Two weeks ago I participated in a hackathon organized by Riga Technical University within the framework of FuturICT 2.0 project (Large-scale experiments and simulations for the second generation of FuturICT). During the hackathon I delivered a talk on educational data. The goal of the presentation was rather modest: I wanted the folks there to reflect on the way in which we engage with data in education. Essentially, the message was that we manipulate data and we can do that in different ways depending on the type of inference that we are drawing “from the data”. We can look for confirmations, similarities among phenomena or explanations. I pretty much referred to Peirce’s well-known classification of inferences – deduction, abduction, induction. I did not talk about a forth category, which is retroduction.
After the presentation, I started thinking about how to present these four types of inferences or, better, epistemic activities (activities that deal in one way or another with knowing). I usually use the syllogism to explain the differences, but I was looking for something simpler.
Let me start with deduction. What do we do with deduction? If deduction is used theoretically, it generates predictions that are meant either to confirm a hypothesis or not. It does that by deriving the consequences that logically follow, when we assume that something is the case. From example, all men are mortal. And if I am a man, then I am going to die sooner or later. When it’s used practically, deduction prescribe a certain course of action. The prescription derives from subsuming a case under a general rule. For example, when one is ill, it’s better to stay at home and rest. This means that, if I happen to be ill, I should not go to the party.
deduction predicts or prescribes
Induction is a different matter and perhaps a bit more complicated. Or not so straight-forward as deduction. Here I have to start with an example. Let’s say that I meet a thin man and I discover that he is a runner. Then, I meet a second man who is thin and who is a runner. Then, I meet a third one, who is the same as the previous two. After a few observations I start thinking that runners are thin. In doing so, I am taking single observations (e.g., the thin man is a runner) and I generalize them. Such generalization – it is very important to note – can be quantitative or qualitative.
Qualitative generalizations are based on numbers. So, I say that I verified that 80 out of 100 runners are thin. Then I generalize that this is true for the entire population of runners in, say, Tartu, which is the city where I live. If in Tartu there are, say, 1000 runners, then I can draw the conclusion that 800 of those are thin.
Qualitative generalizations are different and in a way more interesting. Let’s say that there is a certain disease D that manifests itself with a list of symptoms a, b, c. By making a inductive generalization, I attribute those symptoms to any single instance of that disease D. So, all cases of that disease will have the set of symptom a, b, c.
It’s important to note that induction does not really produce new theoretical knowledge. It allows us to see things as similar. So, it seems that induction is essentially related to making similarity judgements. This can be done independently from the number of observations. If a boy touches the hot stove with his fingers once, he will inductively derive that the next time he does the same, he may expect something similar to happen. So, the generalization that the induction does is related to similarity.
Induction generalizes and finds similarities
Now comes abduction. Abduction is often mistaken for induction. The example that I have just made is perfect to show the difference with induction. The boy touches the hot stove and he burns his finger. He may assume that the next time he is going to do the same, the same will happen, because the second time will be similar to the first time. However, induction does not tell us anything about the finger. This is done via abduction. When the boy touches the stove, he perceives that his finger burns. That is done via abduction and it deals with the recognition that something is the case. In this case the recognition happens at the perceptual level. It’s performed, in other words, by our senses. Abduction performs an act of synthesis, as it brings under the same umbrella different elements. To make an other example, when we see a face, we don’t see the single individual components of that face. We see the face as a whole, which is in this case a perceptual synthesis of all the elements composing it (most of which we cannot really name). So,
Abduction recognizes wholes
Now comes the last in the list: retroduction. This is still the trickiest of all. So, what I am sharing now is very tentative. In general, I tend to see retroduction as a sort of bag containing different molds. For example, I am not a wine expert. And if you give me a glass of wine, I can’t really say much about it. I can essentially say if I like the wine you give me or not. But if you take a wine expert, he or she will tell you way more than me. Now, we both can recognize that something is wine and not coke. The reason is, to put it simply, that we both have the category “wine” and “coke”. What we do is that we are able to create a synthesis of all the clues that are coming to us when we drink something. However, the wine expert has way more “categories” than me. That’s why he or she recognizes many more differences. It’s a little bit like the doctor. The doctor has a sort of library of diseases, which allows him to make sense of the patient’s symptoms. We can call them symptomatologies. Now, retroduction answers to the question concerning the development of those symptomatologies. So, retroruction does not tell us that this is pneumonia or Lambrusco wine. Retroduction provides the “molds” to then capture things in the world. Or, better, to take them off from the background and bring them to the foreground. So,