Predictive knowledge vs predicting the future

I am reading von Mises’ Human action. A Treatise on Economics and in Chapter VI – dedicated to Uncertainty – there is almost an aphoristic passage that says:

If man knew the future, he would not have to choose and would not act. He would be like an automaton, reacting to stimuli without an will on his own.

He then makes another bold statement about the kind predictability that is assumed in the so-called natural sciences:

Natural science does not render the future predictable. It make it possible to foretell the results to be obtained by definite actions.

A few pages earlier he had made another interesting remark:

The temporal order in which knowledge is acquired must not be confused with the logical simultaneity of all parts of this aprioristic deductive system. […] Anteriority and consequence are essential concepts of praxeological reasoning. So is the irreversibility of events.

What von Mises is very clear about is that, whenever we approach human action to study it (hence the term praxeology – the science of action), we tend to generate, say, “systems of thought” – such as those belonging to logic and mathematics – that are simply out of time. This means, more specifically, that all events and occurrences – including our own actions, either as consequences or motives  – are seen synchronously as part of a broader picture that is cast upon reality. There are no lapses of time. Nor are there past, present and future.

The suppressing of time is  a typical characteristic of an objective approach to the world and human affairs. Notions such as after and before, antecedent and consequent, etc. make sense because they are part of a sequence of events that we can arrange before our eyes – this leads to this that leads to this, etc..

As von Mises brilliantly notices, indeed we can come to predict the consequence of an action. For example, we put coffee on the stove and  we can quite accurately predict that the water will start boiling and coffee will be ready. However – and this is a central point – the accuracy of a prediction (in other words, whether we get it right or not) is always relative to the conditions under which we make our prediction. Let’s go back to the example of making coffee.

Indeed, the water on the stove will come to boil. But, if we use an electric stove and after a few minutes there is a blackout, it simply won’t. Is it an exception? As paradoxical as it may sound, It is not an exception to the rule. The conditions under which the rule can be applied is the exception. The simple fact that we can wake up and make coffee implies a concomitant occurrence of a number of events that would make us tremble in disbelief.

The more general lesson that we derive from von Mises’ treatment is that we can predict the consequences of certain specific actions. But that does not mean that what we are doing is predicting the future. And the reason is that predicting the future would imply predicting something regardless of  the actual conditions.

 

Simulations, duplicates and the fallacy of misplaced concreteness

Searle has a very interesting argument – among many others – against the strong AI program (and all the rest of it). He says that simulation is not duplicate (he seems to love punchlines, by the way). This means that we can simulate, say, cognition in a computer, but that does not mean that we have then a duplicate of cognition, that is, something that IS cognition in the computer. Ontologically speaking, what a simulation creates (or embodies) is nothing but a model, a representation. What Searle adds is that it lacks causal power. This is obvious, if we think of a flight simulator. It can be as accurate as possible, as close to reality as possible. Yet, guess what? Nobody would ever think that with that machine we could actually fly. Indeed! As Searle says, it lacks causal power, that is, the power to do exactly the same thing as a real aircraft.

Now, why is it that we think that with cognition (or intelligence or reasoning) things are different? A full answer to this is worth an entire philosophy course of AI and computation. But one short way to answer to the question is that we are committing a fallacy, the so-called “fallacy of misplaced concreteness”, which was introduced by British mathematician and philosopher Whitehead. The fallacy states that when we come up with a theory or a model of how something works, for example, our cognition, we tend to assume that that theory or model is characterized ontologically by the same very way as that which it models or theorizes about. So, a theory of cognition is (erroneously taken for) cognition. Or a theory of the brain is the brain. As we know, theories of the brain have changed over the past centuries, while the brain has interestingly remained the same over the same period of time.

Indeed, there is one thing that changes. A new theory of the brain or cognition may change the way we use the brain or our cognition! And this is worth a few more words. Suppose that I say I have 100 Euros in my pocket when in fact I have nothing. Would saying that  have that money magically make 100 Euros appear in my pocket? Indeed, nobody would say Yes. The difference is eminently ontological. Same thing as above.

