Tuesday, 9 July 2013

A new direction for psychology

I have my doubts about psychology. 

Anyone who's read this blog before knows that Andrew and I are fairly opinionated about what we think is right and wrong with psychology research. This isn't about small effect sizes or falsified data, which are currently popular (and valid) concerns. It's about the the types of questions psychologists usually ask and whether they are useful and likely to move the discipline forward. These questions are dominated by constructs - self esteem, prejudice, working memory capacity, intelligence, motivation - to the point where an alien reading a psychology journal would be forgiven for assuming that the point of psychology was to understand constructs, not people. 

Psychologists ask questions about constructs because the dominant theoretical paradigm (cognitive psychology) says that mental states play a causal role in behaviour and that, to understand how people work, you have to understand the content of these states and the nature of the cognitive processes that operate on them. This doesn't sound crazy. Even though I think cognitive psychology is a fatally flawed paradigm (see here, here, and here), I am enculterated enough in mainstream psychology that this doesn't automatically sound like a bad way of doing things. And anyway, I'm done doing active battle against cognitive psychology, so this post isn't to re-hash what I've written about elsewhere. Instead, I want to lay out what I think psychology should be doing. The idea is simple, but it's radically different from the mainstream. 

Here's the claim:

Information is the primary external cause of behaviour. If psychology is going to make any real progress, it must be grounded in a thorough analysis of the types of information available and the mechanisms by which information is used to control or precipitate behaviour. 

Here's the argument:

Let's say that a behaviour is any potentially observable activity that does not have a mechanical cause (e.g., raising my hand is a behaviour, having my arm raised by a robot is not).

And, let's say that the goal of psychology is to create models that explain behaviour. These models should enable us to predict behaviour, to understand why it has the form it has, and to provide the basis for predicting novel results and changing / manipulating behaviour. What's the best method for reaching this goal? 

Answering this requires first identifying potential non-mechanical causes of behaviour. There are two main places to look for potential causes - things outside of an organism and things inside of an organism. Outside of an organism there is structure in energy and chemical arrays. Structure in light, sound, magnetic fields, molecules. For animals with the right kind of receptors (e.g. retinas), this structure is potentially informative about the world. Information is defined as structure in energy arrays that is used to control or precipitate behaviour. 

Information is the primary external cause of behaviour*. 

Understanding information is what psychologists should be doing. This is the right starting point for psychology because it includes most external causes of behaviour and clarifies the roles of internal contributors to behaviour. Brains and bodies are important because they define an organism's ability to detect and use information. Questions about brains and bodies should be about these problems. Comparisons between the brains and bodies of different organisms should be about these problems.

*External mechanical events can change our brains and bodies (e.g., via damage). Ingesting substances (e.g., via eating) can change our brains and bodies by virtue of the chemical properties of the substances, which may cause harm, be used by the body to carry out vital processes, or alter gene expression. So, external mechanical events can change the nature of the internal contributors, which can alter behaviour. Understanding these events will be useful in clarifying why brains and bodies are the way they are for a particular organism in a particular task. However, this is a step removed from the main goal of psychology. 

Starting points

The perception-action literature following on from Gibson's Ecological Psychology is one good example of psychologists taking information seriously. This literature is useful because it has already begun the work of investigating how one type of information - perceptual information - is used to control or precipitate behaviour and because it illustrates a research strategy that can be used to investigate other types of information. Andrew and I recently wrote a paper explaining how this research strategy could be applied to any area of psychology, not just perception-action (Embodied cognition is not what you think it is).

The field of possible information types, however, is much larger than just perceptual information. To begin getting a handle on this I have created an information taxonomy in which information types can be defined along a number of dimensions. The dimensions, themselves, are defined in terms of the relationship between the world (the structure in light, sound, etc) and the organism. Structure in energy or chemical arrays is classified as an information type by looking at the first person perspective of a particular organism in a particular task. 

