Limits and Language

A side effect of the attempt to simplify - in the interests of marketing (that is to communicate) - an idea, is that we obscure the truth - which is after all obscure at best, and whose plain articulation may seem to verge on the mystical.

So perhaps this is the place for that: a graveyard if you like, a place to speak to future generations. A place to use language both plain and obscure.

So with the weight of marketing lifted, let's talk about learning.

The essence of cognition is concern, and hence learning is primarily affective in nature.

All creatures that can learn are necessarily capable of feeling; this is the common cognitive root shared between human and primitive life. I do not mean that a fly experiences emotion in the same way that a person does - this seems unlikely and is inscrutable in any case - I mean that we know that both flies and people may be subject to classical and operant conditioning, and therefore feel.

How so? In considering such learning mechanisms it is always overlooked that we describe stimuli as positive or negative. What we mean by that cannot simply be circular (that the positive increases the frequency of a behaviour and the negative decreases it) but rather that a creature differentiates positive and negative stimuli affectively - there are some things it 'likes' and others it 'dislikes' in other words.

So it is that affective response lies at the very heart of learning and cognition. We are not conditioned by those stimuli which elicit no affective reaction - indeed we don't notice them (in the sense that we perceive them but register no trace). The driver who doesn't learn from their mistakes is oblivious to them. Unless someone uses their car horn to alert them to the trouble they cause, they may not even notice.

In this way human learning is similar to all organic learning at its core. So how does it differ?

As we develop, like all creatures, we develop an 'affective imprint' for stimuli - a complex pattern of tiny affective responses to stimuli: a reaction to the colour 'red', to the word 'cat', to the sound of the sea. Our limited repertoire of emotional reactions as an infant becomes a sophisticated and idiosyncratic pattern-detection mechanism as an adult.

I also believe that human affective coding is far more complex that that of other creatures: we take much longer to develop and, I believe, make more subtle discriminations of the experiences around us.

Moreover, we seem to be unique in the degree to which we are able to extend these affective comparisons - in part due to our ability to anthropomorphise.

Only a human can understand 'I wandered lonely as a cloud', since only a human can project the affective context of being a single cloud alone in the sky. This ability, it seems, allows us to greatly extend our cognitive repertoire. It may be that an evolutionary capacity aimed at permitting us to guess what another person might intend, allows us to greatly extend our cognitive abilities, by enriching the affective classification of the world around us: we now see the 'weeping' willow, the 'steadfast' rock. We think it a childish quirk that children automatically consider how a tree might feel, or the sun might intend. 

It is perhaps worth reminding ourselves of some of the more significant implications of this approach:

• AI: since artificial cognition (computation) is not constructed with affective response at its core, there is no possibility that it can ever resemble human cognition in any but superficial ways. This is not to say that artificial cognition is impossible - indeed it may well be superior to human cognition - just that it remains of a fundamentally different nature. Where artificial cognition performs functions similar to human cognition it does so in a profoundly different fashion. To put it simply: since computers do not feel, they cannot think (like us). Neural nets may learn through feedback and an artificial system of positive and negative outcomes, however without mapping our 'affective start state' there is no chance that that their cognitive development will take a similar course.

• Psychology: our understanding of human psychology has been warped by a Cartesian tradition and decades of research aimed at confirming an inaccurate view of human beings. We have turned a blind eye to the vast majority of phenomena that characterise everyday cognition and only recently begun to catalogue the numerous biases that pervade our thinking. We can only hope that we will eventually reach a tipping point, where we begin to see humans not as 'rational creature with (affective) biases' but as 'affective creatures with aspirations to rationality'.

• Education: in education we have a situation much like that in alchemy or early medicine - in which we manage to get some things right, but without knowing why. Today, education comprises a concatenation of conventions of religious proportions, many of which serve to actively suppress the instinct to learn, or which 'hijack' the natural mechanisms - for example using punishments, tests or rote learning to imprint information.

To begin to talk about education we would first need a language, for example words to describe the mapping of the learner's affective geography (a working model of their concerns), a way to describe the teacher's affective impact and role, a systematic method for describing the affective paths that lie between learner and environment. In order to develop this language we would need to start with the idea.

And here is the vicious circle of progress: how to develop a language without the idea, how to develop the idea without the proper words to describe it.

For this reason we misappropriate words: bending and warping them so that they point to something new, something as yet undescribed.

The greatest peril faced by an understanding of learning is therefore that our grasp of 'affective context' collapses back onto our simple everyday and undeveloped understanding of 'emotion' where we think of this a subset (rather than a substrate) of our cognition. We imagine that 'happy', 'sad' etc. have some impact on our learning, but only if secondary significance. We do not grasp the central idea that all cognition is underpinned by subtle affective response - just as all vision is underpinned by the detection of patches of light.

To use that analogy, if we were to explain to Martian the function of human sight as 'the detection of patches of light' they might as completely underestimate vision as sometimes happens when we describe cognition as our affective responses to our environment.

Mostly, when I try to describe learning I am pointing. And mostly, people look at the end of my finger.