However, in a way, it is true that those 100 euros exist. But they exist as a mere verbal representation – something that I say. Indeed, 100 euros would not exist as, say, a banknote to exchange at the supermarket. Now, this representation is not without meaning. It is a representation after all.  So, if I say to you that I have 100 euros, and you think that I am a man of my word (pun intended), then you would start acting as if I actually have that money in my pocket. So, you may give me, say, your watch, because, well, I have the money!

Now, coming back to cognition and computers, this simple example shows one important thing. That the issue concerning the attribution of some sort of “cognition-like” power to computer becomes a matter of practical concern. That is, it is more about what we then do than what it actually is.

Interaction, incorporation of new skills and abilities, and the meaning of educational technology

In preparation of one of my courses for the fall – the one dedicated to creative re-use and tinkering – I decided to read again Chapter 1 of Where the action is by Paul Dourish. The book was published in 2001, but I believe it’s still a very good reading even in 2017.

In this first Chapter Dourish merges our ideas of what interaction is with the history of computation. Dourish walks us through some 60 years of history during which computation has increasingly colonized our life. The main argument put forward by Dourish is that over the last decades we have seen a widening of the range of skills and abilities being incorporated in the interaction with computers. There was a time when only experts could interact with a computer, which, by the way, would occupy a space as large as a medium-seized gym. Now it’s completely different.

We have here a curious phenomenon. On the one hand, the actual interaction with computers is dramatically enriched. For example, since the introduction of the so-called graphical user interface (GUI), I can incorporate in the interaction with a computer – and thus make use of – my spatial reasoning skills, which radically improves the usability of a so-called computer. On the other, the incorporation of an increasingly wider range of skills and abilities made the so-called “technical knowledge” not so fundamental as it was before to operate a computing machine. It is not that all of a sudden we no longer need engineers and programmers. We still need them to build a computer, but when it comes to  interaction, and consequently the kind of experience we may or can have, things have changed. Dramatically.

To put it another way: the interaction with a computing machine is less and less tied to the kind of knowledge we need to build it. It follows that what we can do with computers is less affected by our “technical” skills, while it is increasingly affected by other skills and abilities we have, which, in turn, are less and less related to building the thing.

Now, one thing that we may add to Dourish’s argument is that in adding layers to the interaction with computing machines, we lose control, as complexity kicks in. Computers were used at the beginning for specific tasks and they needed an entire task force of engineers to operate. Not quite the same as today. But the price we pay is, as I said, losing control. So, once the interaction of a computer incorporates social skills, this also means that the computer gets embedded in a much more complex “game” – the social one. The result is that we can no longer assume the same kind of linear connection between a machine and its function.

This is to me quite a big issue when it comes to educational technology. It is clear to me that educational technology is not something technical. It is not about becoming engineers or getting the right sets of skills. The computer is already interacting and thus incorporating skills and abilities pertaining with teaching and learning. But here comes the big question: have these skills and abilities changed? Or are we still dealing after all with teaching and learning?

 

Practical deductions vs practical abductions (or algorithmic thinking vs heuristic thinking)

Deduction has the function of a decision. Deduction can be defined as the act of subsuming a case under a general rule. This means that the resulting action (what in a syllogism would be the conclusion) is identified by the application of a general rule (if x, then y) to a particular case b.

If x then y.

b is x.

Then, y

Such a framework for decision is also called “purposive-rational action” (see my previous post). This model or frame of decision-making is in essence the practical side of algorithmic thinking. Algorithmic thinking proceeds this way.  It is the generation of a chain of practical deductions. Since it’s logical, it can easily be embedded in a computer and in this way it makes decisions without any further human intervention (provided that a viable system of input and output is in place).