The major dimensions are: detection (does an organism have to learn to detect the information?), learning (does the organism have to learn to use the information?), and aboutness (is the information about the thing in the world that caused it or is it about something else?). There are some potential minor dimensions, most importantly, continuity (does the information precipitate an action or is it used in the continuous control of action?).

For more info see here and here.

This taxonomy provides, for example, a basis for distinguishing structure that precipitates reflexes from that used to pick up a coffee cup or to engage in a conversation. It also provides a basis for comparing animals in terms of their ability to use various information types. 

Research Activity

One of the most appealing things about this new framework is that it is doable. The information taxonomy provides a principled basis for a large-scale project cataloging animal behaviour in terms of information types. A similar project needs to be done comparing structural brain differences between animals with access to different information types. Not to mention, the very large volume of empirical work (based on our four step research strategy) that needs to be done with people. 

We also co-hosting a special topic in Frontiers in Human Cognition on Radical Embodied Cognitive Neuroscience. And, if you have any ideas about topics you'd like to see tackled, please leave a comment here


  1. Great post and great blog, very much enjoy reading it. Two brief comments; First, I find it interesting when and where the term information is preferred over the most common philosophical term knowledge. Of course, it has been observed that one of the 'dirtiest secrets' of (analytical) philosophy has been its inability to provide a clear and systematic account of what constitutes knowledge and what doesn't (i.e. some of the current debates on knowing how and knowing why). And this brings up my second point, more specifically a question in relation to your post: Can you offer a succinct approximation, if not a precise definition of what counts as "information" for a given organism? Further, would this definition be a human centric account or would it include living organism at large?

    1. 'Knowledge' is effectively a construct. Psychologists talk about what people 'know' about in order to behave but this doesn't explain anything and it's not clear that the term carves nature at any useful joints.

      Information is not a construct; structure in energy arrays created by task dynamics exists. This means that understanding information allows for the possibility of understanding the mechanism by which an organism comes to "know" what it needs to know. So scientifically, information beats knowledge as a thing to study.

      Read Sabrina's taxonomy of information post for a not quite up to date draft of the whole landscape she is mapping out. I've written about information in the specific context of perception and action; check out the links on 'Specification & Its Discontents' and the post I linked to above about task dynamics.

      Different organisms will have access to different information, but the principles by which information is created should apply across the board. Chasing up the comparative psych literature is therefore a big priority for Sabrina's stuff.

    2. Information (for a given organism) is any structure in an energy or chemical array that precipitates or is used in the continuous control of action.

      So, information is always conveyed via a medium (e.g., light, sound) and information is always structured. For a given organism, you could make a general prediction about what types of information it is sensitive to by looking to see what types of receptors the organism possesses. For example, an organism without photon receptors cannot use structure in light.

      I think it is very important that this applies to all organisms, not just humans. It also provides an objective way to discriminate between organisms based on information use, without resorting to value-laded language like "intelligence". For example, I can contrast the set of organisms who cannot learn to detect new structure in energy arrays (those who are 'hard-wired' to be sensitive to certain structural invariants) with the set who can learn to detect new structure in energy arrays (those who can learn to detect new invariants). The organisms in the second set will likely appear to be somehow more sophisticated to a human observer than the organisms in the first set. The information-based comparison formalises this hunch and provides a constraint on how we expect the organisms to differ in neural complexity, for instance.

  2. Interesting post but here's an objection:

    What is information to any organism is determined not just by what kind of organism it is (species, other characteristics) but what state it is in.

    To give a simple example, if I'm thirsty then a glass of water means one thing. If I'm not, it means another thing.

    This is especially true in animals - I suspect that my cat has absolutely no recognition of or interest in water, unless he's thirsty. He simply ignores it - externally, and I presume 'internally' too.

    So what information does a glass of water have? It depends on who's looking - but 'who' includes the internal state (and thirst is almost literally an internal state, electrolyte levels in the blood etc.)

