Sapience

George Mobus

The University of Washington Tacoma,
Institute of Technology

Part 1. An Introduction to Sapience
Part 2. The Relationships Between Sapience, Cleverness, and Affect
Part 3. The Components of Sapience
Part 4. The Neuroscience of Sapience
Part 5. The Evolution of Sapience

 

Part 2. The Relationship Between Sapience, Intelligence, Creativity, and Affect

Decisions, Decisions

The world around us is both dynamic —moving— and forever changing in its characteristics and their statistical properties, or what we call a non-stationary environment. All autonomous agents living in such a highly dynamic and non-stationary environment must continuously make decisions about what to do next given a particular situation. Simple animals living in less complex environments have fewer decisions to make. Humans appear to live in the most complex, dynamic, and non-stationary environments imaginable; this despite their every effort to construct a predictable, convenient environment with technology! The evolution of intelligence and creativity is the development of neural computation systems increasingly able to handle more complex environments requiring more elaborate decision making processes (Geary, 2005).

For humans, only some decisions are made consciously after some form of mental analysis of the situation. Some decisions are the result of intentional thinking in this way. But the vast majority of decisions are made subconsciously or preconsciously (before conscious awareness). This often comes as a shock to people who are unfamiliar with the research in the psychological and neurological bases of decision making.

Additionally there is the scale of the situation with respect to both space and time. Most decisions are trivial, here-and-now-what-do-I-do types. Many are routine. We go through our daily lives hardly thinking about what to do next because we have habits that serve well under ordinary circumstances. It is when the circumstances are not ordinary (or common) that we need to engage conscious thinking to come up with a choice of actions.

Then there are the decisions that our limbic brain makes for us and before we are even aware that a decision is needed. The limbic brain* is the ancient portions of the brain (mid brain) that handle early sensory perception and motor signal relays to the body. It also is involved in automatic responses to semi-complex stimuli that have semantic value to the well being of the agent. The conscious brain experiences these responses through various forms of emotion but only after the limbic brain, the subconscious, has jumped into action.

In what follows I want to identify each of these functions' role in decision making in general. As I have written previously, all of the functions of the brain are integrated to produce the final behavior of the agent. But it helps to understand what is going on in the brain to tease apart the core functions to see their primary responsibilities in the overall scheme of decision making.

The Basic Relationships: A Functional Model

For much of the history of psychology and the study of intelligent behavior the focus has been on cognitive processing, and in particular, rational thinking. The field of artificial intelligence, in computer science, mirrored this focus in its attempt to replicate the human ability to play games like chess. For quite a while the basic belief was that intelligence was best seen in the capacity to win such games. Today both psychologists and computer scientists have gained a much deeper understanding of the realities involved in making decisions.

Formally, a decision process is a temporally sequential set of stages in getting from a starting state to a goal state. At each stage the decision maker is faced with a set of options, moves in game language. Each option is tagged with some kind of ‘objective’ value that should help the decision maker select the best option. This is often complicated by the fact that the valuation is based only on local information that may lead to a less than optimal global outcome or cause a failure to reach the goal state when many more stages need to be traversed before reaching that state. We can represent a decision network graphically as a tree structure where the start state is the root and each stage is represented by some number of option nodes (Fig. 1).

 

Figure 1. A decision process can be represented formally by a tree structure.

 

Many researchers hold that this formal model is an idealization of what takes place in the human brain. Though an idealization, the model of a network of decision nodes may not be far off the mark in real brains. My own work trying to build a primitive brain with simulated neurons that learn options and the weightings associated with their links has convinced me that the model is very valuable in understanding how decisions are made by humans and animals.

But the formalization is just a starting point in understanding what is happening. The intelligence comes from the brain's ability to encode situations in the environment into meaningful concepts. In effect, the concepts form the option nodes in this tree structure (note that the network is artificially represented as a tree because many nodes may be replicated both within a stage and between stages so as to eliminate cross linkages that would complicate the analysis). The concepts need to be learned by experience and so do the links that lead from one concept node (state of the world) to another under the considerations that a given action is taken by the decision maker. When a decision is taken (in the sense of a path through the tree), it generates an action that in some sense will change the situation in the world, i.e., the new state is the concept that obtains from the selection and action.

There are several unanswered difficulties with using this model. One has to do with the granularity or precision of decision-actions. The world appears to be a continuum rather than a set of discrete states. If it were encoded using discrete representations (i.e. neurons) it would be unreasonable to expect that the precision of encoding would be so great as to consider every little slight change in the world as a completely different state. A possible clue as to how to solve this problem comes from the visual and auditory perceptual systems in which it appears that there actually is a sampling rate associated with capturing frames of visual and auditory information. In other words, our perceptual systems discretize the world for us. Given how most neurons operate — communicating with discrete pulses called action potentials — this actually makes sense. But the topic is beyond the scope of this paper. For our purposes we will assume that there is some form of ‘just noticeable difference’ function operating in the nervous system that discretizes the world into small enough chunks that we can approximate continuous dynamics without great error and yet not so small that the computational load is so high we could never keep up with it.

