School of Computer Science


Papers and presentations on affect, in the
Birmingham Cognition and Affect Project
started here in 1991, building on earlier work
at Sussex University.

Aaron Sloman
School of Computer Science, University of Birmingham

This document is available in two formats:

I've always thought that great novelists know more about emotions, and
other affective phenomena than psychologists and neuroscientists ever will.

(DRAFT: Liable to change)
Last updated: 11 Apr 2024

More Recent Items, below

A major new web site is now under development here, subsuming many of the issues discussed or mentioned below:
This is includes references to the largest and most complex collection of related but varied ideas I have ever assembled. I don't know whether anyone else has noticed and written about all those connections, although many others have addressed significant subsets -- different subsets!

Other (mostly much older) papers and presentations more concerned with non-affective aspects, e.g. perception, reasoning, learning are included in:
with talks/presentations here

What is this?

This (still incomplete) document lists some of the papers written at the University of Birmingham (1991 onwards), mainly within the Cognition and Affect (CogAff) project, concerned with a general framework for combining affective mechanisms (involved in wants, hopes, fears, likes, dislikes, emotions, moods, evaluations, ...etc...etc.) with cognitive phenomena (e.g. perceiving, acting, planning, learning, predicting, explaining, learning, hypothesizing, ...etc...etc.)

From its early days the project was committed to a theoretical-comparative approach, i.e. studying not only phenomena that occur in humans, or in humans and other animals, but studying the whole space of possible designs (especially possible information processing architectures) for systems capable of having informational states and processes (involving desires, preferences, values, moods, beliefs, skills, knowledge, uncertainty, etc.) as well as having physical states and processes (including body form, actions available, sensors, motors, internal physiological systems, size, shape, weight, strength, changing physical needs and stored resources, etc.), in various environments (e.g. under water, on land, while flying) on flat terrain, in various types of non-flat environment (e.g. rocky, snowy, icy, muddy mountain slopes, etc.), with various ranges of temperature, resource availability, threats and dangers, etc.

This approach contrasts strongly both with shallow theories of embodied, enactive, expressive ("skin deep"??) aspects of affective states and processes, and utility-based theories of motivation, and also with informationally-restricted theories, e.g. assuming that all information and information processing is logical, or numerically measurable, or symbolic, or probability-based, or restricted in some other way (as happened at various stages in the history of AI theories of emotion).

The emphasis on designs and what they can and cannot do also contrasts with a focus on classifying and correlating measurable, or observable or introspectable or physiological states of humans or other species. A design is something deeper and more abstract, and may have multiple different sub-types of instantiation of the design. Compare the space of possible designs for utterances in the English language, or some other language -- which may have a different space -- or the space of types of communication modality: spoken, written, signed, signalled (e.g. using semaphore), etc.

The CogAff approach to the study of the space of possible minds, and possible mind-based states and processes, aims to bring about a change in science that is partly similar to how the periodic table of the elements initiated deep changes in chemistry. Unfortunately, most researchers on cognition and affect are not educated with the required attitudes, concepts, knowledge and skills, e.g. abilities to design, build, test, and debug working models.

The list below includes some relevant earlier papers from the time (pre-1991) that the project was based at Sussex University (1962-91).

Thanks to help from Dean Petters, the following includes a new architecture-schema diagram, showing overlaps between input, output, and "central" components of the architecture:
(also PDF)
N.B. There is no intended claim that all instances of the design must include instances of all the types of architectural feature: this is intended to be more like the specification grammatical structures for sentences, clauses, phrases, or larger linguistic products (e.g. stories) available in a particular language. However, it is arguable that the variety of types of design specified here is not rich enough to accommodate all relevant types of design for systems capable of emotions or other affective states.

Who knows what?
Great poets, playwrights, novelists and composers often have a much deeper understanding of varieties of affect than (current) philosophers, psychologists, neuroscientists and cognitive modellers. But their deep understanding is implicit and usually only indirectly articulated, e.g. in plot construction, dialogue construction, thought-streams, musical compositions, etc.

One goal for AI is to find ways to make that implicit knowledge explicit and demonstrate the implications by building a succession of increasingly realistic, increasingly complex, working systems. But "working" does not merely mean showing behaviours (including linguistic behaviours) thought to correspond to various cognitive and affective states and processes. There must also be the right kinds in internal/invisible information processing including forms of reasoning, clashes of motivation, resolution of conflicts, growth and modification of attitudes and values, etc.

