Architectures underlying cognition and affect in natural and artificial systems
Biological organisms differ in many ways. Members of the same species can differ according to their stage of development, according to the problems and resources (including information) encountered during their development, according to details of their genome, and details of previous development, growth, and learning opportunities and also in details of their particular environments with different threats, opportunities, resources, obstacles, competitors, helpers, current needs, and so on. Variations across species are even greater. Over billions of years, biological evolution on this planet has produced a staggering variety of forms of life, differing in physical size, change of size during the life of individuals, life span, sensory apparatus, modes of development and motion, types of environment, modes of interaction with the environment including conspecifics and other life forms, food, prey, predators, forms of information storage, modes of reproduction, and many more. All of these differences (most of which are structural not numerical) can affect mechanisms, internal states or processes, and externally visible forms of behaviour or expression, including affective states and processes related to motivation, goals, plans, preferences, desires, attitudes, values, hopes, ambitions, decisions, intentions, concerns, moods, and other affective states and processes. Is this an area that is susceptible of scientific study and accurate modelling, or is there merely a hopelessly unstructured mess/tangle of special cases understood in depth by some novelists, poets, playwrights counsellors and therapists, but unfit to be the subject of scientific investigation? A similar question might have been asked about chemistry centuries ago when alchemists were faced with a tangled mess of special cases with no means of expanding knowledge except by doing more experiments. But that situation was changed by discoveries about the atomic structure of matter, including the details summarised in the periodic table of the elements, along with advances in chemical understanding based on many experiments and applications of new ideas from quantum mechanics - producing explanations that were not possible in the framework of Newtonian mechanics. Chemical reactions could not be explained by Newton's laws of motion, but new explanatory theories emerged from information about the structure of atoms related to the facts assembled in the periodic table of physical elements, later elaborated by developments in quantum physics able to explain chemical structures and mechanisms including some that are crucial for biological evolution analysed in 1944 by Schrödinger. Since then, although huge gaps remain in our biological knowledge, there have been tremendous advances based on theories in physics and chemistry about possible structures and their interactions, often forming new structures essential to processes of biological reproduction, growth and development. In contrast, much (so-called) scientific study of minds has relied on correlation-seeking experiments and the use of independently variable components of vectors to describe complexity - which would be hopelessly inadequate even for the study of complex molecules. There is also a wide-spread assumption that all motivation needs to be thought of in terms of the relative attractions (or repulsions) of various kinds of reward (or punishment) with a common (positive or negative) utility measure. This can be compared with the ancient assumption that all physical masses seek the centre of the universe, which is hopelessly inadequate for the explanation of known physical and chemical phenomena. Even if there are reward mechanisms that explain some motives and preferences, there is much they cannot explain. For example, if someone really enjoys doing mathematical research only because doing it produces some reward (whether chemical or psychological) then in principle he or she should be just as willing to get the reward by doing something much easier than struggling with mathematical problems - e.g. drinking some potion, or stepping into an otherwise harmless machine. But nobody who really enjoys doing mathematics would swap the activity for one of its side-effects. Of course, there may be such people for whom doing mathematics is not its own reward, but they still want to do it, e.g. because they enjoy the admiration it produces in others, or because it is a necessary condition for achieving some other goal, such as getting into university, or a useful aid to attracting an intelligent mate. I believe I first encountered that refutation of popular reward-based theories of motivation in Ryle. There are similar objections to widely used utility-based mathematical theories of decision making, such as theories based on "payoff matrices" (criticised in my 1978 book). I suggest that evolution frequently made use of architecture-based motive-generation mechanisms (ABM) that, unlike reward-based motivation (RBM) allow new motives to be triggered by perceived opportunities or situations without the individual having any ulterior reward-motive. It suffices that ancestors who had such mechanisms acquired useful knowledge that later brought benefits that the individuals could not have predicted, or even thought about. As a result they succeeded in life and produced offspring who were likely to share the same motive-generators. So the ABM mechanisms trigger motives that have been beneficial in one's ancestors, not motives whose achievements produce some special reward-juice. These can be thought of as genetically programmed internal reflexes comparable to genetically programmed physical protective and feeding reflexes.
