School of Computer Science THE UNIVERSITY OF BIRMINGHAM Ghost Machine

Cognition and Affect Workshop
Following AISB 2017 Discussions

Monday 24th April 2017
School of Computer Science
University of Birmingham
(DRAFT: Liable to change)

Organiser: Aaron Sloman
School of Computer Science, University of Birmingham

Installed: 13 Apr 2017
Last updated: 23 Apr 2017; 4 May 2017; 11-12 May 2017; 29 May 2017


The workshop summarised here was held on Monday afternoon 24th April. People who had attended or communicated about the workshop were invited to submit comments related to the discussion at the workshop or the document on anger mentioned in the workshop "homework" below.

I have received some interesting comments from one of the people who attended the workshop. They are now available in a 'Comments' page here.

If anyone else who attended, or has comments or questions, wishes to send them to me, please feel free.

There is also an old 'CogAff' email list maintained in this school. If you would like to join it please let me know (a.sloman [AT] It has recently been used only for announcements, but it also can be, and has been, used for discussions. (It incorporates a spam blocker.)

NOTE: There will be an independently organised philosophical workshop on Cognition and Affect at this university on 9th June. Details here:

A partial progress report on the Birmingham CogAff project

     Monday 24th April 2pm - 4.25pm then "forward look" from 4.30pm
    School of Computer Science
    University of Birmingham
    ROOM 245
This workshop followed a Workshop held at the AISB Convention in Bath the previous week:
Computational Modelling of Emotion: Theory and Applications
at the AISB Convention at Bath University 19-21 April 2017.
(Presented in Aaron Sloman's introductory talk.)
The Cognition and Affect project (originally Attention and Affect and then generalised) began at the University of Birmingham in 1991 led by Glyn Humphreys (Psychology) and Aaron Sloman (Philosophy and AI), funded initially by the Rennaisance Trust and the UK Joint Research Council Initiative, the Canadian National Research Council and Commonwealth Scholarship (grants to Luc Beaudoin the first PhD student), and later by the School of Computer science (PhD studentships), DERA (funding post-doctoral research by Brian Logan), the Leverhulme trust (for post-doctoral research by Matthias Scheutz and Ron Chrisley, a Sony Studentship (for Nick Hawes), and a DSTL(grant to Catriona Kennedy). Project web site

The project attempted to investigate evolutionary origins, architectural (information processing) underpinnings, and the variety of types of affective phenomena that needed to be accounted for, including those commonly referred to as emotions, attitudes, values, preferences, ambitions, desires, moods, and personality traits, among others.

Later, work on attachment (co-supervised by Gillian Harris in psychology), contributed to work on development and attachment by Dean Petters.

The key idea -- inspired by Herbert Simon (1967) -- was that these are all phenomena concerned with information-based control of actions and information-processing ("controls of cognition" according to Simon). Where possible, we tried to demonstrate and test ideas with computational models, though these were mainly illustrative toy examples, since they were not sufficiently complex to model or explain affective states and processes in humans or other animals. (This could be said of ALL such models so far.)

The hope was that an architecture-based understanding of emotions and other affective phenomena would eventually be comparable to how the architecture-based understanding of matter could account for the Periodic table of the elements, as opposed to observations and statistical theories extracted from data (as in alchemy, before modern chemistry).

The most important types of architecture and the hardest to identify are not physical mechanisms but components of virtual machinery -- created by biological evolution long before human engineers discovered the power of virtual machinery during the 20th Century. An introduction to the philosophical importance of increasingly complex forms of virtual machinery can be found in this paper on Virtual Machine Functionalism (VMF), contrasted with Atomic State Functionalism (ASF).

A list of research students associated with the project since 1991 is here:
Tools developed here between 1991 and about 200 to support development of fairly complex information processing architectures, with multiple concurrent types and levels of control included The Poprulebase Rule interpreter, which was part of the Simagent toolkit.