There is some evidence that the brain operates not on single discretized representations but on small populations of semi-independent representations that collectively provide the ‘illusion’ of continuousness by statistical approximations. Again this is beyond the scope of this work, but it should be clear from this that the decision stage model above is viable for understanding intelligence.

That is, it can be if we can explain how the link evaluations are instantiated in the first place. The decision processor is faced with selecting one option out of a set of options at each stage. Each possible option carries a value. But what happens when several values are the same and there is no clear higher valued single choice? If indeed all other factors were equal this is where the creativity generator comes into play (see Fig. 2 below). In the simplest possible version, this would be expressed as a random selection. Later I will discuss a somewhat more sophisticated approach that is not random per se, but does involve novelty.

Most often, however, all other factors will not be equal. It turns out that the evaluation value attached to each node in the tree is not a simple scalar but a complex function involving percepts, explicit concepts, affective inputs, and the subtle effects of judgment or intuitions. Figure 2 shows a functional map of these various components. This map reflects to some degree the psychological construct model from part 1. This model makes explicit how the various constructs overlap, i.e. how they interact with one another.

 

Figure 2. The basic relationships between intelligence, creativity, affect, and sapience are shown in a functional diagram. Intelligence, in the sense of analytical decision making is orchestrated by a central decision processor (see below) that constructs a network of options for action based on the current perceived situation (percepts) and explicit concept knowledge previously learned. These are limited in capacity (see text). Thus the decision will always be made in uncertain and ambiguous conditions. Evolution equipped our ancestors with affective responses triggered directly from perceptions of situations that were inherently rewarding or damaging. This is a very fast reactive system, but is not sufficient to deal with more complex environments. The explicit concept store (the majority of the paleo and neo-cortices) evolved to provide more information to the decision processor. The higher levels of the perceptual system and the concept system interact directly (below). The tacit/intuitive knowledge (implicit) system evolved in higher mammals and made its greatest advances in humans. This system supplies tacit knowledge, which is learned from life experiences, to the decision processor in the form of judgment. Creativity is a kind of co-processor that generates novel links within the decision network.

 

Shortly I will tease out more details as to how the various processors and stores shown in Fig. 2 work. Here I want to describe the interrelations more holistically.

As already noted the decision processor's job is to construct and traverse the decision tree (network) seeking the best sequence of actions that will lead to a desired goal state for the decision maker. It is, for the most part, a pretty mechanical operation, coming as close to what we normally think of as a computer as you will find in neural tissue. Unfortunately, for many Star Trek fans and as I will explain below, it is not an algorithmic machine as is a computer. The choices it makes at each node are guided by many factors. There is no truly objective function that is guaranteed to produce a correct result. The decision processor takes in whatever facts of the matter it can from the perceptual system (telling it the state of the world and self) and from the conceptual system, the store of explicit knowledge that has been acquired through learning. Due to limitations of size and speed, these systems cannot produce absolutely veridical information. They provide, at best, approximations of how the world is and what concepts bear on that state of the world. Thus uncertainty and ambiguity tend to predominate in almost all decisions. The perceptual system may have got it wrong somewhere. The conceptual system may not have a good representation of a concept that pertains. All things considered, the decision will not be straightforward.

That is where the affective and the tacit/intuitive knowledge systems come into play. It has been known for a while that humans make decisions more frequently based on something called ‘judgment’. Psychologists have been working hard to tease out the various forms this takes and how it works. Antonio Damasio (1994), a neurologist researcher, discovered the importance of the role of affective influence on our decision process as an addendum to the conceptual/rational reasoning process. He discovered that decision making without affective input was nearly impossible. Patients with a brain defect that disconnected their limbic systems from the frontal lobe executive functions had remarkable difficulty arriving at decisions on even very simple problems (like when to schedule the next appointment). These patients got stuck in analysis paralysis, only able to evaluate the values attached to conceptual nodes without the benefit of any affective weightings (e.g. good/bad valences associated with pathways through the network). This brought home the difficulty that a character like Spock (Star Trek) would actually have being a purely rational being (the back story on Spock was that he was half human and so, from time to time, slipped back into the more ‘primitive’ thinking abhorred by his Vulcan contemporaries! But it was a great plot device to show how humans really do need their emotions and drives, as evidenced by Captain Kirk.)

Affective inputs to the process are important. But they can all too often be wrong. They come from a system that evolved to cause reactions to environmental conditions that could be rewarding (presence of food or mates) or punishing (predator nearby). For reptiles and earlier genera this was really about all that was needed to survive and thrive. We mammals inherited it because it is still often useful, especially when we were totally dependent on survival in the wild. Nevertheless, in matters of complex social nature, simple emotional reactiveness is not necessarily a good guide to appropriate behavior. For more complex and subtle situations the decision process needs some kind of knowledge that is global in scope, generalized and broadly applicable, and can be learned from experiences in life. This knowledge is implicit. It is held, manipulated, and retrieved subconsciously. It enters the conscious awareness only as it affects our decisions, and then only for those decisions that we are being conscious of making. Sapience is the processing system that manages the gaining and using of this tacit knowledge. It provides subtle but powerful inputs to the decision process (see below). Judgments, then, come in various degrees of tacit knowledge/affect ratios. Sapience facilitates the ratio by what we might call a second order judgment on the contributions of these two systems. For example, if the tacit knowledge system does not have a strong input to provide in some situation for which the decision maker has no prior experience, it may be a good idea to let the heart decide! At other times, especially as the decision maker ages and acquires greater life experiences, the input from tacit knowledge may actually provide a different direction to the decision, in which case sapience needs to override or down-modulate the affective input. I will examine this in more detail later.

