Expertise
Most of us specialize in something in our lives, perhaps even achieving expert status. Gladwell once claimed that 10,000 hours of practice in anything is enough to make anyone an expert. While this has been roundly debunked, large neural network models are now able to achieve expert or superhuman ability when given enough millions of training episodes for a single task.
On the other hand, these models struggle far more than people to generalize this expertis to even slightly different tasks, or indeed to master more than one thing at all. What is expertise in cognition? This session will explore how we can model expertise, how expertise changes our how a cognizer approaches a task, and what trade-offs this can come with.
Primary Readings
Everyone should read these and be prepared to discuss:
| Chi, M. T. H., Feltovich, P., & Glaser, R. (1981). | Categorization and representation of physics problems by experts and novices. Cognitive Science, 5, 121-152. |
| Gobet, F., Lane, P. C., Croker, S., Cheng, P. C., Jones, G., Oliver, I., & Pine, J. M. (2001). | Chunking mechanisms in human learning. Trends in Cognitive Sciences, 5(6), 236-243. |
Secondary Readings
The presenter should read and incorporate these:
| Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). | The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100(3), 363.This article from 1993 discusses the role of deliberate practice in achieving expert performance in various domains. It challenges the notion of innate talent and emphasizes the importance of intense practice for at least 10 years. The article provides evidence on the potential and limits of environmental adaptation and learning. This is the origin of the the idea popularised by Malcolm Gladwell that 10,000 hours in anything will make you an expert. Is this really exactly what Erisson et al found? |
| Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., Van Den Driessche, G., … & Hassabis, D. (2016). | Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484-489.The article discusses the use of deep neural networks and tree search in the development of the computer program AlphaGo, which plays the game of Go. The researchers trained policy and value networks using 30 million positions from the KGS Go server, achieving high accuracy in evaluating board positions and selecting moves. AlphaGo was able to defeat other Go programs and even a human professional player. The article also discusses the techniques and methods used in AlphaGo, including rollout policy, exploiting symmetries, and reinforcement learning. |
Questions under discussion
- What differences do we see in how experts and novices approach a problem?
- What are some features of domain expertise?
- In what ways might expertise trade off with flexibility?
