It is Labour Day week-end! This is a time to celebrate great knowledge work. It is also a time to ask “Have we been using the best conceptual tools?” And “how can we build better knowledge?” Let’s keep these questions in the back of our minds as we consider the following
Fifty years ago this year, Herbert A. Simon published “Motivational and emotional controls of cognition”. There, he expressed one of the most important insights about emotion and motivation (and hence perhaps of psychology) of the last century. It is that in order to deal with its multiple purposes in real-time, given nature’s unpredictable dangers and opportunities, the human central nervous system (CNS) requires an interrupt system; emotions are nature’s designs to meet these requirements. This is not to say that every idea in Simon’s paper was original. The importance of interrupt mechanisms in emotion had been recognized by several other luminaries. Simon’s paper was significant because it approached these phenomena using information processing/ AI (Artificial Intelligence) concepts. He saw that natural constraints have implications for the design of human minds. He recognized that AI’s conceptual toolkit can be used and extended to understand motivation and emotion. It is noteworthy that his paper was published in Psychological Review.
In 1980, D. A. Allport described computational modeling as the “single most important development in the history of psychology” (p.31). In a nutshell, the designer stance involves specifying the requirements and competencies that need explaining (the problems), exploring the space of possible designs that address those requirements, studying the trade-offs amongst the designs, attempting to implement some of the designs in computer programs, analyzing the results. This process is repeated over and again. This approach, which combines engineering and empirical science, is described in Aaron Sloman’s “Prospects for AI as the general science of intelligence” and in Chapter 2 of my Goal Processing in Autonomous Agents. Notice that this approach goes significantly beyond David Marr’s oft cited levels of analysis.
As one might expect, Simon’s ideas appealed to (some) emotion psychologists who were conversant with AI. For example, Keith Oatley and Johnson-Laird’s communicative theory of emotions makes extensive use of Simon’s key insights, as does Michel Aubé’s commitment theory of emotions. There’s a connection with the Cognition & Affect Project here: Oatley and Sloman were both at Sussex University in the 1980s. In 1993, I introduced Aubé to that project.
However, Oatley, Johnson-Laird and Aubé are amongst the exceptions. Uptake of Simon’s idea elsewhere in psychology has been quite disappointing. Sure, Simon’s paper has been cited 1,704 times (according to Google scholar 2017-09-02 3-45 PM PT). This may sound like a lot, but not if you consider the number of papers published on emotion every year! I’ve attended several pertinent conferences where not only was the paper not mentioned, but I could rarely detect its influence. Moreover, as Sylwia Hyniewska, Eva Hudlicka and I argued at AISB-2017, an evolution of Simon’s concept, perturbance, can be used as basis to explain some of the most important characteristics of normal and pathological mentation, namely “repetitive and intrusive thought”. (I discuss perturbance a bit more below.)
To illustrate the point further, Simon’s paper only received a token citation in the following noteworthy and otherwise very helpful publications on emotions:
- Ortony, A., Clore, G. L., & Collins, A. (1988). The cognitive structure of emotions. Cambridge, UK Cambridge University Press.
- The Emotions (1986) and The Laws of Emotion (2007) by Nico Frijda.
- Wells, A., & Mathews, G. (1994). Attention and emotion: A clinical perspective. Hillsdale, NJ: Lawrence Erlbaum Associates Publishers.
More recently, the following otherwise quite helpful books completely ignored Simon’s theory:
- Pessoa, L. (2013). The cognitive-emotional brain. Cambridge, MA: MIT Press,
- Lisa Feldman Barrett How Emotions Are Made , and
- Panksepp, J., & Biven, L. (2012). The archaeology of mind: Neuroevolutionary origins of human emotions.
If one subscribes to the ideas that emotion is a scientifically salvageable concept, and that emotions have multiple components, one might expect the evolution of Simon’s concept, perturbance (mentioned below), to be recognized as a component of emotion. Yet the major component theorist of emotion, Klaus Scherer, wrote of Simon’s paper in 1984, “Much of our experience with affect, however, seems to contradict the notion of emotion as interruption” (p.295). And to this date, perturbance is not included in Scherer’s theory. To be fair to his theory, one might argue that its low- and high-level prioritization functions capture what is helpful of Simon’s theory. In particular, the goal/need significance check of Scherer (1987) is related to the concept of insistence that Sloman first proposed in 1981. However, Scherer’s concept and overall theory are significantly different —for example, his theory doesn’t posit attention filtering mechanisms.