This has been a general overview of the interrelationships between the various constructs from a functional perspective. I would now like to examine each of the constructs/functions in greater detail and then I will explicate the interrelationships at the micro (i.e., the decision node) level.

Circumscribing Intelligence, Creativity, Affect, and Sapience

In the first installment I mentioned Robert Sternberg's work on the integration of intelligence, creativity, and wisdom (Sternberg, 2003). In the above I have indicated that intelligence and creativity are involved in making decisions and are modulated by both affect and judgment. In order to put a slightly finer point on the distinction between sapience and cleverness (the combination of intelligence and creativity) and to distinguish between judgment and affective modulated decision making I want to de-integrate these functions and clarify what each contributes to the final decisions and behavior of the individual.

Intelligence

Let's start with intelligence since that is the mental capacity most people think about when talking about the human mind. The Wikipedia article (linked in the last sentence) does a good job of distinguishing some of the functions of intelligence, so I won't repeat their analysis here. What I want to focus on is the information processing function of general intelligence, a facility which is part of even the multiple intelligences model of Gardner (Gardner, 1999).

Information processing involves taking in raw data, extracting correlations in both space and time and detecting patterns that have semantic content. Next those patterns and their semantics are used to generate sets of options for action. If the options can be weighed (salience or importance) directly, that is local information is sufficient to make a choice, then the machinery of intelligence can do the weighing and select the 'best' option. More often, with complex and fast changing patterns the number of options generated are great and the weights attached may not be distinguishable purely from local information. Choices are ambiguous and outcomes are uncertain.

Roughly speaking, intelligence is responsible for the more rational approach to problem solving. Rational, here, includes inductive and abductive reasoning processes, not just deduction (at which humans are actually not very good in general). Given the information at hand, the machinery of intelligence assembles the components of interest in a decision, weighs the evidence, and selects an appropriate choice (see below, Structure of Decisions...).

Other factors which are often attributed to intelligence are things like memory capacity, speed of learning and recall, accuracy and appropriateness of encoding memories. These are the characteristics that can be measured (more or less) in tests (see IQ). It might be better to restrict the concept of intelligence to the notion of rational decision processes and collect these background capacities under a general heading like 'memory management' competency. At least doing it this way has helped me identify functional aspects of intelligence. Below I dissect the nature of decision making in the mind.

Intelligence + Creativity = Cleverness

It is difficult to separate creativity from intelligence. The two work so smoothly together to produce intelligent problem solving. Intelligence is responsible for capturing information and sifting through options looking for ways to exploit that information to solve the problem. Creativity, on the other hand, is responsible for constructing novel options, for putting concepts together that previously were not related (e.g. the unicorn).

When the number of options available to choose from gets too large the process of weighing and selection would become untenable (Schwartz, 2004). It would take too much time to analyze the options relative to the time frame in which a choice needs to be made. As we will see below, the more primitive brain (the so-called limbic system) contains automatic pattern recognizers for life and death situations that work to reduce the number of choices we have to make in such situations. Also, our affective brain works to bias choices according to prior emotional experiences in similar situations. These affective mechanisms work to trim the number of choices where emotional content is concerned and so make it easier for intelligence to do its job, with fewer choices to consider.

Similarly, we will see that sapience does the same kind of job, but from the storehouse of learned tacit knowledge that one develops over time. These systems act to keep the number of choices down to a manageable number but only if prior experience can be brought to bear. In other circumstances there is no 'precedent' by which to trim choices and no guidance available to intelligence in how to proceed. In those instances the brain invokes one of several methods to drive what might appear to be a random choice. It conducts an experiment!

We have reason to believe that the brain doesn't actually make a random choice. Rather there is some evidence that brain circuits have a built-in capacity to produce deterministic chaotic signals that have random-like qualities. My robot's search pattern, which I proudly called the 'drunken sailor' search, and which formed the basis of a foraging behavior, was generated by a central pattern generator (CPG), a circuit of simulated neurons (Mobus, 1999). These circuits, in real animal nervous systems, generate oscillatory signals that can have a roughly sinusoidal pattern. In my work I was able to show that by coupling two of these oscillating circuits I was able to generate a chaotic signal that had just the right amount of random-like qualities in variations of the amplitude and frequency of the combined signal. It turned out that the robot search pattern emulates the kind of quasi-random search pattern used by many foraging animals when they are hunting for prey but have not yet found a trail to follow.

