In this post I tackle some problems concerning the capacities of predicting future events and performing actions based on these predictions.
What is prediction in the mind? How does it come about? What are the necessary mental faculties that allow for predicting future events? How does prediction relate to other mental phenomena such as understanding and simulation? The same can be asked about expectation and anticipation. What are the relations between these three, seemingly similar concepts?
The ideas and speculations are based on Complex Mind Theory, which most of the posts on this blog are about.
A tour of prediction, expectation and anticipation
I define the relevant concepts in a particular order which will become clearer as the post progresses.
- “Prediction is a representation of a particular future event.” (Pezzulo et al., 2008, p. 25). To be more precise, prediction is “having in mind” a thought about something that the mind thinks will happen.
- Expectation is being ready for something that the mind thinks will happen. Expectation does not imply that the mind wants for it to happen, only that the mind thinks it will happen and the mind is somehow prepared for it.
- “Anticipation is a future-oriented action, decision, or behavior based on a (implicit or explicit) prediction.” (Pezzulo et al., 2008, p. 25). When a mind predicts something and expects it to happen, it can initiate an action or an action chain that will, in a way, “catch” the predicted event. That is, the anticipatory behavior is not directed toward the events immediately present, but toward the event that is expected to happen. Anticipation may also refer to making decisions based on future, expected events, and only on events that are happening now.
What we can observe is a kind of progression of complexity: prediction is just an inference from the presently observed events, knowledge, with the conclusion being a thought regarding what is likely to happen. Expectation is prediction with something added to it: a mind is getting ready to observe a predicted event. Anticipation is expectation with something added to it: an organism prepares and executes behavior based not only on observed environment but also on thoughts about the expected event.
Systems capable and systems incapable of predicting
To get a clearer picture of the distinctions between the three concepts let’s take a look at some simple examples. A small moving bacteria happens to move onto a place when there is some chemical (that we know leads to food). The bacteria moves a bit to the left, there is less of the chemical. The bacteria returns, moves to the right, there is more of the chemical. The bacteria moves where there is more of the chemical than where it was before. This phenomenon is known as chemotaxis. At this point we can ask: does the bacteria predict what will happen if it moves to the right (there will be more of the chemical and maybe food)?, does the bacteria expect to find food at the end of the chemical trail?, are the movements of the bacteria anticipatory?
I propose to leave the bacteria for a while to ponder these questions so that we can look at something even simpler than that: tiny bits of iron in a magnetic field (see figure 1). Let’s imagine that we have a magnet on a table and we spread some iron fillings around. Let’s focus on a single piece of iron. It falls on the table, it is subjected to the magnetic field (we can say that the filling “senses” the field). We see that the filling moves, it aligns with the field in a particular way, depending on its length and position.Did the filling predict that the movement will align it with the field? Did it expect that? Were its movements anticipatory? In this case we would not say that. We could say that the iron filling “senses” the magnetic field, but in this case it is irrelevant. Let’s look closely at these questions starting from the last one: Where the movements of the iron filling anticipatory? No, the movements of the iron piece were only due to the magnetic field, the filling did not initiate the movement. Did the iron piece expect that the movement will align it with the field? Again, no. We would have to perform some non-trivial intellectual acrobatics just to imagine what would “iron filling expects” mean. Can and did the iron piece predict what will happen shortly in the future? To be able to predict something, a system would have to make inferences from the environment and/or its own knowledge. We could try hard and say that the different between the direction the filling is point at and the specific of the magnetic field constitute information that the filling could use to predict the unfolding of the situation (the difference between the direction of the filling and the specific property – a “direction” of the magnetic field – decrease). The thing is, we can predict that. The iron piece can not, as it is not able to “think” or to “represent” the situation it is in. In short, the iron filling does not act, predict, make expectations nor does it anticipate. It reacts to the environment.
The not complicate example of the iron filling in a magnetic field will hopefully help is in our problem of the bacteria in its environment with a specific gradient of a chemical. In a similar fashion to a piece of iron, our bacteria is subjected to the chemical “field” and it’s being brought closer to the place where the concentration of the chemical is the highest. In this case however the bacteria uses its energy reserves, it initiates the action of moving its tentacle-like “limbs” (the so called flagellar filaments). Bacteria, being in one position remembers the concentration of the chemical, then moves, checks the concentration there. If the concentration is greater than before, it moves forward even more. If the concentration is lesser than before, it sets a new course and moves in a new direction. The bacteria clearly compares the current concentration with what it remembered from the previous place. But does the bacteria predict the concentration of the chemical farther on the path? The thing is: it doesn’t have to. The only thing it has to do is to remember, compare, take appropriate action. The idea of bacteria predicting, expecting and anticipating higher concentration of a chemical along the path is our inference, our projection of these concept onto the bacteria based on the bacteria’s behavior and our own modes of thinking. The bacteria reacts and it is completely sufficient for it.
