Artificial Intelligence Demystified
Artificial Intelligence (AI) continues its ascent as a major factor in many fields, including healthcare. Although I have had a pretty good general sense about how it works, I gained some great new insights at a November 13 symposium on AI sponsored by Leadership Atlanta. David Lee, VP of Innovations UPS Ventures, delivered the keynote that provided a helpful framework that I wanted to share.
AI is loosely defined as using machines to replicate certain cognitive abilities, such as drawing conclusions from prior experiences or solving problems, that are normally associated with humans or animals.
David Lee outlined three ways AI operates:
Supervised learning where researchers feed many, many examples of a particular item or concept into a computer which then extrapolates principles from the input data. For example, after ingesting thousands of pictures, images, and drawings of a cat, the machine can determine the defining characteristics of a cat to the point of being able to conclude whether an additional image is cat-like enough to be positively identified as a cat.
Reinforced Leaning where the respondent is provided feedback on whether or not it accurately identified the correct answer. Scientists observing a lab rat navigating a maze and then rewarding it for successfully finding it was way through is one example of non-computerized reinforced learning. A machine learning version might involve two robots where the first one attempts a task and the second observes how well the first on did. The second robot learns by extrapolating the principles that led the first robot to success.
Unsupervised or Deep learning where huge amounts of absolutely raw data are fed into a processor with no hypotheses about unifying characteristics. The computer’s job is to detect underlying patterns, if any, contained in the data.
Lee went on to explain that the most powerful applications of AI in the next five to ten years will involve supporting human intelligence, not replacing it. AI can play a vital role in identifying potentially significant patterns to alert human observers to intervene but will not do so itself automatically.
Healthcare is already seeing this type of application with programs like IBM Watson, which can narrow diagnostic possibilities down to the most probable ones and allow clinicians to determine the most appropriate treatment. Although IBM Watson has taken huge steps toward practical application of AI in a clinical setting, Lee indicated it still has a long way to go.
Lee concluded that AI truly has the potential to revolutionize many aspects of how we live and work and has many implications for future workforce requirements. Many jobs may be eliminated, but many others will be created to support the industry and to tap into AI’s potential. For example, although it seems counterintuitive, there are now twice as many bank tellers today as there were in 1975 when ATMs were first introduced.
In closing, he asked, “Are we all doomed?” His wry response: “Yes, but not for a while.”