Surprise Minimization or Free Energy Minimization (see post Prediction Machine et al) as presented by Andy Clark and including the ideas of Karl Friston and others
I continually look for commment on and expansion of these ideas, and I often do this in the most lazy of ways, I google them. Recently I seemed to find the last two mentioned on the same page of a philosophy book. That was not actually true, but it did remind me of similarities that I could point out. The idea of a compensatory process where one changes his belief a little to match the current set of “facts” tracks well with the idea that we can get predictions correct by moving our hand to catch the ball so that it does not have to be thrown perfectly. Both clearly try to match up the environment and ourselves. The Parallel Constraint Satisfaction model minimizes dissonance while the Free Energy model minimizes surprise. Both dissonance and surprise can create instability. The Free Energy model is more universal than the Parallel Constraint Satisfaction model, while for decision making PCS is more precise. The Free Energy model also gives us the idea that heuristic models could fit within process models. All this points out what is obvious to us all. We need the right model for the right job.
This is more or less a continuation of the previous post based on Andy Clark’s “Embodied Prediction,” in T. Metzinger & J. M. Windt (Eds). Open MIND: 7(T). Frankfurt am Main: MIND Group (2015). It further weighs in on the issue of changing strategies or changing weights (see post Revisiting Swiss Army Knife or Adaptive Tool Box). Clark has brought to my attention the terms model free and model based which seem to roughly equate to intuition/system 1 and analysis/system 2 respectively. With this translation, I am helped in trying to tie this into ideas like cognitive niches and parallel constraint satisfaction. Clark in a footnote:
Current thinking about switching between model-free and model based strategies places them squarely in the context of hierarchical inference, through the use of “Bayesian parameter averaging”. This essentially associates model-free schemes with simpler (less complex) lower levels of the hierarchy that may, at times, need to be contextualized
by (more complex) higher levels.
As humans, we have been able to use language, our social skills, and our understanding of hierarchy to extend our cognition. Multiplication of large numbers is an example. We cannot remember enough numbers in our heads so we created a way to do any multiplication on paper or its equivalent if we can learn our multiplication tables. Clark cites the example of the way that learning to perform mental arithmetic has been scaffolded, in some cultures, by the deliberate use of an abacus. Experience with patterns thus made available helps to install appreciation of many complex arithmetical operations and relations. We structure (and repeatedly re-structure) our physical and social environments in ways that make available new knowledge and skills. Prediction-hungry brains, exposed in the course of embodied action to novel patterns of sensory stimulation, may thus acquire forms of knowledge that were genuinely out-of reach prior to such physical-manipulation-based re-tuning of the generative model. Action and perception thus work together to reduce prediction error against the more slowly evolving backdrop of a culturally distributed process that spawns a succession of designed environments whose impact on the development and unfolding of human thought and reason can hardly be overestimated. Continue reading →
This post is based on a paper by Andy Clark: “Embodied Prediction,” in T. Metzinger & J. M. Windt (Eds). Open MIND: 7(T). Frankfurt am Main: MIND Group (2015). Andy Clark is a philosopher at the University of Edinburgh whose tastes trend toward the wild shirt. He is a very well educated philosopher in the brain sciences and a good teacher. The paper seems to put forward some major ideas for decision making even though that is not its focus. Hammond’s idea of the Cognitive Continuum is well accommodated. It also seems quite compatible with Parallel Constraint Satisfaction, but leaves room for Fast and Frugal Heuristics. It seems to provide a way to merge Parallel Constraint Satisfaction and Cognitive Niches. I do not really understand PCS well enough, but it seems potentially to add hierarchy to PCS and make it into a generative model that can introduce fresh constraint satisfaction variables and constraints as new components. If you have not read the post Prediction Machine, you should because the current post skips much background. It is also difficult to distinguish Embodied Prediction and Grounded Cognition. There are likely to be posts that follow on the same general topic.
Having the good fortune to be lost in Venice, I was reminded of the nuances of the recognition heuristic. My wife found the perfect antique store which was unsurprisingly closed for lunch. We went on, but a couple of hours later we tried to recreate our steps. For much of the journey, we did well having only to recognize that we had seen a particular store or archway or bridge before. Unfortunately, that broke down when we realized that we were retracing our steps in a five minute period. We still remembered that walk to the restaurant the first night, but it was unfortunately not differentiated in our minds. This was certainly an example of less knowledge being more. Eventually, using a combination of GPS and maps we found our way back to our hotel, but we never did find that antique store. And I was trying.
This post is based on a paper by Benjamin Hilbig, Martha Michalkiewicz, Marta Castela, Rudiger Pohl and Edgar Erdfelder: “Whatever the cost? Information integration in memory-based inferences depends on cognitive effort.” that was scheduled to appear in Memory and Cognition 2014. Fundamental uncertainty is a not uncommon situation for our decision making. There is an ongoing argument with the fast and frugal heuristics toolbox approach and the single tool approaches of evidence accumulation and parallel constraint satisfaction. However that argument depends on the particular task, the type of task, and on and on. I am still waiting for a giant table that puts all those things together.