Much of problem solving is trying to find a path through a complex web of decision points. If there are no clues at each nexus as to which next path to take then a quasi-random selection might seem as good as any other. However this isn't entirely the case. It is likely that one finds one's self at a particular point due to a history of clues, even weak ones, which means one has been on a 'right' path. In that case it would be best to pick a new out path that was not too far different from the direction one is already going (see the example of a labyrinth below). So the choice should not be completely random. Some novelty needs to be injected into the path selection process so that the chances of finding a good path are actually increased, say, over a systematic (straight) path. The reason is that in nature and the brain, resources are often chaotically distributed so that a systematic search would tend to fail more often. On the other hand, a random walk search would also fail since it could just as easily produce a clumped search before getting out into the territory (see figures 5 & 6 in Mobus, 1999).

CPGs are evolutionarily very old circuits in motile animal nervous systems. I suspect that some kind of chaotic CPG (or more likely many) are at the base of creativity in the brain. This is, of course, highly speculative. But it would not surprise me if someone were to report on such a circuit (probably to be found in the basal ganglia) that modulates attentional search in the cortical tissues of the brain. Such a mechanism would go a long way to explain how seemingly novel but not strictly random thoughts are generated. More research is needed!

In any case, creativity can solve a big problem in searching for conceptual solutions to problems. By breaking ties, or simply causing a choice to be made irrespective of rational processes, creativity helps the brain keep from getting stuck in traps (local minima, c.f. simulated annealing). THe right balance between intelligence and creativity solves the perennial problem of exploitation versus exploration in non-stationary environments. The truly intelligent agent has to strike a balance between the energy efficiency gains from exploiting a known resource and the potential discovery of new and better resources that would come by foregoing exploitation and spending some time exploring. The agent needs to invest some effort in finding new resources because there may be a better source out there somewhere and because current exploitable resources may run out.

Cleverness is behind the human propensity to invent. Invention covers a wide range. It can be the creation of a new or improved tool, or it can involve a new or improved process (procedure). The motive behind invention is generally always the same; how can one do this job better, faster, more efficiently? And this motivation comes from somewhere in the affect system.

Affect

Antonio Damasio writes about a patient who had lost the connection between his limbic (emotions) brain and the frontal cortex where decisions are made (Damasio, 1994). One of the most dramatic effects on this patient was the loss of an ability to decide! His rational brain was intact, but had lost contact with his emotional brain. One would think that this would produce a Spock-like (Star Trek) super rational being. After all, with no more emotions clouding his reasoning he should have been able to make better decisions. In fact he had trouble making decision at all. Damasio has concluded from numerous such cases that emotions, or at least some kind of emotion-based valence (positive or negative) attached to aspects of decisions (the attributes of the world resulting from making a specific choice) act to reduce the size of the decision space. The decision processor can prune out all negatively marked choices and focus on only the positively marked ones.

The marking comes from prior similar experiences that evoked low-level emotions or feelings at the time and is generally subconscious. This means that decision nodes that were involved in obtaining a positive or negative outcome were marked with that valence by the experience and that mark is used each time that particular decision node is used in future decision making. He speculates that virtually all life experiences are encoded with an emotional response tag that provides this valence factor. Then in current experiences when some choices need to be made, the valence tags can be used by the underlying processor to follow those choices that have had positive outcomes associated with them in the past.

My robot, MAVRIC (Mobus & Fisher, 1994), learned in this manner. It learned to associate a signal representing pain with objects in its environment that it should avoid. It learned to associate a signal representing reward with other objects. These somatic markers then compelled MAVRIC to avoid the potentially painful objects and seek out the rewarding objects.

Thus, ironically, good decisions really do depend on our emotions and are not the result of pure reason. This is something we inherited from our animal ancestors. Reptiles, for example, have basically limbic brains with a thin veneer of cerebral cortex to handle very simple learning functions. Early mammals had little better facilities. Most of the limbic pattern recognition, taking place in the amygdala, is largely based on genetically-controlled behaviors that proved useful in evolutionary terms. For most of animal evolution these limbic-based decisions (to approach positively marked or avoid negatively marked stimuli) have served well. As long as the eco-niche was relatively simple and non-changing over the course of many generations animals could rely on their limbic system to guide their decisions. It is when the environment changes and drives speciation toward higher use of learned patterns to modulate decisions that we see the cerebral cortices, and especially the frontal lobes, increase in relative mass and importance.

In humans this evolution has led to the preeminent place of cleverness and learned knowledge. It has also led to an expanded role for judgment in guiding decisions. Sapience includes the capacity for down modulating, if not directly overriding, limbic signals. But it also includes the monitoring of limbic subsystems in order to provide an affective assessment to the current decision point.

Sapience

Sapience is the fourth and newest tool for living agents to use in making decisions. Put simply, once the environments of evolving animal life became sufficiently complex, ambiguous, and uncertain, cleverness and affect were not sufficiently reliable in guiding behavioral decisions. Something more was needed, something that could work based on acquired experiences to adapt behavior to the more complex worlds. That something was a reliance on learned tacit models of the world against which to judge current situations and bias decisions based on, essentially, what had worked in the past in similar situations. This goes far beyond simple conditioning, as in MAVRIC. It is highly malleable and can coopt creativity to consider alternative models. Sapience is even able to coopt affect to give color and motivation to decisions. This is the basis for wisdom, for an elder who has many life experiences being able to bring those to bear on current situations. She does so not as one recalling episodes, but as one who intuitively knows the right things to do and the wrong things to avoid, no matter how complicated things seem.