Across all the variation in forms of life, are there any common principles? One seems to be the ability to acquire and use information for purposes of control, such as generating options for consideration, selecting options, working out consequences of various options. There is also information-based control of chemical and physical processes of reproduction, development and growth. Information is used during interaction with inert physical features of the environment and also during interaction with predators, prey, offspring and other conspecifics - which often requires information about information, e.g. using information about what something else wants or can perceive. In many cases passive individuals are acted on by the environment, for instance when seeds are dispersed by wind, or when seasonal or daily changes in temperature or availability of light, air or water currents, or supply of nutrients or dangers are out of the control of individuals and they can at most resist, react to avoid or react to make use of (e.g. consume) contents of their environment. In more complex cases information about threats, opportunities, resources, and obstacles can be acquired and put to use, either immediately or at a later time when a need arises. Coping with threats from other organisms, may involve purely physical avoidance or escape actions. But in some cases it requires other-directed meta-cognition: inferring intentions, knowledge, reasoning processes and choosing means of avoidance or escape accordingly. So information of many kinds plays many different roles in living things, unlike non living but interacting physical objects and processes, such as weather features, geological features shaped by and shaping one another, including tectonic motion, earthquakes, volcanoes, floods, tornadoes, other weather patterns, seasonal changes caused by motion around the sun and tides caused by rotation of the moon around the earth. This notion of information is much older than the notion developed by Shannon around 1948. Since Shannon, information is often discussed as if it were primarily the content of messages, with senders and receivers. But sending and receiving messages would be pointless if the message contents had no other use than to be transmitted, received, encoded, decoded, compressed, decompressed, stored, retrieved, etc. The fundamental fact about information that is often ignored in discussions of the nature of information is that it can be used in CONTROLLING what happens. This can take many forms: in some cases information directly triggers a response, e.g. a defensive reflex such as blinking or rapid withdrawal, or an opportunity taken, such as motion towards water, food, shelter or a mate, or use of a body part to acquire or consume something edible. In other cases the information can be stored for future use, e.g. information about where a resource or a danger is located, or information encoded in a genome that is used at a particular stage during during reproductive processes to control aspects of development and growth of tissues and parts of new individuals. Other forms of information in a genome can generate and control behaviours of organisms once they are functional, e.g. controlling breathing, pumping of blood, digestion, begging for food, following parents, and triggering new motives to be acted on later (ABM). Such uses of information could be ignored in Shannon's famous work on information  because he was working for a company (Bell Telephone Company) providing information services, for whom the main problems were reliable transmission and storage, not use of information. The USE was the concern of their customers. In contrast, the novelist Jane Austen was very much concerned with ways in which her characters could not only transmit, acquire and store information, but also use it, as discussed in http://www.cs.bham.ac.uk/research/projects/cogaff/misc/austen-info.html She frequently referred to information, not in Shannon's sense, but in the much older sense in which information is used, not merely transmitted or stored.