One more example, this time of a simple system build by man. Figure 2 depicts a feedback system: fire heats the water above and the alcohol in a glass tube. Heated alcohol expands raising a rod connected with another rod with a damper, a lid on the other end. The the first rod is raised by the expansion of the alcohol, the lid is being moved down, covering the exit of the fumes. By covering the exit of the smoke, fire is dimmed. Alcohol is getting colder, contracts, the first rod is pulled down, bringing up the lid, which makes exit for the smoke, leading to increased fire. In effect, the system revolves around a given temperature. The system acts as a simple thermostat. The questions we can ask now are: does this system or a part of it (alcohol in a glass with a rod) predict, expect and behave in an anticipatory fashion?
It would not be crazy to think that the level of expansion of the alcohol somehow “represents” the current state of the environment (the strength of the fire, the position of the lid). Elements where the alcohol and the rod meet can be thought of as action (sub)system. Decision (by how much to raise / lower the rod) is made accordingly to the strength of the fire, as a reaction to the strength of the fire. Now, can the system, as a whole or one of its parts, predict what the effect of its actions (raising / lowering of the rod and the lid) will be? We’ve just seen that we can think of the level of expansion of the alcohol as acting as a knowledge center of the system. What about an analogue for “representing” the future state or future events? Examination of each element in isolation will surely lead us to the conclusion that there is no single thing that “knows” what will happen. There are two possibilities that come to mind: a) prediction may be visible on the level of the entire system, and b) prediction may not be visible in the “state” of any part or the whole system, but may be what the system does: predicting is doing.
Let’s examine the idea of prediction being visible on the level of the entire system. We can easily infer, or predict, what will happen in a short while if the fire burns strong for some time. How could the system in question make that prediction? We can discern the strength of the fire and the level of expansion of the alcohol. We can similarly differentiate between the level of expansion of the alcohol and the position of the lid above the hole of the oven. The second difference looks much more promising. The level of expansion of the alcohol tells the system how hot is in the oven, the position of the lid tells the system how much air can circulate in the oven and consequently how strong the fire can burn. We have the present state of the environment (alcohol expansion) and the constraint the system gives on the environment (the fire) by means of its actions (position of the lid). If we simply compare the two, we see that what follows is a difference between what is and what can be, which will later lead to what will be (what will happen with the fire). For the system to be able to predict the future events / states of the environment it would have to know the difference between the two relevant factors: present state of the environment (the level of expansion of the alcohol) and the constraint it poses on the environment (the position of the lid), but also how does the system’s actions (lowering / raising of the lid) impact the environment. Having these capabilities the system would need one more thing: know all these things, predict its effect on the environment, but refrain from making any changes. Without this constraint we would be immediately pushed into ascribing predictive faculties to practically every system imaginable: bundles of atoms, galaxies, rocks, dying animals. If the system can be thought as having all the necessary ingredients of knowing (strength of the fire, position of the lid, effect of the lid on the fire) and refraining from acting, then it could not only merely react, but also predict its actions.
The system doesn’t have a lot of moving parts, and in particular it doesn’t have any parts that would not lead to actions on the environment (the fire). In this case there doesn’t seem to be a way for this system to be anything but a reactive system.
We can now see how the idea that prediction is a form of action performs. It may be the case that the system is intrinsically linked with action, that is for “intelligent oven” that we’re talking about action is prediction, or – to be in line with the definitions presented at the beginning of this post – prediction, expectation and anticipatory behavior are inseparable. Here, the system acts in reaction to the fire. For the behavior to be anticipatory, the behavior should be directed toward events that will happen, not toward events that are happening now.
I have presented these example to show to why we don’t have to ascribe predictive capacities to everything that moves, that does something, even to everything that lives.
But something has to be able to make predictions, right? Sure. When you walk toward your dog that lays on the floor, the dog predicts (based on what it has learned) that you will walk around / over it, but you will not step on it. When there’s 15:40 the dog expects you to be here any minute now! And when you unpack your bags, your dog anticipates your feeding it by moving its food bowl in your direction.
For a system to be able to make predictions, expectations and anticipations, it has to have a non-trivial internal structure and dynamics so that it is able to construct a non-trivial internal model. In short: it has to have a mind, a complex mind.
Prediction, expectation and anticipation in Complex Mind Theory
At first, it may seem as prediction is the same thing as simulation, that is constructing a dynamic scene that goes further than the aggregated ideas it started with. A few things distinguish prediction from simulation, the most important thing is context. Context, in which a given process unfolds has as much, and sometimes even more, bearing on the meaning of the process than the process or its contents itself. Prediction is, at the start, more akin to visual pattern completion. Simulation is contextualized by the events, then it is running “forward”, thus predicting future events, but with lesser degree on context dependence (as the context may not always provide as much information as would be needed to constrain predictive simulation more). When we see a ball falling on the ground, a process of simulation is taking place in our minds. In this context, it acts in the mind as a shallow prediction. When we see a ball falling on the ground, a process of simulation acts as imagination. The context – state of the body, other thoughts, motivations, emotions, situation in which we are – determines the meaning of the simulation process in the mind. Simulation, which is the unfolding of the thought by means of interaction between the ideas, is directed by the context and constrained by the context.