The facilities of sapience are not just based on more or higher intelligence. Sapience is a self-management function that evolved out of the learning and recall control structures in the prefrontal cortex (see Fig. 2). It is, in fact, the strategic manager in a hierarchical cybernetic system. More than just a learning controller, it is a planning system as well. This will be discussed further below.

Now that we have dissected the individual components of the mind, the psychological constructs, it is time to look much deeper into how they all work together in the process of intelligent and wise decision making.

Making Decisions: Putting The Constructs Back Together

Picture yourself in a vast underground labyrinth composed of tunnels and chambers. The chambers are large enough so that, on average, ten tunnels lead in/out, including the one you just came through to get there. Some may have quite a few more, some fewer. The number doesn't really matter for this discussion, I just wanted to give a sense of definiteness to the 'allegory'.

Each chamber is well lit so you can make out any local features. Different chambers have different features, like wall color, or pictures, tables, etc. so that you can identify a chamber if you've been there before and you have a memory. The tunnels are long and possibly twisty so that you can't see what the next chamber has in it by looking down the tunnel. Chambers may contain food, water, a shower, or possibly an ogre that bites. Your life is wandering from chamber to chamber looking for resources and avoiding the ogres.

Here is a view from above some portion of a labyrinth.

 

Figure 3. A labyrinth is a model of decision processes. While inside a chamber, the searcher must decide which tunnel to take. How should the decision be made?

 

Note that tunnels can penetrate in three dimensions so that one tunnel can cross the path of another without intersecting. Thus one tunnel connects only two chambers.

Now picture yourself in a chamber. You got there by traversing a tunnel from another chamber. Currently there is a small amount of food in a plate on a table, which you gobble up because you are hungry. Now, you must decide which tunnel to go into to get to the next chamber. Say there are five tunnels to choose from. Which one should you choose? Just to make things interesting suppose you had never been in this particular chamber before (that you know of). You are still hungry. It would be nice to find something to drink. So what is your decision?

You might be able to eliminate the tunnel you came in through since you were just in that chamber and left when there was no resource left. So you might be thinking you don't need to go back. But wait. This world is dynamic and non-stationary so it is possible that a resource (mysteriously) has appeared. Of course there is some likelihood that an ogre has entered that chamber too. So your decision looks like a random choice. Just pick one of the tunnels and take it to the next chamber.

Another possibility is that while there seems to be a lot of randomness associated with how the chambers are arranged and maintained, there might actually be some kind of organization involved. Indeed there might be some causal relations between chambers that, if you could learn them, could be exploited as cues. For example suppose you discovered that if the chamber had a table with food, it almost always sat near a tunnel that led to a chamber with something to drink in it! Or perhaps the carpet next to a tunnel has scratch marks left by a transiting ogre. That tunnel might lead to a chamber where the ogre is in waiting. Thus there are possibilities for learning cues that might serve well to guide your travels. Your choices will be made based on a combination of knowledge and emotions (fear of an ogre) and drives (hunger).

This world is a model of the fundamental nature of decision making. We can characterize decision making in problem solving as a sequence of multiple choices. Making a choice at one juncture takes you to a new state. A choice leads to an action that changes your relation to the world, and hence, the world itself. Now you and your world are in a new state and find that you need to make another choice. Amazing as it may sound, all information processing boils down to such a sequence of choices.

Now let's add a bit more motive to the chamber world. Suppose you know that there exists a chamber somewhere in this labyrinth that allows you to exit to the surface where your problems will be over. You have no real information provided in your current chamber so you don't have, say, a sign that says, "This way to the exit". Your task is to solve the problem of getting to the exit and it will involve using your memory and discovering cues (patterns that involve causal relations) that do point in the right direction. Or at least they point in a "better" direction.

The problem as posed is one of searching through a world of options, learning causal relations that can assist you in making increasingly better choices in the future, and eventually (it is hoped) finding a solution to the search problem. The payoff is not only solving this one instance, but using the accumulated knowledge of causal relations to generate general solutions to all similar circumstances; say you are thrust into another labyrinth in the future (Mobus, 1994)!

Structure of Decisions and How Cleverness, Affect and Sapience Contribute

I'll get back to the labyrinth in a bit and try to introduce some more elements that make it more realistic. Meanwhile I want to delve more deeply into what it means to take a decision.

Figure 2, below, is another view of decision making called a decision tree. The node labeled 'current decision node' represents the current state of the world and all of what you can observe of the world around you. Below the current node are the set of choices that you can make and the resulting states of the world that obtain from making one of those choices. The down pointing arrow at the left indicates that these choices are in route toward a goal state — the solution to the problem that got you started choosing in the first place. Say for example that you are a carnivorous hunter looking for prey. And you are hungry! Your goal is to find food and your mode of operation is to hunt for said food. Thus your decisions are based on finding choices that lead to ever closer to your goal state. How should you choose such that the future state of the world, achieved by actions taken after making a choice, gets you closer to your goal? That is the fundamental problem in decision making.