The significance of Simon’s paper was recognized by several “affective AI” research teams. Most R&D in affective AI, however, does not address the requirements of “grand AI” to which Simon’s pertains. Alas, affective AI currently focuses on much simpler problems, such as the measurement of affect and “emotion recognition”. I know of no AI lab that is working on the detection of perturbance, though I often mention it to AI “emotion recognition” experts. (For example, Rosalind W. Picard’s seminal Affective Computing book mentions our concept of perturbance. She does work on emotion recognition. So, after her keynote presentation at CogSci 2015, I asked whether she has tried to, or would create, machinery to detect higher aspects of human emotion such as perturbance. She said this would require new behavioral tests.) In short, my impression is that the detection of perturbance is not on the radar of emotion recognition researchers. Maybe my renewed research programme on perturbance (e.g., with respect to limerence and insomnia) will have an impact here… we’ll see.
I believe it is fair to say that Aaron Sloman and his Cognition and Affect project team (of which I was a member) pursued Simon’s insight more extensively than any other group in the world. (As noted, Sloman had discovered overlapping ideas and acknowledged Simon in 1981.)
Sloman’s theory however goes significantly beyond Simon’s. Some of the ways in which it does are discussed in Sloman & Croucher’s “You don’t need a soft skin to have a warm heart: Towards a computational analysis of motives and emotions”. In particular, Sloman originally argued that emotions are not special purpose mechanisms, they are emergent side-effects. Sloman argued that any system that meets the requirements of autonomous agency will be subject to emotion-like states. This applies not merely to humans and similar organisms, but to the space of possible minds —e.g., AI and life on other planets that deals with similar requirements (should it exist). Sloman proposed the need for minds to compute insistence measures that heuristically co-determine whether and how a motivator will be considered by higher-order mental processes.
Circa 1992, because —as James Russell convincingly argued in 2009 and Elizabeth Duffy before any of us — the term “emotion” means too many different things, I proposed that we relabel Sloman’s (designer-stance) concept of emotion as “perturbance”. Sloman adopted the terminology for a while, but later he replaced it with “tertiary emotion”. However, I have since reverted to using the term “perturbance”, which is more general (this is in the spirit of the space of possible minds approach: architectures with more than three levels can also support perturbance), and avoids the highly problematic term, “emotion”.
In an upcoming blog post, I will describe other ways in which the concept of perturbance differs from Simon’s notion of interrupts. Meanwhile, interested readers can check out “Perturbance: Unifying Research on Emotion, Intrusive Mentation and Other Psychological Phenomena with AI”.
I would like to point out another important insight that is implicit in Simon’s paper. The idea is that motivation and emotion need to be considered together. Notice how Simon talked about motives. In the 1980s, Sloman would publish papers on “motive processing”, and I joined him on this in the 1990s. (For example, we published “A study of motive processing and attention” and my thesis is “Goal processing in autonomous agents”). The perturbance theory of emotion essentially states perturbance is a state that emerges from motive processing.
Baumeister recently published an already influential (2015) paper, “Toward a general theory of motivation: Problems, challenges, opportunities, and the big picture”, in which he argues that “cognition, emotion, agency, and other psychological processes exist to serve motivation.” With a few qualifications, this is quite consistent with my view.
Looking back: I feel gratitude towards Herbert Simon, Aaron Sloman and many others whose work I have leveraged in my own academic work this year. At AISB-2017, I had occasion with Hyniewska to show that the concept of perturbance is relevant to many problems in psychology. At ISRE-2017, we argued that perturbance is an important contributor to insomnia.
We are continuing to build on the knowledge work of giants, writing a paper that combines many diverse literatures in AI and psychology that had never previously been applied to sleep onset and insomnia. Working title: “Towards a somnolent information-processing theory of human sleep-onset: A broad, architecture-based alternative to cognitive and arousal theories of proximate insomnia mechanisms”.