Cognitive Science

  • The Ola Project, subjective evaluation, rating versus choosing, time control
  • The Aoe Project, subjective evaluation, the role of expectation, bivalence
  • The Crystal Project, subjective evaluation, gaze cascade, spatial cueing, emotional priming
Rats and Humans
  • The Noha Project, context, prior learning, exploration, gaze cascade, risky choice
  • The Tokka & Ryu Project, stress, cortisol, bias versus sensitivity in decision-making

Philosophy of Mind

Beyond Prediction

(or the 20/20 Hindsight of Meaning through Post-Dictive Fixation)

In recent years, predictive coding has gained considerable popularity in neuroscience and philosophy of mind as a powerful theoretical approach (e.g., Hohwy, 2013; Clark, 2016). The basic concept of “predictive coding” is that the brain endeavors to minimize prediction errors: it tries to anticipate the current situation so as to reduce the loss of energy required to deal with the unexpected. However, one recurring problem with predictive-coding models is that, in order to minimize prediction errors, the most efficient strategy would be to avoid unpredictable situations as much as possible. Yet, as pointed out several decades ago in a critique of behaviorist models (Berlyne, 1966), curiosity is a very typical and basic characteristic of many animals, particularly also humans: we tend to seek a certain level of new stimulation even if there is no guaranteed material or immediate benefit, sometimes even if there is a considerable risk of adverse outcomes. Moreover, much brain activity is devoted to a type of coding in which the relevant information is fully known and requires no analysis in terms of prediction (e.g., savoring the taste of wine, or ruminating on a negative experience). Humans often spend large amounts of time in such “postdictive” information processing. How can these phenomena be reconciled with the framework of predictive coding? I propose to use the concept of “intrinsic reward” as a key addition to predictive-coding models, where the ultimate goal of prediction is not to minimize error, but to maximize reward. A new model based on predictive coding with intrinsic rewards would explain not only how the typical utilitarian behaviors work, but also how seemingly spurious activities such as reprocessing can be tied to the mechanisms of predictive coding. Essentially, the reprocessing would lead to the self-organization of a richer, intrinsically rewarding experience of meaning, with real-world information as a constraint. This “intensive approach” to information processing would tend toward the expansion of meaning and predictive power by continually trying to “fail better.”


A Mismatch Between Micro-Motives and Macro-Behavior in Animal Ethics

(or Why the Three R’s Are Easy to Like but Hard to Achieve)

Viewed at the microscopic level of each individual lab, the principles of the 3 R’s may appear to work reasonably well. Yet, turning to the macroscopic level of the entire field, the principles are essentially ignored. This is largely due to our research culture, which works on the basis of free inquiry, with every principal investigator as an autonomous entity. Researchers organize their studies at the microscopic level: each lab steers its own course, usually staying with a particular paradigm and animal model, aiming to get the smallest sample size that provides statistically reliable results for any experiment, and continuing to refine the techniques and knowledge extraction. At this level, then, we see Reduce and Refine at work. However, the investigations in one lab may overlap with those done in another lab; sometimes the fear of being scooped will lead to rushing the completion of a study. As a result, the present research culture leads to considerable levels of redundancy and a vast literature with too many premature or trivial research papers. In addition to the problem of redundancy, the current research culture makes it practically impossible for an individual researcher to really consider the Replace principle. Shifting to another animal model (e.g., from nonhuman primates to rats) would imply a great individual cost, in time and effort, at the expense of the publication flow. Young scientists usually cannot afford to take this kind of risk; established scientists may lack the energy, motivation, or flexibility to reconsider their way of working. However, support from the general public for animal research hinges on the quality of its output. Good research with important payoff will be welcomed; poor research with suboptimal payoff will lead to disinterest and fading support. Here, it should be noted that the general public naturally views scientific research at the macroscopic level. At this level, people ask tough questions: not lab-by-lab, but for the entire field, about the returns for the vast amounts of resources that are invested in animal research. People outside science also wonder about the efficiency of the three R’s, or why, for instance, nonhuman primates are being used for certain types of research when rodents, at first sight, seem a viable alternative. Indeed, several recent calls from within the neuroscience community have suggested that the levels of public support are dangerously low already (e.g., Holder, 2014; Roelfsema and Treue, 2014). The calls for support, however, were limited to a demand for more effort toward public outreach. This is essentially a conservative attitude, aiming to change the communication about science, without actually addressing the 3 R’s. Instead of a conservative attitude toward animal ethics, I argue that we can adopt a proactive strategy to promote the three R’s at the macroscopic level. The task is to make collective decisions about which research could offer a justifiable contribution: not at the level of the Principal Investigator, but at the macroscopic level of organizations: research communities, universities, funding agencies.