This blog has existed for two years. It includes nearly 150 posts. I should note that I remain the only one to have set eyes on a few of them. My most popular post What has Brunswik’s Lens Taught? has only 625 views. Clearly, the blog is all about me. My inability to recall what I have written in the past became obvious even to me recently. My post Parameter P -Slowness Factor (6 views)? is about a month old. My post Parallel Constraint Satisfaction- Plugging in the some of the Numbers (10 views) is a little more than a year old. While writing the Parameter P post the older post did not cross, enter, or even come close to my mind. Interestingly, they seem to concern exactly the same thing, but at different stages. This post will look at that. As for future posts, maybe it is time to look at the existing posts as source material and try to integrate the most significant ideas.
This post looks at Parameter P, a specific construct of the PCS-DM model, as elaborated in “What is adaptive about adaptive decision making? A parallel constraint satisfaction account,” by Andreas Glöckner, Benjamin E. Hilbig, and Marc Jekel (Cognition 133 (2014) 641–666). (See post Revisiting Swiss Army Knife or Adaptive Tool Box.)
Glockner et al state that transformations in Eqs. (3)–(5) (See figure at top of post.) are commonplace and sensitivity analyses have shown that the selection of specific values has little influence on predictions as long as inhibitory connections are relatively strong compared to excitatory connections. PCS-DM predictions, however, strongly depend on Eq. (2). In this equation for calculating connection weights, validities are corrected for chance level (.50) to avoid that irrelevant cues have a weight. Parameter P allows PCS-DM to capture individual differences in the subjective sensitivity to differences in cue validities. Low sensitivity is captured by low P. By contrast, high sensitivity for cue validities is captured by large values of P with high values as special cases in which less valid cues cannot overrule more valid ones. P captures sensitivity at the level of individuals, that is, it determines how an individual transforms explicitly provided or learned information about a cue’s predictive power (i.e., cue validity) into a weight. Glockner et al suggest that P describes a core property of a psychological transformation process that precedes decision making.
This post is based on a paper: “What is adaptive about adaptive decision making? A parallel constraint satisfaction account,” that was written by Andreas Glöckner, Benjamin E. Hilbig, and Marc Jekel and appeared in Cognition 133 (2014) 641–666. The paper is quite similar to that discussed in the post Swiss Army Knife or Adaptive Tool Box. However, it reflects an updated model that they call the PCS-DM model (parallel constraint satisfaction-decision making). From what I can tell this model attempts to address past weaknesses by describing the network structure more fully and does this at least partially by setting up a one-free parameter implementation which can accommodate individual differences and differences between tasks.
I occasionally like to go far afield from judgment and decision making, and here I go again. This post takes a look at Michio Kaku’s 2014 book, The Future of the Mind–The Scientific Quest To Understand, Enhance, And Empower The Mind, Doubleday, New York.
Decision models can sometimes seem very explanatory, but they seem so simple minded when I read in Kaku’s book that we have two separate centers of consciousness and that we may all have photographic memories.
This post is based on the paper: “Single-process versus multiple-strategy models of decision making: Evidence from an information intrusion paradigm,” written by Anke Söllner, Arndt Bröder, Andreas Glöckner, and Tilmann Betsch. It appeared in Acta Psychologica in January 2014. It is a well done overview of multi-attribute decision models (Multi-attribute decision making deals with preferential choice e.g., “Which dessert do you like better?” and probabilistic inferences e.g., “Which dessert contains more calories?”), along with clever experiments. I am confused that it is single process vs multiple strategy. I would think that it would be process vs. process or strategy vs strategy.
This appears to me to be another polite skirmish in the continuing battle between fast and frugal heuristics and compensatory connectionist models. Do we change strategies or adjust decision thresholds or weights? However, the researchers have moved back to broader frameworks to get a different way to study and attack. This paper has an interesting group of authors. Sollner and Broder wrote a paper last year that looked at similar issues, but focused on the importance of looking separately at how information is acquired and how information is integrated. Glockner and Betsch are prime proponents of parallel constraint satisfaction theory– a single process model that apparently is weak on information acquisition. I will expect a Gigerenzer or Marewski counter move soon for the fast and frugal heuristics side. I should note that there seems to be much respect between those with differing views, and the idea that probably everyone is a little bit wrong and a little bit right seems to pervade.
This is clearly an example of the blind trying to lead when sight is a real advantage. Glockner displays PCS1 and PCS2 in some figures in his January 2014 paper in the Journal of Judgment and Decision Making. Since I tend to look at the pictures, this got my interest. Was this some different model or some innovation? I have provided some narrative explanations of Parallel Constraint Satisfaction in earlier posts, but here I am going to look at the difference between PCS1 and PCS2. I am doing this by cobbling together explanations from a few of Glockner’s papers. This is a little dangerous since the experiments are different.