There are arguments and then there are arguments...
Deductive, inductive, and analogical reasoning
Reasoning is so ubiquitous that our first job has been to find a workable definition. We need to come to grips with what reasoning is (and is not) in order to figure out how generative AI can help us be better at it. We have settled on a definition that will limit our focus to three types of activities: deciding, problem-solving, and arguing. But of these three activities one is the queen: arguing. This is not only the one that is most studied, but is also so multifaceted, that often what applies to some element of arguing can be adapted to either decision-making or problem-solving.
So let the arguments begin.
Even restricted to just argumentation, the act of reasoning remains varied and complex. There are at least three types of arguments that are often used by people. The first, and rarest, is deductive reasoning, where the conclusion is guaranteed to be true if the premises of the argument are true. You may be familiar with the type of reasoning if you have studied logic or syllogisms. The second type is inductive reasoning, much more widely used, especially in the sciences, where the conclusions are likely, but not guaranteed, to be true if the premises are true. The final type is analogical reasoning in which characteristics of two events or objects are compared to demonstrate that something that is true of one is also true of the other.
Of all the types of reasoning, deductive was once thought to be the crowning achievement of human intelligence. Because the conclusion of a deductive argument is guaranteed to be true if the premises are true, this form of reasoning offers the surest route to irrefutable beliefs. So sure, in fact, that it was championed by a very influential psychologist of the 20th century, Jean Piaget, as “the crown of cognitive development” (Stenning and Lambalgen, 2008). Piaget posited that deductive reasoning was in fact how humans reasoned once they achieved a certain age.
Piaget’s empirical claim was tested and did not fare well. The most famous of these tests is the Wason selection task, a logic problem developed in the 60’s. Participants are shown a series of four cards, two of which are face-down. The two face-up cards are a three and an eight. One face-down card is brown, and one is red. The participant must decide which cards to flip over to confirm the claim that every even-numbered card has a red back. The answer requires the application of a series of logical inferences, and yet less than 10% of participants were able to solve the puzzle (Mercier & Sperber, 2011). Even in a controlled environment people rarely use deductive reasoning well.
So can we deductively reason? Of course. Do we do it well? No.
More promising, and much more common, is inductive reasoning. This is the form of reasoning that has powered the scientific revolution, and arguably, our success as a species. Inductive reasoning depends on a series of observations, usually observed regularities like the sun rising in the east, and generalizes over them to reach a conclusion, for example, “the sun always rises in the east”. Inductive arguments are not truth-preserving, they are truth-promising. The conclusion is likely to be true if the premises are.
Finally, analogical reasoning is as common as inductive, perhaps more so. It compares two events or items, and concludes that because they share some relevant features, some other thing that is true of one must also be true of the other. It is often used in our professional and personal lives. Here is just one such example. The movement to decriminalize/legalize marijuana has been fairly successful in the United States, with new states constantly being added to the list of those that have made the drug available to their residents. Advocates often note that marijuana’s risks and benefits are like alcohol. If one is legal, then the other one should be too. Although easy to execute, analogical reasoning has a noticeable flaw: there are no hard and fast rules for how similar the two things need to be. This leads to a lot of faulty analogies. It is nevertheless a common and important form of argumentation.
With our new army of definitions — deductive, inductive, and analogical reasoning — we are now in an excellent position to figure out how generative AI can help us. By understanding its abilities within each sub-type, we can pinpoint the best ways to use it to improve our thinking.
It is now time to leave theory behind and begin exploring just what generative AI can and cannot do in the domain of reasoning.
References (and very interesting reading on reasoning):
Mercier, H., & Sperber, D. (2011). Why do humans reason? Arguments for an argumentative theory. Behavioral and Brain Sciences, 34(2), 57–74.
Stenning, K. and van Lambalgen, M. (2008). Human Reasoning and Cognitive Science. MIT Press.