Friday, 28 February 2014

121. Marvin Minsky’s ‘Society of Mind’

'Our minds did not evolve to serve as instruments for observing themselves, but for solving such practical problems as nutrition, defense, and reproduction' (Marvin Minsky 2006).

Marvin Minsky is a pioneer in the field of machine intelligence. Efforts at developing machine intelligence have resulted in deep insights into how the human brain functions. In 1986 Minsky published the book The Society of Mind, in which he formulated his ideas about human cognition. His next book, The Emotion Machine, published in 2006, reflected the progress made at that time in gaining insights into the workings of the human mind via the machine-intelligence approach.

Minsky’s ‘society’ of mind comprises of ‘agents’ or ‘resources', which are the simplest individuals that populate the brain. Each agent or resource can be visualized as a typical component of a computer program, like a simple subroutine or data structure. The agents can get connected and composed into larger systems called agencies or societies of agents. The agencies self-organize into still larger conglomerates that can perform still more complex functions, and so on into still higher and higher levels of self-organization and complexity, ultimately leading to the emergence of abilities we attribute to minds. There is a hierarchical structure and organization, like in any complex adaptive system.

The idea of hierarchical levels of organization was well documented in an earlier publication of Minsky (1980):

'One could say but little about "mental states" if one imagined the Mind to be a single, unitary thing. But if we envision a mind (or brain) as composed of many partially autonomous "agents" — a "Society" of smaller minds — then we can interpret "mental state" and "partial mental state" in terms of subsets of the states of the parts of the mind. To develop this idea, we will imagine first that this Mental Society works much like any human administrative organization. On the largest scale are gross "Divisions" that specialize in such areas as sensory processing, language, long-range planning, and so forth. Within each Division are multitudes of subspecialists — call them "agents" — that embody smaller elements of an individual's knowledge, skills, and methods. No single one of these little agents knows very much by itself, but each recognizes certain configurations of a few associates and responds by altering its state'.

As is the case with any complex adaptive system, we cannot predict with certainty the properties of the mind-system in terms of the laws of physics applied to the constituent agents, nor can we start from the observed complexity of the brain and work our way downwards all the way to understand why the increasing complexity took a particular route in phase space [please note that 'deterministic' and 'unpredictable' are not mutually exclusive propositions in physics]. To quote Minsky (1990):

The functions performed by the brain are the products of the work of thousands of different, specialized sub-systems, the intricate product of hundreds of millions of years of biological evolution. We cannot hope to understand such an organization by emulating the techniques of those particle physicists who search for the simplest possible unifying conceptions. Constructing a mind is simply a different kind of problem — of how to synthesize organizational systems that can support a large enough diversity of different schemes, yet enable them to work together to exploit one another's abilities’.

Here is Minsky’s (1986) take on consciousness:

In this book, the word (consciousness) is used mainly for the myth that human minds are "self aware" in the sense of perceiving what happens inside themselves. I maintain that human consciousness can never represent what is occurring at the present moment, but only a little of the recent past  --  partly because each agency has a limited capacity to represent what happened recently and partly because it takes time for agencies to communicate with one another. Consciousness is peculiarly hard to describe because each attempt to examine temporary memories distorts the very records it is trying to inspect’.

Minsky also described ‘free will’ as a myth, the myth that human volition is based upon some third alternative to either causality or chance.

The ‘Single-Self’ concept

Some people still subscribe to the concept that there is creature (or a set of creatures) inside us that does all the feeling or thinking for us, and makes all the important decisions for us. It is our ‘identity’ or ‘self’. Even our legal system distinguishes between deliberate wilful murder, and murder that was not pre-planned. This Single-Self concept may possibly be useful as a meme, but has no scientific basis.

Why do humans entertain such fiction? It may be partly because it makes life look pleasant ‘by hiding from us how much we're controlled by all sorts of conflicting, unconscious goals’. According to Minsky:

That image makes us efficient, whereas better ideas might slow us down. It would take too long for our hardworking minds to understand everything all the time. However, although the Single-Self concept has practical uses, it does not help us to understand ourselves — because it does not provide us with smaller parts we could use to build theories of what we are. When you think of yourself as a single thing, this gives you no clues about issues like these: What determines the subjects I think about? How do I choose what next to do? How can I solve this difficult problem? Instead, the Single-Self concept offers only useless answers like these: My Self selects what to think about. My Self decides what I should do next. I should try to make my Self get to work’.

He goes on to say that: ‘Whenever you think about your "Self" you are switching among a huge network of models, each of which tries to represent some particular aspects of your mindto answer some questions about yourself’.

