education (Do we need teachers or algorithms?) for TechCrunch in January. My PhD research touched on expert systems and artificial intelligence, and my early interest in knowledge management was centered around interesting technological capabilities that sprang from AI. As a result, my eyes have always caught on articles on the topic of AI, even if it isn't directly what I do these days.
As these articles and other retrospectives on Artificial intelligence say, the founding technologies and capabilities of AI have branched into many aspects of our lives today. Most articles lament that these technologies are so common or familiar that AI doesn't get credit: voice recognition, sophisticated data analyses, interactive voice response phone systems, wired cars, awesome gaming systems, and a lot of elements of why we love the Internet. The one area that usually still gets credit as "being AI" are the computers that can play games against human masters: most famously, IBM's Deep Blue and Watson.
Khosla focuses on two areas of our lives where AI doesn't seem to have made as many inroads: healthcare delivery and education. There are certainly aspects of AI that have shown up in both of these fields. But Khosla's discussion is on the primary interaction of doctor-patient and teacher-student: these interactions remain a familiar model with very little in the way of technology support. He argues that we have the tools around us to make these interactions very different. Rather than asking routine questions (over and over again) for routine diagnoses, why not take advantage of in-home capabilities to get improved care. While it sounds like taking the doctor out of the equation, I don't see it as that at all. It is one of the ideals behind AI (and technology in general): replace the routine and simple activities so that we can focus our expertise on the usual situations. Similarly in education, we have plenty of adaptive systems available out there that can gauge the skill level / expertise of people interacting with them: why not make use of them to help customize development of skills and knowledge of our students? Don't remove teachers, but give teachers the ability to focus on the children with the right timing and tools.
An AI anecdote: I remember reading about the massive efforts to build and train Cyc in the mid 1990's. One of the funny things that came out in the early days was the funny fact that "all dead people are famous." Cyc had been trained with a lot of specific information about important or famous people throughout the ages. It had probably been given statistics about population trends as well. But its training hadn't included baseline information that important people are drawn from that larger population.
This leads me to think about the larger conversation of experts and expertise. It is probably beneficial to businesses and society to have experts focus on those areas for which they are uniquely qualified. But in order for them to get that qualification, they need the baseline training and experience and background to get there. If we can automate away a lot of the baseline, does that dis-able us from being able to get beyond the baseline? Or does it enable us to get there faster? Or does it make it look like everything is non-baseline?
"All dead people are famous."
[Photo: "Famous People Players" by Danielle Scott]
Previous entry: Why Plans Fail or It's Your Own Darned Fault
Next entry: Evidence of KM, redux