This approach based on practical deductions “works” – so to say, if we can connect the particular case at hand back to a known category –  a sort of specimen. This is what the act of subsuming a case under a general rule actually means. So, a case is never taken for what it is, but as a member of a broader category. This makes very hard for a practical deductive system to handle any case that is not immediately connected back to a known category.

Interestingly, a practical deductive system may actually force a particular case into a known category. Practical deductive systems are in fact procrustean beds (I will come back to this point in another post). There are indeed false positive and false negative to deal with.

Practical deductions never quite deal with individual cases. A case is always a token of a type – to use a more technical jargon. In other words, it needs to engage with abstraction. Which means in this case to reduce or attenuate the number of levels of (possible) descriptions (see Mark Johnson’s post on this). Assigning a case to a broader category means precisely reducing the variety of ways in which we can describe something to a single one (or a few). This is the problem when we simply classify things.

Leaving procrustean beds aside for the time being, practical deductions are not particularly good when it comes to actual decision-making. Indeed, they look good. Very good. That’s because they rely on purposeful deliberation: we know what we are going to do and why. However, they are not able to adapt. They need explicit knowledge. They do not tolerate uncertainty and ignorance. While practical deductions are safe and provide a good deal of certainty and control, they inevitably deal with fixed abstract entities and therefore do not adapt to circumstances. (Seemingly, practical deductivism maps pretty much on what the great Alexander Luria said about classical science, which he viewed as “the reduction of living reality with all its richness of detail to abstract schemas”.)

Let me now turn to practical abductions, which I started dealing with in a previous post.

Essentially, in practical abductions we are not able to recognize the case at hand. And that impairs the possibility to apply a practical deductivist approach. An interesting example is unlocking the door. If everything works, what we have to do is to put the key in the lock and turn the key left (or counterclockwise). The thing is, if something goes wrong (because of a faulty lock, for example),  we are plunged into a state of ignorance: what should we do? Practical deduction is a powerless option precisely because it requires that the case is known. The case at hand cannot be subsumed under the general rule, simply because we don’t know what the case is.  We need therefore a creative type of practical inference. Here is where practical abduction comes into play.

In the syllogistic framework, an abduction is composed of a major premise, which is a piece of knowledge, a habit, a rule, or, more in general, something that we assume to be true; a minor premise, which is something like a cue, a trace left that we detect in the here and now. The conclusion of the abduction establishes a state of things – a case. So, I can make the following example:

Paul comes to office with his laptop (Major premise – habit).

I see Paul’s laptop on his desk (Minor premise – cue)

Paul has come to the office (Conclusion – case)

What makes abduction particularly interesting is that the generation of a conclusion depends on the selection of a major premise, which happens in connection with the cue, the trace left. In a non-practical abduction the conclusion we reach is an explanation.

In a practical abduction we have something similar. First of all, we have something like a cue, a trace left. In our example it is that we can’t turn the key right. The major premise is not a piece of knowledge, but an action, a doing. It can be, for example, pulling the key out of the lock by a few millimeters. This action is somehow cued by the situation. We simply can’t turn the key right. Now, this action – this doing – has then an effect on the world. So, the “conclusion” is not an explanation like in the case of a non-practical abduction. It is some sort of change that we see happening in the world. If we prefer, it is a manipulation of the world brought about by an action (pulling out the key) cued by something in the environment (we can’t turn the key right) that we took as relevant.

What is interesting is that the conclusion in the practical abduction is known only after we have actually acted. This is what gives the sense of tentativeness to practical abduction. Indeed, the resulting change can meet our expectations or not. Or it may create something that eventually meet our expectations, although it was not intended to be so. This is the case of serendipity. Now, what is interesting to stress is that this way of proceeding is not algorithmic, that is, governed by rules that we apply to known cases. The selection of the major premise – the action – has a pure heuristic function. We do something in the expectation that the change produced will somehow be helpful for us to understand a bit better the world around us. But there is no certainty.