A full model should include "genetic time-bombs" i.e. potential at various late stages of development to produce new motives, values, preferences, abilities, etc. In humans the motivational (and consequential emotional) changes at puberty are obvious examples. But genetic time-bombs may have even longer fuses concerned with how to use a large volume of acquired knowledge, skills, experience, etc. after enough time has been spent on acquisition. (The corresponding mechanisms in humans seem to be highly erratic, and often over-ridden by self-interested motivation.)

Beware of arguments purporting to prove that "AI systems (including robots) can't do X" by proving that "Computers can't do X".

They are as valid as arguments showing that "Molecules can't do Y (e.g. have emotions, or discover geometric theorems), therefore objects composed of molecules can't do Y".

Turing machines, are irrelevant to AI for reasons explained in this paper:

On the other hand complex systems composed of large numbers of interconnected digital computers, sensors and motors, in a machine located in a complex, changing, partly unpredictable environment, are another matter. (As H.A. Simon pointed out in "Motivational and emotional controls of cognition", 1967.)

The deep, still unanswered, question is: what sort of (self-extending) information-processing architecture could replicate the required functionality in future machines?

The answer may be related to the question whether sub-neuronal molecular computations are essential to biological competences of poets, playwrights, squirrels defeating squirrel-proof bird-feeders, and ancient mathematicians.

Clearly some motives e.g. hunger, are based on molecular processes. Perhaps far more aspects of mentality are than we realise. I have raised that question in connection with ancient spatial reasoning abilities underlying discoveries in geometry and topology, here:
Key Aspects of Immanuel Kant's Philosophy of Mathematics
(That's a companion-piece to a discussion of Turing's distinction in his PhD thesis between mathematical intuition and mathematical ingenuity: he suggested that computers could replicate the latter but not the former.

Some incomplete remarks on requirements for types of computer capable of replicating spatial reasoning in humans and other intelligent animals:

Installed: 11 Mar 2018
Updated: 20 Jan 2019; 7 Feb 2019; 6 Apr 2024
27 May 2018; 19 Aug 2018; 2 Oct 2018; 24 Oct 2018;
More papers still to be included, annotated, etc.
This paper is
This is part of the Birmingham Cognition and Affect (COGAFF) project:

(Main contents below)

Some of what follows is now explicitly or implicitly subsumed (and in some cases updated) in the more recent document mentioned above
(Still being updated.)

Work on evolution of consciousness, from its very simple (or precursor) life/proto-life forms (4.5 page abstract for invited talk at APA conference 14th Jan 2021):
The meta-configured genome project:
Some of the ideas are updated in the metamorphosis.html document cited above.

Work in 2021 on the deep problem of explaining how chemical processes inside eggs can produce cognitively competent hatchlings (e.g. baby avocets), apparently with significant knowledge about where food is to be found and how to obtain it:

Work by Luc Beaudoin PhD Thesis 1994
Learning from works of Art
News from Luc Beaudoin's Cogzest project, continuing the work he began here in 1991 and others listed below. Some of his more recent work

On Monday 28 January 2019, two researchers in the School of Philosophy gave a talk in the School of Computer Science, on
Robots, Emotions, and Epistemic Rational Assessability
Slides for the talk: PDF with linked videos Rationality_Robot_emotions.pdf
Speakers: Matilde Aliffi and Helen Ryland
(Both doctoral Researchers, Department of Philosophy, University of Birmingham)
There is a current lack of philosophical research on whether robots could have emotions. In this talk, we argue that the idea that a robot could have emotions is more plausible than currently assumed. We will demonstrate this by giving examples of robots that appear to have some of the emotional components that are usually involved in human emotional experiences. This opens up new philosophical questions specifically about the rational status of these robots? emotions. We claim that if a robot can have emotions, or ?robot-like emotions?, then these emotions may be open to epistemic rational assessment.