Expectation happens when the context focuses the simulation on the outcome. It also has the quality of “waiting for”, when the general pattern is strong enough to try to subdue other smaller processes into itself, thus interpreting even weak signals as potential signs of the event that is expected by the mind.
A second obvious factor by which simulation and prediction differ is the non-linearity, or the “jumping” character, of some predictions. Having a prediction process going on in the mind, prediction does not necessarily progress in a simple linear fashion, from one thing to the next like in the case of perception, when each thing moves smoothly. Prediction often uses associated knowledge to sharply jump in the future, to the conclusion. In this case, instead of one smooth simulation we have many smaller simulations that correspond to stages through which the prediction process jumps. In the brain, the dynamics of pattern formation that corresponds to a simulation of one stage, through strong connections with other neural ensembles, activates the formation of another stage and is annealed by the second pattern: attention of neural ensembles immediately shifts to the second pattern that is forming. The sudden shift just mentioned is facilitated by the context, in which the processes occur: the context “prediction” sets a goal for the processes of simulation. When a second pattern is more conformant with the goal (the context), the general context focuses on unfolding this second thought.
The analogous case can be made for expectation and anticipation. When a mind expects something to happen, the prediction has created a simulated event, and the general context is richer, it has something more – attention is on looking out for occurrences or signs of the event that we say the mind expects. When a mind anticipates something, the event is expected and the context is even more extensive, as it contains possible ready, just as well simulated, actions and thoughts ready to be used, to be incorporated into the situation which the mind is immersed in. In short, anticipation has the quality of “do when it happens”, that is do (an external (physical) action, an internal dynamics – a thought) something when the anticipated event occurs.
How can prediction work on large time scales
Some species of animals are able to make predictions on time scales that are larger than the immediate experienced context. Usually animals live in the “now”, that is they act in the environment they are currently immersed in, and don’t predict events that are too far out into the future. However sometimes predictions they make span longer periods of time: hours, days. A squirrel that sees a nut on a branch two trees away makes predictions and anticipations about the route to the nut. A raven expects that if it throws enough rocks into the glass of water on which a worm is swimming the level of water will rise and the raven will be able to reach the worm. An ape having a coconut predicts that it needs a solid rock with which to crack the nut. Examples of predictions made by people need not to be stated here.
We observe factual predictions and anticipatory behaviors that jump into the future by hours. One possibility is that the predictive simulations are run very fast, last a little longer, but don’t break or decay. The second possibility is that, through accumulated experience, predictions are made by gluing together chunks of remembered events that fit the situation and the ongoing context of a prediction that is being built chunk by chunk.
Currently we have no tools to rule out any of the alternatives – maybe both of them are being used in different circumstances? Most likely experimental means or concrete models will be required to pin down the specifics of these processes.
Progression from prediction to expectation to anticipation
Does prediction imply that the expectation will follow? And does expectation mean we should see anticipation building up? The answer to both of these questions is most likely “no”. Prediction is more similar to visual pattern completion – it happens automatically, fast, and with little valuation. The mental context in which the prediction is building up does not valuate the progression of the simulation, then the mind is less likely to expect the event to occur. Similarly, if the processes of the mind do not recognize expected events to be valuable or possible to influence, then no relevant anticipatory processes can be constructed and merged into the simulation in such context.
Interruption of prediction, expectation, or anticipation
Prediction, expectation or anticipation of an event that does not happen leads to a sudden interruption of ongoing simulation. The case is very similar to that of understanding. The mind understands the current situation through the perspective of the simulation which is unfolding. The simulation constitutes understanding by forming a pattern of connection and unfolding of relevant ideas, some of which are most likely connected with the situation and some of which pertain to event that are being predicted. The valuation of the future event that is one of the central ones in the simulation varies across prediction-expectation-anticipation spectrum, thus the disruption of the process will lead to stronger reorganizations of the patterns and collapse of the context. Understanding will fade into confusion. Many emotional reactions will arise in the mind, most likely astonishment.
Pezzulo, G., Butz, M. V., Castelfranchi, C., & Falcone, R. (Eds.). (2008). The challenge of anticipation: A unifying framework for the analysis and design of artificial cognitive systems (Vol. 5225). Springer.
Chemotaxis. (2015, January 16). In Wikipedia, The Free Encyclopedia. Retrieved 13:26, January 25, 2015, from http://en.wikipedia.org/w/index.php?title=Chemotaxis&oldid=642723808