 

Figure 4. The structure of decisions. At each node in a decision tree there is a set of possible future states of the world that include some state that will bring the decision maker closer to a goal state. Each decision is associated with an action output (not shown) that changes the state of the world. So the problem is to choose the next node such as to take an action that advances the agent toward that goal. Choosing the best option depends on having information at each node that will give a strong indication as to which option to take. In this figure, no information is indicated. In such a case a random choice would have to be made.

 

There is a whole science of decision making that is devoted to formal methods for solving the immediate problem of choosing. But unfortunately humans do not do very well in terms of thinking formally to make choices (c.f. the delightful Marcus, 2008). No animal does, and we are, after all, animals. Still we must make choices and carry on with our business.

Animals have evolved very clever mechanisms for carrying out heuristic decision making (Gilovich, et. al, 2002). As with Damasio's recognition, above, that we mark (or tag) our experiences with valences so that we can use those experiences in the future to help guide our choices in similar circumstances, there are a number of pattern recognition 'tricks' that can be used to tag memories of states of the world such that we can use those to provide guidance as well.

Remember the problem with trying to achieve a global optimal outcome based on just local information? This issue is related to a well-known decision process called local optimization? As I pointed out, taking a decision based on a local optimum can lead to a global sub-optimization. The same kind of problem exists with respect to decision points. It may be the case that a prior experience (and its set of attributes and tags) has been coded by the limbic system with a negative valence. So our local information suggests that we should avoid that choice. But it is also possible that what was a temporary negative experience led to a later positive experience of much greater value. This is the classic 'face the danger for the greater reward' problem (it is also related to the reward postponement problem). A beast operating strictly on limbic signals will avoid the local negative situation. But a beast with a memory of pathways through the danger will be able to choose the dangerous selection on the off chance of reaping a bigger reward. Choosing the path with the highest immediate reward is called the ‘Greedy Method’ and there are instances in computer applications where local information is all one needs to make the right choice. But far more often local information is not enough (as in the above example). There needs to be more global information available at the decision point in order to make the right choice with respect to reaching the goal state.

To begin to understand this more elaborate mechanism of decision making in intelligent animals take a look at Fig. 5.

 

Figure 5. The context surrounding a decision node in a creature with extensive memory. The state of the decision process is shown where a prior decision had been made leading to the current node. This node is surrounded by a ‘knowledge milieu’ that will help interpret the various aspects of the state of the world to guide in the choice of one of the possible nodes (green). In this figure, the inputs from the affect systems, the valence ‘tags’ or somatic markers (per Damasio). Without any other information operating on the decision, the choice would be driven by the feeling that B would be the best one would be made. With a sufficiently rich knowledge milieu, especially from the base of tacit knowledge, the affective valences might be overridden.

 

There is a great deal more going on in this figure (and in the brain) than most people might imagine. Remember these apply to every minuscule decision that the brain makes, especially unconsciously. The 'knowledge milieu' in the figure represents a host of background knowledge that can be brought to bear on every choice (decision point, see Fig. 6 below). This includes the tacit knowledge I've discussed previously, as well as the affective motivations (drives toward a goal state), and facts of the situation, meaning state of the world at that instant. The latter aspect is very confusing to most people who think we apprehend the world through our sensory perceptions moment to moment. The truth is we are conscious of the world through our memory systems. Our prior concepts are more responsible for our present perceptions than we realize. But, unfortunately, that is another whole story that will have to be dealt with in the future. The so-called facts of the situation are really our subjective experience of facts and not objectively determined facts as we ordinarily think of them.

 

Figure 6. Knowledge encoded in various areas of the brain can influence the decision through the knowledge milieu. At the very least, the knowledge milieu will modulate or override the affective valence effectors so that a decision may be made not on the basis of how good it feels, but on how beneficial it may be later on.

 

The figure includes the affective valences for choices regarding predicted states of the world. These states are encoded in memories based on experiences, in the past, of similar situations. Or they are conjured from roughly similar experiences. The brain, remember is a magnificent modeling machine that can project future states even when no direct historical experience was had. It does this by using creativity through analogical thinking. It finds past experiences that were sufficiently similar and uses those to estimate the likelihood of future states.

So the basic decision problem is to take all current information and background and affective knowledge into account while estimating the supposed best choice among the options presented. Now just to make things even a little more complicated, we add the role of creativity to the mix. This comes into the picture in three possible ways (at least). Suppose of all the choices presented two or more seem to have roughly the same affective and knowledge-based value. What to do? One obvious approach would be to choose one of the options at random (Mobus, 1999). Little is known about how the brain resolves such choices. But there is a rich literature on creativity suggesting that indeed the brain 'makes guesses' in some sort of quasi-random fashion (Andreasen, 2005; Csikszentmihalyi, 1997).