Papers and notes on the Cognition and Affect Project

Related work done since mid 1960s, including later work with students and colleagues.
This work was first influenced by AI during 1972-3, when I spent a year in Edinburgh, having my brain rewired, for a new approach to philosophy.
(This list is still incomplete.)
External summaries
Here's a very short summary, with diagrams, of the H-Cogaff architecture schema, produced by someone I've never met

(To be expanded)
Early papers relevant to emotions/motivation/affect/preferences/values...
How to derive "better" from "is"
"Ought" and "Better"
1978 Book: The Computer Revolution in Philosophy
Papers more centrally concerned with varieties of affect
You don't need a soft skin to have a warm heart
Sloman and Croucher
Why robots will have emotions
Sloman and Croucher
Towards a Grammar of Emotions
Motives Mechanisms and Emotions
Prolegomena to a Theory of Communication and Affect
A Proposal for a Study of Motive Processing
Luc Beaudoin (Thesis proposal)
Appendix to JCI proposal, The Attention and Affect Project
Aaron Sloman and Glyn Humphreys
This paper was mostly written by the first author, although it is partly based on, and develops, ideas of the second author.
What are the phenomena to be explained?
1992 Towards an information processing theory of emotions
Silicon Souls, How to design a functioning mind
(Professorial Inaugural Lecture, University of Birmingham 1992)
The mind as a control system
A study of motive processing and attention,
Beaudoin and Sloman (April 1993)
The Terminological Pitfalls of Studying Emotion
Tim Read
Computational Modelling Of Motive-Management Processes
Sloman, Beaudoin, Wright ISRE 1994 Poster
Goal processing in autonomous agents
Luc Beaudoin (PhD thesis)
An Emotional Agent -- Detection and Control of Emergent States in an Autonomous Resource-Bounded Agent
Ian Wright (Thesis proposal)
Information about the SimAgent toolkit
Aaron Sloman and Riccardo Poli (later Brian Logan)
Playing God: A toolkit for building agents
     Information about the SimAgent toolkit
Aaron Sloman and Riccardo Poli
Date: November 1994 to March 1995
SIM_AGENT: A toolkit for exploring agent designs
Aaron Sloman and Riccardo Poli
Towards a Design-Based Analysis of Emotional Episodes,
(Grief paper.)
Ian P. Wright, Aaron Sloman, Luc P. Beaudoin,
Cognition and affect: Architectures and tools
Brian Logan and Aaron Sloman
Architectures and Tools for Human-Like Agents
Aaron Sloman and Brian Logan
PhD Thesis Proposal: Distributed Reflective Architectures,
Catriona M. Kennedy
Patrice Terrier interviews Aaron Sloman for EACE QUARTERLY
(August 1999)
Title: Architectural Requirements for Human-like Agents Both Natural and Artificial.
(What sorts of machines can love? )

Aaron Sloman. Invited conference talk, later published in Human Cognition And Social Agent Technology
Ed. Kerstin Dautenhahn,
How many separately evolved emotional beasties live within us?
Aaron Sloman
Invited Talk: at workshop on Emotions in Humans and Artifacts Vienna, August 1999
Final version published 2002.
Evolvable architectures for human-like minds
Aaron Sloman and Brian Logan
Invited talk at 13th Toyota Conference, on "Affective Minds" Nagoya Japan, Nov-Dec 1999
Published in Affective Minds, Ed. Giyoo Hatano, Elsevier, October 2000
Progress report on the Cognition and Affect project:
Architectures, Architecture-Schemas, And The New Science of Mind
Aaron Sloman
AAAI 2004 Workshop invited talk: What are emotion theories about?
Aaron Sloman
Simulating Infant-Carer Relationship Dynamics
Dean Petters
How to Determine the Utility of Emotions (At AAAI-04)
Matthias Scheutz
The Architectural Basis of Affective States and Processes
Sloman, Chrisley and Scheutz: invited book chapter for "Who needs emotions" (eds. Arbib and Fellous).
2006 2009
Architecture-Based Motivation vs Reward-Based Motivation
Strongly challenges almost all published theories of motivation, especially in experimental psychology, neuroscience, and AI, and some in philosophy. They all grossly over-simplify the biological facts.
Aaron Sloman
Architectures underlying cognition and affect in natural and artificial systems
(Extended Abstract for invited talk at AISB 2017)
Cognition and Affect: Past and Future
Cognition and Affect Workshop, Following AISB 2017 Discussions.
University of Birmingham, 24th April 2017
Anger, an example of conceptual analysis,
(Background material for workshop)
Aaron Sloman
Early papers relevant to emotions/motivation/affect/preferences/values...


A partial index of discussion notes in this directory is in

Maintained by Aaron Sloman
School of Computer Science
The University of Birmingham