Wednesday, 19 February 2014

120. Can We Take Decisions Without Involving Emotions?

"What sorts of ‘rules’ could possibly capture all of what we think of as intelligent behaviour however? Certainly there must be rules on all sorts of different levels. There must be many ‘just plain’ rules. There must be ‘metarules’ to modify the ‘just plain’ rules; then ‘metametarules’ to modify the metarules, and so on. The flexibility of intelligence comes from the enormous number of different rules, and levels of rules. The reason that so many rules on so many different levels must exist is that in life, a creature is faced with millions of situations of completely different types. In some situations, there are stereotyped responses which require ‘just plain’ rules. Some situations are mixtures of stereotyped situations - thus they require rules for deciding which of the 'just plain’ rules to apply. Some situations cannot be classified - thus there must exist rules for inventing new rules ... and on and on. Without doubt, Strange Loops involving rules that change themselves, directly or indirectly, are at the core of intelligence. Sometimes the complexity of our minds seems so overwhelming that one feels that there can be no solution to the problem of understanding intelligence - that it is wrong to think that rules of any sort govern a creature's behaviour, even if one takes ‘rule’ in the multilevel sense described above" (Douglas Hofstadter, Gödel, Escher, Bach).

We cannot take decisions without involving emotions. This conclusion of modern psychology goes against the grain of what was believed to be the case about the nature of rational behaviour for most of the 20th century. The conventional picture was that at the bottom of the hierarchical complexity of the human brain is the brain stem, which controls bodily functions like heartbeat, breathing, and body temperature. At the next higher level is the diencephalon, which regulates hunger pangs and sleep cycles etc. Then comes the limbic region, which generates and controls emotions (violence, lust, impulsive behaviour, etc.). These three levels of brain complexity are common to all mammals, including humans. Lastly there is the prefrontal cortex, predominantly responsible for our reasoning power and intelligence etc.  Although it enables us to suppress emotions to a small or large extent, it is wrong to think that this ‘rationality’ portion of our brain can completely overpower or overrule what the three hierarchically lower parts of the brain tend to do. In other words, it is impossible for us to make decisions which are completely dispassionate or ‘reasoned’.

It is also true that a substantial portion of the prefrontal cortex is involved in our emotional behaviour. How do we ‘manage’ our emotions? We do so by thinking about them, and the thinking is done mainly by the prefrontal cortex.

The term 'metacognition' is used for the capacity of our prefrontal cortex to contemplate about our own mind.

The frontal cortex knows when we are, say, angry. In fact, practically every emotional state comes with self-awareness attached to it. This enables us to figure out or ‘think’ why we are feeling the way we are feeling. Thus we humans are able to exercise a certain degree of control over our emotions by what is commonly called ‘rational thinking’. This is also how we make decisions. The emotional brain is constantly sending out signals about its likes and dislikes. The prefrontal cortex monitors these emotional outputs and tries to decide which signals to take seriously and which ones to overrule. Although the rational brain cannot silence emotions, it can help figure out which ones should be followed.

A highly readable account of the role of intuition and emotions in our decision-making process has been given in the 2009 book How We Decide by Jonah Lehrer.

Unlike other regions ('columns') of the cortex, which specialize in processing specific types of stimuli, the cells of the prefrontal cortex can process whatever kind of data they need to process. This enables our brain to look at a given problem from a variety of vantage points, and even come out with creative solutions.

How does the prefrontal cortex accomplish this? The answer has to do with its special kind of memory called the working memory. It is a short-term memory, but it has a persistence feature. It is a meeting ground, and also a melting pot, of information from various sources. Neurons in this part of the brain fire in response to a stimulus, and then keep on firing for several seconds after the stimulus has disappeared. This allows the brain to make creative associations. This is the so-called restructuring phase of problem-solving: Here information is mixed together in new ways and overlapping of ideas occurs, leading to new insights. The resultant novel neural wiring enables you to identify the answers you were looking for. This is an important feature of human intelligence.

The emotional brain is important too

Excessively rational thinking can backfire, because it often amounts to suppressing what the primitive brain is trying to tell us. This problem arises because the rational brain is not an infinitely powerful supercomputer, meaning that rational analysis cannot always provide the best solution to a complicated problem. The cumulative wisdom buried in the (much larger) primitive brain must also be used.

The psychologist George Miller demonstrated in his essay ‘The Magical Number Seven, Plus or Minus Two’ that the conscious brain can only handle about seven pieces of data at any one moment. The computational circuitry of the rational part of our brain is only a tiny fraction of the total capacity of the brain, ‘just a few microchips within the vast mainframe of the mind.’ As a result, too many choices, or too much data, can overwhelm the prefrontal cortex, leading to bad decisions. The trick lies in learning when to trust your intuitions more than your reasoning power. ‘Because working memory and rationality share a common cortical source  -  the prefrontal cortex  -  a mind trying to remember lots of information is less able to exert control over its impulses. The substrate of reason is so limited that a few extra digits can become an extreme handicap’ (Lehrer 2009). The fact of life is that the rational part of our brain (which is really a very recent novelty on the evolutionary time scale) has a rather slow and small, even erratic, CPU. Too much information can interfere with understanding. When the prefrontal cortex is overwhelmed, correlation is confused with causation, and people tend to make theories out of coincidences.