    Someone infected with rabies might have a furious aversion to a glass of water (hydrophobia). To him, it carries information of a very different kind compared to anyone else. But the glass is the same and the person is the same except for their internal state.

    So it seems that internal states are important. Now this is a simple example but it could be extended to anything.

    I would propose that it's simply a matter of perspective whether you say that a thirsty man and a hydrophobic man "get different information from water" or "have a different attitude to water".


    1. As I view Sabrina's model, an organism learns a set of context-dependent behavioral dispositions in response to a given sensory input array. The information in the array is context-dependent in that the external environment can vary resulting in different information contents for sensory input arrays that include contributions due to a given external surface or event. For an input array with a specific information content, the particular behavioral disposition actualized is context-dependent in that it is a function of the organism's internal state.

      From that perspective, the description of the external environment - including the "glass of water" - seems to assume (implicitly) that the external context is fixed, in which case the information content of the array is also fixed. Then the different hypothesized responses are due only to different internal contexts (AKA, states).

      Ie, in my view the input "information" isn't dependent on an organism's internal state, but the organism's responses to a given information input are.

    2. You are absolutely correct that internal states matter. This is why I partitioned external and internal causes of behaviour and located information as the primary external cause. Brains and bodies play enormously important and complicated roles in our behaviour.

      I'm suggesting that the right question to ask about internal contributors to behaviour is how they change access to or ability to use information. This is critically different than the dominant question type which is how internals states influence performance on a task (e.g., how does motivation influence athletic performance?).

    3. Information has no intrinsic meaning!

      Information is a measurable quantity used in information science and physics. Information and physical systems are linked through the concept of entropy. Shannon Entropy may be loosely defined as the amount of bits needed to describe the unique features of an information structure. Analogously, entropy in physics could be described as the amount of information that is needed to describe the unique modes of behaviour of a physical system. A completely random system has many such modes, is disordered, anything can happenand and thus a high entropy. A deterministic system has low entropy, is highly ordered, only a few things described by a deterministic rule can happen. The information of a physical system are therefore its degrees of freedom and the mathematical description of the entropy of a physical system is equivalent for all intents and purposes to Shannon entropy.

      What is described here is the decoding of information for meaning by using intelligence (control the flow of information to select what is relevant given current needs, expectations and priorities). This "control" and "select" are should not be considered actions by an agent, but are purely a phenomenological description: knowing a door is pass-thruable because of the biomechanic properties of your body is also using intelligence (intelli-gere).

      The thirsty and hydrophobic man decode one and the same information structure in a different way. If you start to view each code as a unique piece of information (that needs to be stored internally) you'll end up with the cognitive representation library nightmare of the information processing approach.

      At least that's my pov: http://anti-ism-ism.blogspot.nl/2013/06/codes-information-meaning-and.html#.Ud05Pha9YwI

    4. Fred -

      Being a comm engr, I've previously raised the same objections to use of the word "information" in this forum. But over time I've come to accommodate it as follows.

      As you note, the essence of "information" in the comm context is resolution of uncertainty in the occurrence of a random variable whose possible outcomes are known to sender and receiver. (I assume a discrete case because it captures, I think, the crux of the issue at hand while being much simpler to deal with than the continuous case.) In the present context, there is obviously no explicit a priori agreement between the "sending" external environment and the "receiving" organism as to the possible values of a sensory input array. But over time, there comes to be an implicit "agreement" in the sense that the organism learns to effect specific responses to specific arrays (actually, to ranges of array values). As I suggested above, responses are determined by the overall internal state of the organism, one component of which is the neural activity consequent to the sensory input. In that sense, there is "information" in the array. The external environment "sends" an energy array some possible values of which can be "recognized" by the receiving organism, thereby resolving the uncertainty in the sensory input stream.