Since 2011 the Turing-inspired Meta-Morphogenesis project has been developing some of the ideas in the context of an attempt to understand major transitions in biological information processing between the very earliest organisms or proto-organisms and current life forms of many kinds.

Several of those who were involved in the project were participants in the
Symposium on Computational Modelling of Emotion: Theory and Applications
at the AISB Convention at Bath University 19-21 April 2017.
(Three accepted papers, and one invited keynote talk.)

A closely related symposium at the AISB convention is on
The Power of Passion: Human Reason and its Emotional Foundations

(Does not presuppose prior formal study of emotions)
(Added 23 Apr 2017)

Ask yourself: What is it for X to be angry with Y?
Can you be angry with Y without being angry about something?

Which of these can be, or necessarily must be, involved in anger:

    beliefs, preferences, desires, intentions, plans, decisions, hopes, values,
    fears, regrets, percepts, inferences, reasons, reasoning processes,
Where possible justify your answers with examples.

Has any necessary feature of anger been left out of the above list?
Are there some features that are not necessary but may be involved as part of a state of being angry? (I.e. can episodes of anger vary in their contents as molecules can vary in their contents?)

Does being angry have to be a static state (like being tall, or heavy, or dead), or does it, sometimes, or necessarily always, involve processes? Physical processes? Mental processes?

Which of the states and processes vary along linear scales, or combinations of linear scales, e.g. with numerical values, and which do not, e.g. because they include grammatical or semantic structures)?

How much of this can be, or has been implemented in artificial intelligence systems -- e.g. robots, or artificial companions?

Are any of the features impossible for such machines?

Do answers to these questions have implications for any well-known theories, or methodologies, in psychology, neuroscience, AI, or philosophy?

PARTIAL ANSWERS to the above homework questions are available online
in an article published in 1982, here:
     "Towards a grammar of emotions"
     New Universities Quarterly, 36, 3, pp. 230--238, 1982

This exercise provides an example of the important philosophical technique of "conceptual analysis", which can be done well or done badly. A fairly detailed tutorial on how to do conceptual analysis (mainly inspired by work of J.L. Austin, Gilbert Ryle, and Ludwig Wittgenstein) can be found in Chapter 4 of The Computer Revolution in Philosophy (1978) here:


Cognition and Affect Workshop

Our aim in this workshop is
(a) to present some of the key ideas of the Cognition and Affect project summarised above,
(b) to present some recent work emerging from it, presented by Luc Beaudoin and Dean Petters.
(c) to report briefly on the AISB Symposium and
(d) finally, to discuss possible future directions, in an optional session at the end.
-- 2pm-2.20
Aaron Sloman (CS@Birmingham)
    Summary of the history of the project and some of the key ideas
    that have emerged, including unanswered questions.

    The project overview can be found here
    (Summarised briefly above.)
    -- 2.20-2.30
    Comments and questions arising

-- 2.30-3pm
Luc Beaudoin (Simon Frazer University and CogZest)
    (Work done with Sylwia Hyniewska and Eva Hudlicka)
    Perturbance -- a unifying, architecture-based concept.
    Luc Beaudoin's slides (PDF)
    Related papers
    -- 3-3.15pm
    Comments and questions arising

3.15-3.25pm SHORT BREAK

-- 3.25-3.55pm
Dean Petters (Birmingham City University)
    (Work done with Emily Coyne Coyne-Umfreville and Jason Martens)
    From Attachment to Defensive Exclusion and Pride
    For more information related to Dean's talk see these recent papers:
    Towards Modelling Adult Attachment Patterns as Control States.
    Bowlby's Attachment Control System Approach: An Alternative History of
    Emotion Modelling

    -- 3.55-4.10pm
    Comments and questions arising

-- 4.10-4.25 General discussion (Part 1)

SHORT BREAK followed by optional final session.

-- 4.30-5pm
    General discussion (Part 2), including What Next? Where Next?
    E.g. what are the hard and important unanswered questions worth
    pursuing, and how should that be done?

Possible continuation in University bar in Staff House.