There are two fundamentally different roles that useful information can have, as Hume noted in distinguishing "is" (information about what is the case) from "ought" (information about what to do) in his argument that "ought" can never be derived from "is". This distinction was elaborated by Elizabeth Anscombe  as a difference in "direction of fit". For an information user there are some information contents (which we can crudely label "desire-like information") whose role in an organism determines what should be done to the world to make the world match the information content, and other information contents (which we can crudely label "belief-like information") whose role is such that the information should be altered when there is a mismatch with how the world is. Both sorts are required for intelligent, or purposeful action, or deliberate inaction. Moreover, in both cases there is always the possibility of an organism not being in a position to determine whether the information item does or does not match reality - e.g. whether some belief is true, or whether some desire or goal has been satisfied. This can generate a new second order desire-like information state, which specifies that an information gap needs to be bridged. That new state can trigger action to fill the information gap - which may either be done relatively simply (e.g. by looking, sniffing, touching, etc.) or by engaging in some sort of information-gathering research, e.g. to find out whether food is available nearby and if so where it is. As these examples show, there can be many processes by which combinations of belief-like and desire-like information states can generate actions to determine whether the belief-like states actually fit the world or actions to make the world fit the desire-like states. A rich theory of varieties of cognition and affect can be based on the implications of this distinction as pointed out (by Sloman, Chrisley and Scheutz) in  (building on ideas developed by Beaudoin). The time scales involved and the scale of action required to bridge these information gaps (finding out whether X is true, or making X true) can vary enormously according to the complexity of the information specification and the amount of effort involved in checking whether X fits the facts or making X fit the facts. Things get even more complex if individuals can have a large and changing collection of desire-like and belief-like information states, unlike a simple thermostat which has a target temperature and a sensor providing information about the gap between the current and target states, along with a mechanism for turning on or turning off a heat generator or heat remover. It is often assumed that all desire-like information states are concerned with achievement or maximisation of some measurable reward or utility, but life is far too complex for that: organisms have many different needs at different stages of development and at different times and places, often needs that coexist and conflict, e.g. a need to approach a source of food when energy stores are low and a need to avoid detection by a dangerous predator or rival. The assumption that these needs can be compared on a common scale are as misguided as the assumption that strength of materials and fuel energy of materials can be compared on a common scale.
This is very different from a form of learning or development that uses a uniform method for repeatedly finding patterns at different levels of abstraction, e.g. using statistical generalisations. Instead, on this model, the genome encodes increasingly abstract and powerful creative mechanisms developed at different stages of evolution, that are "awakened" (a notion also used by Kant) in individuals only when their time is ready, so that they can build on what has already been learned or created in a manner that is tailored to the current environment.
As living things become more complex, increasingly varied types of information are required for increasingly varied uses. The processes of reproduction normally produce new individuals that have seriously under-developed physical structures and behavioural competences. Self-development requires physical materials, but it also requires information about what to do with the materials, including disassembling and reassembling chemical structures at a microscopic level and using the products to assemble larger body parts, while constantly providing new materials, removing waste products and consuming energy. Some energy is stored and some is used in assembly and other processes. The earliest organisms can acquire and use information about (i.e. sense) only internal states and processes and the immediate external environment, e.g. pressure, temperature, and presence of chemicals in the surrounding soup, with all uses of information taking the form of immediate local reactions, e.g. allowing a molecule through a membrane. Some of the changes in types of information, types of use of information and types of biological mechanism for processing information have repeatedly altered the processes of evolutionary morphogenesis that produce such changes: a positive feedback process. A familiar example is the influence of mate selection on evolution in intelligent organisms, since mate selection is itself dependent on previous evolution of new cognitive mechanisms. This is a process with multiple feedback loops between new designs and new requirements (niches), as suggested in [Sloman 2000]. Compare also the author's presentation at the Computing and Philosophy symposium at AISB 2017. As Figure 1 suggests, evolution constantly produces new organisms that may or may not be larger than predecessors, but are more complex both in the types of physical action they can produce and also the types of information and types of information-processing required for selection and control of such actions. These ideas, and those in [Sloman, (2014--)] suggest that one of the effects of biological evolution was fairly recent production of extremely, but not totally, abstract construction kits that come into play at different stages in development, that produce much more rapid changes in variety and complexity of information processing across generations than ever before. This idea is fairly familiar as regards the role of a common genetic inheritance in enabling hugely varied languages to be developed by humans in different cultures. This pattern can be generalised to other aspects of development, as suggested in Figure 1. (There are loose connections with Chomsky's ideas on evolution and development of language. I don't think he ever realised that human language evolution and development must be a special case of something deeper and more general.) The talk will present some ideas about possible information processing architectures capable of supporting diverse kinds of variety among humans and other animals. I suggest that within a century or two our ideas about how human minds work, and the requirements for modelling them in intelligent machines, will have changed at least as much as our ideas about physics and chemistry have changed since the time of Galileo. Some suggestions, regarding mechanisms and architectures can be found in references.
Some annotated extracts are available