      While also agreeing that there is no intrinsic meaning in Shannon information, I've come to think of the "meaning" of information in any communication context as being the response intended by the sender (I assume this is along the lines of your "codes"). Again, there is obviously no explicit "intent" on the part of an inanimate environment, but one can look at an affordance as having implicit "intent" in that it can cause a response in an organism that "recognizes" the accompanying sensory stimulation from previous occurrences during the learning process. In which case the meaning (in my sense) of the information in the affordance is that response.

      All, of course, IMHO.

    5. We aren't talking about Shannon information. That is a very powerful and efficient way to do some things, but it's not what biological organisms are interacting with. We're talking about Gibsonian, structure in energy arrays kind of information. It's an important difference.

    6. Also, if you poke around this blog, you will see that adopting this theoretical perspective on information removes the need for internal representation. This is an extended, embodied, dynamical systems model of cognition, not an information-processing one. There is still internal mediation, obviously, but this is not in the form of internal storage of information.

    7. The problem remains that many more people know words like "information" and "detection" in the comm sense than in the sense assumed here, so non-eco-psych visitors (like me) are much more likely to assume the former. And many in the general area of phil of mind are applying comm theory concepts (eg, Crick and Koch, Edelman and Tononi) resulting in even more people visiting this site primed to assume that sense. Entering an unfamiliar area of knowledge requires first learning the relevant vocabulary. Encountering a familiar vocabulary from a better-known area in a lesser-known area that assigns new meanings adds the additional step of unlearning the previous meanings. I was just describing my way of relating the different meanings as a possible aid to making that transition.

      The focus here is on continuous processes, with dynamical systems theory apparently the preferred analytical tool. I'm unfamiliar with DST, but presumably its relevant versions assume an input stream with some content to be extracted and used to effect responsive action. Is the vocabulary of that area similar to the comm vocabulary? If not, why not eliminate the middle-man and use the DST vocabulary?

    8. Charles, yes, I always mean to put a "not Shannon information" disclaimer at the beginning! As much as I respect Gibson, it wouldn't have hurt for him to choose a different word.

      DST doesn't require content to be extracted to effect appropriate action. Intelligent action emerges from the joint activity of all participating systems hitched together with the appropriate relation (as keeping accurate time emerges when the parts of a watch are assembled in the correct way).

    9. Thank you for those replies.

      I am quoting from Michaels & Carello (1981):

      "We need only look at the structuralists’ notion of sensation, or the variables examined by psychophysicists to see that light has been dealt with in terms of intensity and wavelength, sound with respect to amplitudes and frequencies. Gibson, on the other hand, proposed that a more psychologically relevant treatment of stimulus would involve not energy, but information. The term “information” has a variety of meanings (for example, from computer science) but, as Gibson uses the term, information is structure that specifies an environment to an animal. It is carried by higher-order patterns of stimulation—neither points of light nor collections of such points (images)—but, rather, complex structures often given over time. These patterns are information about the world"

      1. Information that acts as a "stimulus" for an animal is always a complex information structure that specifies an environment to an animal.

      "James Gibson and those who follow his approach adopt an eco-
      logical stance: they believe that perceiving is a process in an animal- environment system, not in an animal. Proponents of the ecological view argue that perception is, quite simply, the detection of information. This approach is labeled direct because a perceiver is said to perceive its environment. Knowledge of the world is thought to be unaided by inference, memories, or representations."

      2. An animal-environment system perceives information, knowledge about the world unaided by inference, memory, etc.

      I see no problems with information and codes as used in information science (Shannon) and physics (Degrees of Freedom) to describe the Gibson's arrays: The information is a structure that may be decoded for meaning, knowledge about the world by an animal. This act is called direct perception (Smitsman: there is no perception, only directed action) and the "percept" or perceived affordance and the act of perceiving cannot be reduced to parts of the animal environment system. Moreover, these structures are constantly changing.