And yet, excessive dependence on the emotional brain can be risky too. The ideal situation is that exemplified by, say, a champion chess player. Through an unhurried analysis of the games he won or lost, he builds up experience (turning mistakes into educational events) which gets ‘internalized’ into his emotional brain. In due course, it becomes ‘second nature’ for him to make the right moves, not having to consciously analyze the consequences of too large a number of prospective moves. The emotional brain is a huge supercomputer, with massive parallel-processing capabilities. It is neither easy, nor perhaps desirable, to shut off this supercomputer, no matter how hard you try to do so through your prefrontal cortex.

Saturday, 15 February 2014

119. The Human Neocortex

The human brain, along with the spinal cord, comprises the central nervous system. The top outer portion of the brain, just under the scalp, is the neocortex (or cortex for short). It covers most of the R-brain (R for reptilian), and has a crumpled appearance, with many ridges and valleys. The R-brain is rather similar in reptiles and mammals, and has a number of parts, including the thalamus and the hippocampus.

Humans are special compared to other mammals because of their very prominent prefrontal cortex (or frontal lobe). The prefrontal cortex (particularly the upper two-thirds of it, including the dorsolateral prefrontal cortex) can be regarded as the rational centre of the brain; or the rational brain. The rest of it is the emotional brain.

The human cortex, if stretched flat, is the size of a large napkin, and ~2 mm thick. It has six layers, each roughly the thickness of a playing card. There is a branching hierarchy among the layers. Layer 6 is at the bottom of the hierarchy, and Layer1 is at the top. The inputs from the various sensory organs are received in Layer 6, and then interpreted and correlated. Then more and more abstract and generalized versions of the information are sent up the hierarchical layers. There is a very high degree of feedback and feedforward among the layers, as also cross-correlations.

There are ~1011 nerve cells or neurons in the human cortex. Most of them have a pyramidal-shaped central body or nucleus, as well as an axon, and a number of branching structures called dendrites. We can think of the axon as a signal emitter, and the dendrites as signal receivers. When a strand of an axon of one neuron (the presynaptic neuron) ‘touches’ a dendrite of another neuron (the postsynaptic neuron), a connection called a synapse is established. A typical axon is involved in several thousand synapses.

Portions of the cortex can be identified as different functional areas or regions. For example, a portion of the frontal lobe is the motor cortex. It controls movement and other actuator functions of the body.

The cortical tissue can be functionally divided into vertical units or columns. Neurons within a column respond in a similar manner to external signals with a particular attribute.

When a sensory or other pulse (‘spike’) involving a particular synapse arrives at the axon, it causes the synaptic vesicles in the presynaptic neuron to release chemicals called neurotransmitters into the gap or synaptic cleft between the axon of the first neuron and the dendrite of the second. These chemicals bind to the receptors on the dendrite, triggering a brief local depolarization of the membrane of the postsynaptic cell. This is described as a firing of the synapse by the presynaptic neuron.

If a synapse is made to fire repeatedly at high frequency, it becomes more sensitive; i.e. subsequent signals make it undergo greater voltage swings or spikes. Building up of memories amounts to formation and strengthening of synapses.

The firing of neurons follows two general rules:

(1) Neurons which fire together wire together. Connections between neurons firing together in response to the same signal get strengthened.

(2) Winner-takes-all inhibition. When several neighbouring neurons respond to the same input signal, the strongest or the ‘winner’ neuron will inhibit the neighbours from responding to the same signal in future. This makes these neighbouring neurons free to respond to other types of input signals.

The functionality of the cortex is arranged in a branching hierarchy. The primary sensory regions constitute the lowest rung of the hierarchy (Layer 6). The sensory region for, say, vision (called V1) is different from that for hearing etc. V1 feeds information to higher layers called V2, V4 and IT, and to some other regions. The higher they are in the hierarchy, the more abstract they become. V2, V4 etc. are concerned with more specialized or abstract aspects of vision. The higher echelons of the functional region responsible for vision have the visual memories of all sorts of objects. Similarly for other sensory perceptions.

In the higher echelons are areas called association areas. They receive inputs from several functional regions. For example, signals from both vision and audition reach one such association area.
Although the primary sensory mechanism for, for example, vision is not the same as for hearing, what reaches the brain at higher levels of the hierarchy is qualitatively the same. The axons carry neural signals or spikes which are partly chemical and partly electrical, but their nature is independent of whether the primary input signal was visual or auditory or tactile. Finally they are just patterns.

Creation of short-term memory in the brain amounts to a stimulation of the relevant synapses, which is enough to temporarily strengthen or sensitize them to subsequent signals.

This strengthening of the synapses becomes permanent in the case of long-term memory. This involves the activation of genes in the nuclei of postsynaptic neurons, initiating the production of proteins in them. Thus learning requires the synthesis of proteins in the brain within minutes of the training. Otherwise the memory fades away.

Information meant to become the higher-level or generalized memory, called declarative memory, passes through the hippocampus, before reaching the cortex. The hippocampus is like the principal server on a computer network. It plays a crucial role in consolidating long-term memories and emotions by integrating information coming from sensory inputs with information already stored in the brain.