 

 

 

The practical side of abduction. Some initial thoughts

Ten days ago I met in Helsinki Merja Bauters and Sami Paavola. We had a long conversation about big data, algorithmic thinking, learning analytics and inquiry. Both Merja and Sami are very familiar with the work of Peirce and to me that was a very good reason to share some of the thoughts that I had about abduction.

The conversation made me think (a lot) about the practical side of abduction. Abduction is usually mentioned to describe the process whereby we select or generate conjectures or guesses. It’s interesting that abduction is somehow still widely under-recognized. It is often mistaken for deduction or induction. It has been acknowledged only recently as an important concept guiding and informing qualitative research in the social sciences (see the term abductive analysis).

A very simple example of abduction is this: I come to the office, I see that there is a laptop on the desk of my colleague. Therefore, I infer that my colleague has already arrived to the office. So, abduction accounts from the process in which I starts from a set of clues to formulate a conjecture that would explain the presence of those clues.

If this is a very important step forward towards understanding a bit better inquiry and creative processes, there is still one thing missing: how did we come up with those clues in the first place? Clues are not “locked”, but they dynamically appear, disappear, and re-appear. This question leads us to wondering if abduction has also a practical side. Abduction provides a fantastic tool for understanding the way  in which we explain the world. Can it do the same with understanding the way in which we inter-act with and in the world?

Interestingly, Habermas in his Knowledge and Interest (1972) – commenting on Peirce – points out:

deduction has the function of a decision.

And then he adds:

The conclusion to which it leads is a specific behavioral reaction resulting from the application of a general rule of behavior to a singular case” (p. 122 in the English translation).

He goes on citing Peirce who writes:

the cognition of a result is of a nature of a decision to act in a particular way on a given occasion

One page later Habermas claims that

the act of a purposive-rational action can be understood as the performance of a deduction.

So, deduction has its own behavioral correlate in what he calls “purposive-rational action” or  “instrumental action”.

When it comes to abduction, Habermas simply follows Peirce who says that what correlates with abduction is “the stimulus that sets off an action”. That is, a “sensory element”, which is immediate. For Habermas claims that abduction identifies the conditions under which a rule can be applied. The abductive inference is an inference to the case. All this does not really help us understand if abduction has actually a practical side.

Magnani in his Abductive, Reason and Science (2001) contemplates the possibility that some abductive inferences are made outside language. He actually claims that there is a type of abduction that is manipulative. That is, it happens “through doing”. In his Abductive Cognition (2009), he writes:

In this perspective manipulative abduction is a specific case of cognitive manipulating in which an agent, when faced with an external situations from which it is hard or impossible to extract new meaningful features of an object, elects or creates an action that structures the environment in such a way that it gives new information which would be otherwise unavailable and which is used specifically to infer explanatory hypotheses. (p. 173)

This echoes somehow the distinction between pragmatic and epistemic action introduced by Kirsh and Maglio in mid 90s. Unlike pragmatic actions, epistemic actions are meant to uncover new and potentially useful pieces of information. Imagine that we have to unlock a door. We put the key in the lock, but we can’t turn it. Then we start doing something until we sort of feel that the key is turning. The resulting chain of actions we are engaged with is meant to uncover new information. What is interesting in this case is that the action is not meant to “apply a rule” like in the deductivist frame Habermas describes. The application of a rule (the major premise in the syllogistic framework) would imply that the case (to which the rule applies) is known, when in fact it is not. Normally, we would say that what we need to do is to engage in some trial-and-error. But this would miss the point, because it would represent what is an inquiring process as somehow random. When in fact the process is heuristically driven. That is, it is carried out so as to find out something. This stands in opposition to the algorithmic way of proceeding typical of deduction.

In this way we arrive to the core of the practical side of abduction. the cognitive manipulations Magnani refers to can be considered as the result of abductions. In direct analogy with non-pratical abductions, the action is analogous to the case that is abduced from a fact – a stimulus – and a rule – the major premise, which however is not applied, but used to uncover the possible course of action – what should be the case.