Even more interesting than choosing one of several equally attractive options is choosing an option that is not attractive. This could be done essentially in the same way as choosing by quasi-random selection. The choice may be made based on a need or drive to explore (more below). The third possibility is to add an option or two to the set of options that are not really part of the original (intelligent) construction. In other words, the creative brain adds a seemingly unrelated node to the mix of options on an off chance that it will lead to something creative and, hopefully, constructive. This may, again, be in response to a need or drive to explore; what we would call curiosity. However creative choices are made we do know that they can sometimes be completely irrational or, as we call it, outside the box. These mechanisms for creative choice might also go awry in mental diseases causing people to become permanently irrational (e.g., schizophrenic or psychotic).

The thing about creative choosing is that we only positively acknowledge it as creative if it works! Otherwise we write it off as a foolish mistake. Still there is a fundamental need for occasionally trying something not fully indicated by the information at hand. There seems to be a balance in dynamic autonomous systems between pure exploitation of a situation, choosing the best option, and pure exploration, choosing a non-best option to see what happens. Animals show a range of where this balance lies but in general they trade off between the two extremes. Exploitation is not betting. It is choosing a certain outcome. But the problem is that there is no such thing, in nature, as a sure thing. The world is forever changing, even if just a little, what time-series statisticians call a non-stationary process (see Stationary process for a basic description). That being the case pure exploitation is guaranteed to fail at some point as a life strategy. Every species thus incorporates variation in form and behavior of its members, always exploring, at the edges, new possibilities. This is especially important for a species that is dealing with a highly non-stationary eco niche. The more complex that niche is, the more non-stationary it is. Hence the species needs more exploration of what Stuart Kauffman calls the 'adjacent possible' (Kauffman, 1996). Darwinian evolution is such a long time scale exploration of genotypes, while over the short term species exploit their phenotypes. Animals that rely heavily on experience-driven adaptation (learning to behave!) such as humans require the tradeoff between exploitation and exploration in their individual lives.

Sometimes exploration will fail, or get the individual into trouble. Teenage and even younger boys, for example, seem to be very creative in the ways they get into trouble. Not all creativity leads to opportunities for new forms of exploitation. But when it does we gain new possibilities. And since some old situations that we have exploited in the past may disappear (non-stationarity) its good to have new possibilities at hand.

Of course when people we recognize as creative are exploring their domains they are not randomly trying this or that. They are applying judgment to their created options, judgment based on the tacit knowledge they have accumulated over their lives (Andreasen, 2005; Csikszentmihalyi, 1997; Sternberg, 2003). Truly outstanding individuals are often called geniuses.

And, at last, we come to the role of sapience in this process. In humans, tacit knowledge seems to have the greatest effect on decision making. Strong emotions can rule under the right circumstances, of course. Rational decisions can be made for very small problems (or through the exercise of external formal methods). But the vast majority of human decision making involves social, complex, and long-term problems with significant levels of ambiguity and uncertainty. Our models of the way the world works need to include moral sentiments (concepts of what is right and wrong behaviors), systemic knowledge (how things are interrelated and what effects will derive from what actions), and some number of scenarios for what consequences might be expected in the long run from different decisions.

There is reason to believe that creativity is always generating many more possibilities than we could ever hope to explore, at a subconscious level. One of the jobs that sapience does is help filter these possibilities at an early construction stage so that the processing load on intelligence in making a decision is reduced. Thus judgment plays a role in keeping out the most deviant forms of creative ideas so that what does reach the decision process is at least feasible. In all likelihood, much of the filtering is actually done within the tacit knowledge modeling process itself. If the model breaks down on incompatible situations then it probably dies. Many potential ideas may simply be too weak in activation to make it closer to the decision process. I suspect that the real role of sapience is to modulate the creation and filtering processes as well as provide some kind of final say in what gets through. This would help explain several anomalies we see with creativity in individuals. Some people are very creative in the sense of having interesting and different ideas that make it into the public sphere (through their consciousness). Their sapience may be promoting the generation of these ideas if the judgment is that the creativity processor generally does a good job (e.g. society rewards their creativity). Other people are dull and rarely have a creative thought. Their sapience may be underdeveloped when it comes to such a promotion function. Such people's creation generation might be weak or their pre-filtering may be too strong. Finally, there are a few people (fortunately) who are over possessed with wild creativity that reaches the public sphere that shouldn't. Some forms of extreme sports risk taking may be subject to this effect.

Sapience involves judgment of what to learn, what to attend to in life, how to organize it for most effective future use, how to access it when needed. It involves shaping current decisions in such a way that a good outcome is increased in likelihood; an outcome that is best for the greatest number and for the longest time. Such judgments are applied unconsciously, intuitively, even if they later come to conscious awareness after the fact. Sapience rounds out the toolkit of decision making methods that are feasibly implemented in neural tissues, as far as we know. Animal life started with simple stimulus-response behavior with some built-in adaptiveness (Mobus, 1999a). It developed nervous systems to provide the coordination level of hierarchical control as a response to the increasing exploitation of eco-niches by new species and more complex competition. The first versions evolved minimally modifiable reactive programs, automatic pattern recognition, instinctive response repertoires. These served well for most of evolutionary history. But evolution toward increasing complexity continued to favor larger and more complex brains supporting more varied and modifiable responses. Affective response systems sufficed until evolution happened on the neural networks able to represent models of the more complex environments (cortical sheets with macro-cellular cortical columns). Learning and adaptive behavior took off. Intelligence and creativity were enabled at a new level of sophistication. They had obvious selective advantages and so generated species more quickly able to adapt to changing environments; essentially the birds and mammals.

