In a recent article, I wrote about three types of cognition that are changing how we think about how we think (how meta!). All are interesting, but they also have implications for understanding for supporting us in doing things. I think it’s important to understand these cognitions, and their implications. First, I want to talk about situated cognition.
The psychological models of thinking really started with the behavioral models. The core argument was that we couldn’t look ‘inside the box’, and had to study inputs and outputs. Cognitive psychology was a rebellion from this perspective. The new frameworks started showing that we could posit quite a bit about what went on ‘in the box’. We got concepts like sensory, working, and long-term memory, and processes like attention, rehearsal, encoding, and retrieval. With most of our learning prescriptions. However, both were about the ‘the box’.
However, the observed behavior didn’t match the formal logical reasoning that underpinned the model. We needed new explanations. The computational model fell apart. And, despite rigorous attempts to create logical models that described human behavior, they were awkward at best. The shift came when Rumelhart & McClelland, in their PDP book, described what became known as neural networks. Associated with this was a new model of cognition.
What gets activated in the brain is not a reliably pure representation, and is strongly affected by the context. Thinking is ‘situated’ in the context it arises in. If our thinking is the emergent behavior of patterns across neurons, and those patterns are the result of both internal and external stimuli, then we’re very strongly influenced by what’s happening ‘in the moment’. And that means that we can be captured (and fooled) by elements that may not even be consciously processed.
What this means in practice is that it’s harder than we think to get reliable performance across a range of conditions. That we should ensure that patterns are generated across ‘noise’ so that they’re reliable in the face of the appropriate triggers, despite any accompanying contextual patterns. and recognize that decisions can be biased, and design scaffolding to prevent in appropriate outcomes. Developing mental models that provide reasoning abilities about causes and outcomes are useful here. This flexibility is advantageous (and why machine learning struggles outside it’s range of training), but we want to tap into it in helpful ways.
Our approaches should reflect what’s known, and therefore we need to keep up. Situated cognition is a perspective that’s relevant to more effectively supporting individual and organizational performance and learning. So, what is your thinking about this?
Situated cognition, as well as distributed and social to a extent, seems like a return to the behaviorist methodology, focusing on contingencies instead of mental processes since they can’t be trusted. To be honest, most of neuropsychological research seems to only add noise to perfectly fine functional explanations.
David, I don’t agree it’s just about contingencies. It’s about the interaction between the cognitive and the context, not just a total abandonment of the cognitive. So implications like: ensure you’ve considered the contexts of application is an important component, combined with exercises like mental models. Agreed about the neuroscience, however, most of the important work is at the cognitive level, not the neural. Thanks for weighing in!
Thanks for your reply, Clark. I am not sure there is a practical difference in the outcomes of each model. After all, the behaviorist approach takes into account not just the current context but the individual’s history also (or rather considers that both are part of the current context), pretty much the same as the cognitive approach considers the internal processes and the external variables. If what makes a mental model reliable is its adjustment to context, I can understand why behaviorsts don’t bother with the cognitive construct at all… Sorry if this is a bit off topic!
Clark,
Did you mean to imply a causal line from neural networks to situated cognition? I was lucky to be in the room — literally, one room, with small adjoining side offices — for 8 months at UCSD in 1979, as part of a Sloan post-doc program. That was when Hinton, McClelland, and Rumelhart, were reinvigorating neural network research, and Ed Hutchins was developing an early situated model of cognition for Micronesian oceanic navigation. Both lines of research were well aware of each other, but the common causal factor was cognitive science. Cognitive science then was still in its early stages, sending exploratory roots deep into existing research areas. At UCSD, those roots went into the brain, and out to the world, but the center remained “how do we think?”
So close! I didn’t get there ’til 1983 for grad school (was there as an undergrad, too, until I finished in ’79). Yes, I do mean to imply a causal line, and yes it’s all cognitive science. Stay tuned for next week on distributed cognition! Thanks for the memories.