      What prompted the invention of effectivities of the animal necessary to define perceivable affordances for the animal (Turvey, 1981) is a perhaps a pseudo-ontological distinction, because the animal itself, its composition an history, is also an information structure. Consider predator and prey, but the ability to perceive whether a staircase is mountable based on an invariant measure of biomechanical properties of the leg and the physical properties of the riser is also a coupling of information structures providing meaningful information about the world.

      About DST: I teach my student to think of it as a Dynamical Systems Toolbox. Although people (psychologists mainly) will speak of a DS theory, it mostly refers to the mathematics of change that allow modelling of complex dynamics in classical state spaces. In a way it is completely ridiculous that psychology claims to study change of behaviour, without knowing anything about the formal description of change processes used to describe many natural and social phenomena.

      Add to classical DST the mathematics of deterministic chaos and fractals and you have a very rich set of mathematics, models, tools and metaphors to describe complex dynamical patterns, but still not a theory. The theory of synergetic control, is a theory. But Catastrophe theory? Dynamic Field theory? Not sure, they seem to be model based theories instead of principled theories.

      Many of the present day ecological psychologists are exploring these complexity science metaphors and tools in the context of Gibsons legacy.

    10. The information is a structure that may be decoded for meaning
      This kind of talk is the main problem with using information in the physics/Shannon sense. The ecological approach is to get away from 'decoding' and towards the more dynamical notions of 'coupling' and 'resonating'. Decoding makes you think that you have to convert the information into an understandable form. This isn't the case (other than the transduction required at the level of the sensors, which isn't waht you mean here).

      What prompted the invention of effectivities of the animal necessary to define perceivable affordances for the animal (Turvey, 1981) is a perhaps a pseudo-ontological distinction, because the animal itself, its composition an history, is also an information structure.
      I'm not particularly down on effectivities. I think the distinction between affordances and effectivities is a useful analytical tool, regardless of the fact that the whole things is just coupled dynamical systems. I'm all for information being where the real action is, but this is useful for getting some science done.

    11. I have no problem to consider decoding as coupling dynamics or resonance.

      The key point is that one and the same information structure can evoke many kinds of resonance / coupled dynamics, depending on the properties and history of the agent with which the information resonates.

      If someone comments on your Facebook profile picture: 'dapper!', to an English language user this would mean 'Neat and trim in dress, appearance, or bearing' , but it means 'brave' in Dutch and knowing their love for sarcasm this could have the opposite meaning of the English language meaning. In this case one and the same information structure can give rise to different meaning (= be decoded differently) depending on the conventional rules of the native language, writing system and culture of the agent.

      Equivalently, there may be different information structures that can be given the same meaning (decoded, or resonate to elicit the same action) in different contexts (e.g., different spelling of the word father in different alphabetic languages)

      In fact, I have a coupling hypothesis about speech perception in developmental dyslexia modelled as a 2D potential model, including an attempt to fit model parameters from data: http://fredhasselman.com/main/wp-content/supmat/btb/ch5/CHAPTER5_BEYOND_THE_BOUNDARY.pdf

      Its not really a draft anymore, but not the final version yet. I guess I use resonance as in recurrent neural computation, stability of a state after perturbation.

  3. Information is the primary external cause of behaviour.

    I would prefer to reverse that, as:

    Behavior is the primary cause of information.

    To illustrate, suppose that I pick up a ruler and go measure the window. The measurement is information that was caused by my measuring behavior.

    The more traditional view is that the information is out there, and by measuring I am just picking up information that already exists. But that seems more like the cognitivist view that the world is full of abstract propositions.

    I'm aware that Gibson did talk of picking up information. But he also emphasized the role of behavior, such as in the saccadic motions of the eye.

    1. Gibson was clear that we needed flow to pick up invariants - so either the world needed to move or organisms needed to move to create access to structure in energy or chemical arrays. The role of causality is definitely complicated here and our behaviour does, indeed, change the information landscape (Andrew's go-to example of catching a flyball is a great illustration of how our behaviour creates information).