These earliest 'learners' and 'thinkers' were also relying on a primitive kind of judgment just as they relied on affective input from the lower brain as described above. Simple judgment, as I have mentioned, is a guide to decision making in relatively simple situations but which are more complex or uncertain than rational decision making can handle. In social primates we see an expansion of judgment and integration with systems thinking and moral sentiments such that the earliest glimpses of what would develop into full-blown sapience, apart from raw intelligence. Finally, in the genus Homo we see the full basic model of sapient thinking applied to decision making. The domain of decisions humans have to operate in is vast compared to any other living primate. Language facilitates but also expands these domains. The capacity to use judgments to guide decisions has reached a significant level in Homo sapiens. But as we are beginning to learn, it is a level only able to deal with the measure of complexity and uncertainty experienced by pre-agricultural man. Moreover, it is far from reliable in the average individual. We all suffer lapses and biases that are genetically mediated. Our native capacity for judgment is limited to problem spaces much smaller than we encounter in the world we have created with our cleverness.

Figure 7 summarizes the relationships between intelligence, creativity, affect, and sapience. The latter two are directed at guiding decision processes in intelligence with the help of creativity to force exploration of new possibilities or simply keep the brain from getting stuck. The arrows represent major communications pathways and the direction of influence. Note that sapience (as processed mainly in the prefrontal cortex) has influence over the other three areas as well as monitors them. The biggest influence is over decision processing in the intelligence function. In Part 3 I will dissect the functions of sapience in greater detail.

 

Figure 7. Functional relationships between components of cleverness (intelligence and creativity), affect (emotions), and sapience. Affect and sapience provide biases and search "control" to the intelligent finding of solutions to problems. Creativity and intelligence interact to provide novelty to searches as needed.

 

Back to the Labyrinth

Now recall the labyrinth. You are in a chamber and need to decide which tunnel to take in your quest for an exit. Along the way you are also motivated by two basic things, the need for food and water to maintain your strength and the need to avoid ogres that have sharp teeth and consider you food.

I allowed that the chambers could be richly decorated. It turns out that the interior decorator had various but consistent themes that she used in various areas of the labyrinth. In other words there are actual patterns of decoration within chambers such that there is a conceptual connection among chambers. Moreover, the decorator left little clues at the entrances of some of the tunnels to suggest the theme that will be found in the next tunnel.

Your job is to learn these patterns and cues such that you can make a reasonable prediction of what will be found in the next chamber upon following a particular tunnel. If you go through chambers repeatedly, but each time take a different 'out' path you start to build up a model of the labyrinth including a pretty good ability to guess what is in a chamber that you had not been to before. You might also note associations of things like the kind or amount of food and drink with certain themes. You might even discover that ogres prefer to hang out in certain thematic regions, which would allow you to use the cues to avoid those. As you build up a knowledge base of these cues and themes you begin to get better at finding your way through the complexity. You may find an association between a sequence of themes that lead to the exit!

As you first wind your way through the labyrinth you will need to focus attention on details, and reason explicitly about the choices you have to make. But over time, the patterns that you begin to encode in your memory start to provide a scaffolding for automatically adding to the knowledge with each new experience. Also, the decisions you make seem to come more easily and automatically, without reasoned thought. You are building a storehouse of tacit knowledge about the labyrinth that begins to guide your decision processing. You are using judgment more than logic to constrain your choices and generally (though by no means guaranteed) make better choices as time goes on.

Life is a very complex labyrinth for humans. It not only involves making choices for one's self, but making choices on behalf of your family, friends, and tribe. The number and kinds of choices are immense, even for more primitive people living in more natural settings. Imagine what it is for cosmopolitan humans surrounded by strangers and dealing with all of the rules, mechanisms, and bustle of modern life!

Conclusion

Life is dynamic and chaotic. Change is always in front of us. To be alive is to be making decisions all the time. Formal methods of decision making (like computer programming) have taught us that problems of reasonable size and complexity can be solved once all the necessary information has been gathered. But once you introduce huge scale, uncertainty, high risk, time constraints and other factors of real life, the ability to solve these problems with such formal methods dissolves. Instead, the human brain uses a variety of non-formal methods to provide approximate or satisfactory solutions under a wide variety of situations.

One method is the quasi-random or chaotic choosing of creativity. Another is the evolutionarily, tried and true, method of affective response. But for higher mammals and especially man, the method that has been instrumental in allowing us to adapt to extremely complex environments is knowledge-based judgment. In mankind we see the first glimmers of judgment based on moral sentiment, systemic and strategic thinking in social contexts. Indeed, I will be arguing that it is the latter element of strategic thinking that turns mere judgment into sapience. In man, sapience emerges in a nascent state. But as I shall also argue, that state is not yet able to handle decision making in the world that cleverness has wrought.

If, and how we might manage to transcend this latter point will need much thought.

References