A primary mechanism that drives learning is a process called "chunking".

If you haven't encountered the concept of chunking before you might want to consult the wikipedia page on Chunking.

My interest in chunking was reignited because I've been (re)learning server technologies as part of the lead up to my recent server upgrade. I was interested in exploring how theories of chunking might be applied to what I was experiencing and learning.

There is considerable trial and error involved in getting servers and software working together. There are times when you can be working on a problem for 2 or 3 hrs trying to figure out how to get something to work and suddenly some path leads to a working solution. Once it works, then you can setup the same solution on another server in much less time. You've built up a chunk of expertise. As you build up these chunk of expertise your overall level of expertise rises.

The buildup of expertise can occur at a cognitive level like this chunk, or it can be happening beneath your awareness (i.e., implicit learning). In a study of cigar rollers, they found that the speed of cigar rolling increases according to a power law - speed up occurs most at the beginning and continues to improve with time on task. There may be discontinuities in speed up if new techniques are consciously employed initially to overcome some limitation in technique.

In Allan Newell's theory of chunking (as expressed in Unified Theories of Cognition 1990), a chunk is built around an impass. You bang your head on a problem for awhile and if you figure it out you can potentially create a "production" that takes as input your problem state and goal, and welds that information into an action to take, namely, the action you took to overcome the impass. Symbolically, a production has this form:

If State + Goal THEN Action.

In John Laird, Allen Newell, and Paul Rosenbloom's SOAR model, chunking plays an important role of converting declarative knowledge into procedural knowledge. The power of chunking comes from that fact that chunks can be organized hierachically. Lower level chunks (O V E L) can be combined into a singular higher level chunk (L O V E). If I give you list of random letters to remember, if you can break that list into higher level chunks of meaningful information you can recall many more letters in that list.

Most of the chunking we do takes place beneath our awareness at a very low level. The level at which we operate as adults builds upon a foundation of low level chunking we do in childhood. Over time the chunks that we are aware of in our problem solving are at a higher and higher level.

The view that chunking can be realised in software as a system of productions is a very powerful idea coming out of Carnagie Mellon, where Allan Newell, Herb Simon and John R. Anderson were some prominant proponents. Herb Simon also won the Nobel Prize in economics and applied some of these ideas to how to think precisely about firms.

One way you could model a firm is as a production system consisting of a dynamic set rules that fires off certain actions when certain goals and situational states occur. What happens when that production system encounters an impasse such as the present coronavirus situation where you still have the same goals but the states of the system are changed and the available actions are changed?

To overcome an impass humans often use declarative knowledge to building new chunks of expertise, ideally building up those chunks quickly to adapt quickly. Many will be faced with the decision of whether it is worth learning new chunks of expertise in the current industry or move to a new industry that may have been a sideline but becomes more of a focal point.

What about Grit and other ideas that are used to explain how high level expertise is acquired? This is where we need to retreat a bit from the view that chunking can explain all aspects of learning and say that we are only concerned with the cognitive aspect of learning and not about the emotional, motivational, and socio-economic aspects that are important drivers of learning as well. These other aspects, however, are not learning theories per se because they alone cannot perform adaptively - they need the guts of a production system chunking along as the underlying driver of adaptive performance.

In summary, the purpose of this blog is to touch on the concepts of a chunk, chunking, and production systems. These are practically useful concepts not only for thinking about how learning occurs, but also in thinking about how firms work and adapt over time.