AI and Simulations: Accelerating Expertise
by Ruth Clark
Recently I watched a short documentary on AI. One part of it especially caught my attention. They taught the AI the basic rules of chess and turned it lose on simulated chess games. It played against itself 24-7 millions of times. In a matter of hours it learned new strategies – some not even recognized chess strategies. This reminded me of research by Ericsson that focused on expertise. He reported that to build high levels of human expertise in any domain requires about 10 years of deliberate practice. By playing millions of games 24-7 the AI application was able to compress years of experience into just a few hours.
Assuming we continue to rely on human expertise, simulations are a proven method to accelerate expertise. A good simulation offers compressed experience. A recent meta-analysis (Chernikova et al, 2020) analyzed 145 studies that evaluated simulations in several professional domains including health sciences, teacher education and management skills. They reported an impressive positive overall effect size of 0.85 on learning outcomes. They found simulations had highest learning benefits on technical performance of procedures such as laparoscopic surgical skills (1.06) followed by diagnostic skills (0.82), and problem solving (0.88). They also found that simulations based on higher levels of technology such as programmed mannequins or virtual reality resulted in better outcomes compared to simulations that did not use technology. An important feature of effective simulations involved higher levels of task authenticity that represented relevant aspects of a highly realistic situation or just one aspect of a situation set in a realistic context.
Chess of course is a highly structured domain with clear cut rules involving pattern analysis. I wonder to what extent AI learning can apply to domains that are not as black and white? What roles will simulations play in the future to accelerate human and artificial expertise?
To Read More
Chernikova, O., Heitzmann, N., Stadler, M., et al. (2020). Simulation-based learning in higher education: A meta-analysis. Review of Educational Research, 90, 499-541.
Ericsson, K.A. (Ed.). (2009). Development of professional expertise. New York: Cambridge University Press.