      I am aware of the traditional view regarding the persistence of information, but I don't think it hold up. Structure is the thing that endures in the world whether or not anything can detect it. There are all kinds of potential invariants that may or may not be of any use to anything. Psychologists can, of course, look at these invariants and the events in the wold that create them and make predictions about how they might be used in a task. But, invariants can only be informative if they are actually usable by a given organism in a given task. I can't make predictions about the usefulness of an invariant in the optical array, for example, without knowing that organisms exist that are sensitive to structure in light. This is why it is necessary to separate out structure and information.

  4. Interesting post! I think you'll find a number of other researchers who agree with your logic but disagree with some of your axioms and assumptions. Certainly not all psychologists will agree with you that the goal of psychology is to explain behavior, for example.

    A clarifying question: My understanding is that perception refers to the processes by which the mind takes in external (environmental) and internal (somatic) information. Yet you refer to perceptual information as only one type of information. What might be an example of non-perceptual information, and how is this non-perceptual information different from an internal state?

    1. Good question re non-perceptual information. First, the concept of perception we use on this blog comes from Gibson's ecological psychology, which is different from older approaches that treat perception as a secondary process of interpretation (a primarily cognitive act) following on from sensation (receiving input from the world). For Gibson, perception was direct because structure in energy arrays is lawfully related to the event in the world creating it, meaning that no intervening process of enrichment or interpretation is needed to guide appropriate action.

      The notion that there are more types of information than just perceptual information is new and follows on from an analysis of the different ways in which organisms use structure in energy or chemical arrays. I'll provide a few examples below:

      Perceptual information is defined as structure in an energy or chemical array where 1) the organism must learn to detect the structure, 2) where the meaning of the structure (for that organism in that task) is lawfully related to the event in the world that caused the structure (i.e., a dog's bark is about the event of a barking dog), and 3) where an organism must learn how to coordinate action with respect to this structure.

      Reflex information is defined as structure in an energy or chemical array where 1)the organism does not need to learn to detect the structure, 2) the meaning of the structure (for that organism in that task) is lawfully related to the event in the world that caused the structure (i.e., a dog's bark is about the event of a barking dog), and 3) an organism does not need to learn how to coordinate action with respect to this structure.

      Linguistic information is defined as as structure in an energy or chemical array where 1) the organism must learn to detect the structure, 2) where the meaning of the structure (for that organism in that task) is conventionally related to the event in the world that caused the structure (e.g., the word 'dog' is not about the physical process of creating the spoken word 'dog', it's about a domesticated mammal that is often kept as a pet), and 3) where an organism must learn how to coordinate action with respect to this structure.

      For more information, you can check out the link to the information taxonomy post in the "Starting Points" paragraph above.

  5. Great post. Have you ever read the article, Behavior Analysis and a Modern Psychology: Programs of Direct Action by Morris (2003)? It was the first article I came across that spoke of trying to bring together various non-representational, non-mediational approaches to psychology.

    I believe that more of a collaboration between these camps (e.g. behavior analysis, embodied cognition, ecological) would help alleviate much of the conceptual confusion and move psychology forward into more workable territory.

  6. David,
    I'm jumping into this discussion quite late (playing catch up), but if you like Morris's suggestion about bringing together these related approaches, you should really check out this month's issue of The Review of General Psychology. It brings them right together, though it stops short of forcing them to talk with each other... that's the next step.

    End shameless self promotion.

  7. Sabrina,
    I agree! I've been trying to figure out why this is so hard a perspective for people to adopt. It occurred to me that because we see THINGS (objects and events) via Gibson-information, the information itself is transparent (psychologically). Even when we learn to see things in "artistic" ways, i.e., in perspective, as composed of component shapes, etc., we are still not being explicitly sensitized to the information by which we normally see.

    A weird problem,