More thoughts to come.

Learners as chance-seekers

I have recently completed the draft of a paper, which will be published (in English) in a collection, whose title is: Digitalität und Selbst – Interdisziplinäre Perspektiven auf Subjektivierungs- und Bildungsprozesse.

The article is theoretical and it is an attempt to make explicit the connections between learning and chance-seeking.

Here is the PDF of the draft.

Tinkering, subversiveness and conviviality

During the last EARLI conference here in Tartu Heidrun Allert and Christoph Richter had an interesting presentation in which they claimed that tinkering is subversive. Or, better, it is potentially so. I agreed wholeheartedly. The idea is that a piece of technology (a smartphone) or a service (Instagram or Facebook) comes, say, “in a package”. And users are not supposed to “unpack” it. I remember when I bought my second Mac and to my surprise the new design did not allow the owner to replace the battery. The iPhone itself was “locked”, but skilful users managed to unlock it and spread the word on the net.

Recently, I have bumped into an interesting example of tinkering, in which its subversiveness is quite on display. Instagram has a quite strict policy when it comes to  sexual content pictures. So, for example, pictures exposing nipples or bare breast are banned. The same for the male naked body and all the rest of it. Yet for many Instagram is a tool of full personal expression and its policy may limit users’ freedom to use it. Interestingly, this is partly true. Indeed, some simply put a little sticker on their nipples or crotch. In some other cases I saw that some women have erased their nipples and that made them look like a doll – de-sexualized. However, this limitation can be turn into something to tinker with. Instagram user louielewisss provides an interesting illustration of tinkering with censorship: https://www.instagram.com/louielewisss/

What I would like to draw the attention to is that tinkering exploits the same very constraints that are operating, but at the same time via tinkering we potentially creates something different. So, to simplify, there is no need to put sticker (which would make one’s attempt to be erotically artistic an epic fail). But one can play hide and seek in ways that only one’s imagination can really know. This is also capture by the proverb: necessity is the mother of invention. Necessity is in this case the set of  constraints that are forced upon us, which, however, become the sort of trigger for invention.

Subversiveness can also be played out in a different way. And here comes my second example.

I have a facebook friend who always posts incredibly suggestive photos that she herself takes. I sometimes comment, but sometimes I feel that my words are powerless to communicate my feeling of awe. I am also a bit tired of this “Like” culture that Facebook has contributed to spread. A Like is now an utterance that means anything and nothing. In addition to that, Likes have become for Facebook a powerful weapon to allegedly know what we want and therefore the kind of post to display in our feeds. But to come back to my friend’s post, on that occasion I decided to try to use a picture to express and so communicate my feeling (see the picture right here). And I picked one from a film. I did this pretty much by chance. I actually stumbled upon it using the image search. But I realized retrospectively that that is not a bad way. Who better than an actor can express sentiments and feelings?

Indeed, I was tinkering. Facebook Like does not provide me with an opportunity for full expression. But I can tinker. Accidentally, it turns out that this way it’s much harder for Facebook algorithms to “guess” what I am doing – whether I liked that picture or not. And it feels much more human, because the other person can have a better hint as to what her picture made me feel and experience (a hint, indeed!). Here again, tinkering is subversive.

There is now a last thing to add, which is not about subversiveness but conviviality. I refer here to Ivan Illich and its specific interpretation of the concept. Illich wrote in his Tools for conviviality:

I choose the term conviviality to designate the opposite of industrial productivity. I indent it to mean autonomous and creative intercourse among persons, and the intercourse of persons with their environment.

This is something I see implicit in the notion of tinkering. That is, tinkering is an attempt – either tacit or not – to turn a tool into a convivial tool. That, tinkering can be viewed as an activity in which we use our tool so as to to create zones of conviviality. More thoughts will come.