Locked and unlocked strategies are at the center of this article, as ways of shedding new light on the cognitive aspects of deep learning machines. The character and the role of these cognitive strategies, which are occurring both in humans and in computational machines, is indeed strictly related to the generation of cognitive outputs, which range from weak to strong level of knowledge creativity. I maintain that these differences lead to important consequences when we analyze computational AI programs, such as (...) AlphaGo, which aim at performing various kinds of abductive hypothetical reasoning. In these cases, the programs are characterized by _locked_ abductive strategies: they deal with weak (even if sometimes amazing) kinds of hypothetical creative reasoning, because they are limited in what I call eco-cognitive openness, which instead qualifies human cognizers who are performing higher kinds of abductive creative reasoning, where cognitive strategies are instead _unlocked_. (shrink)

This article presents a computational model of the learning of diagnostic knowledge, based on observations of human operators engaged in real-world troubleshooting tasks. We present a model of problem solving and learning in which the reasoner introspects about its own performance on the problem-solving task, identifies what it needs to learn to improve its performance, formulates learning goals to acquire the required knowledge, and pursues its learning goals using multiple learning strategies. The model is implemented in a computer system which (...) provides a case study based on observations of troubleshooting operators and protocol analysis of the data gathered in the test area of an operational electronics manufacturing plant. The model not only addresses issues in human learning, but, in addition, is computationally justified as a uniform, extensible framework for multistrategy learning. (shrink)

Default reasoning occurs whenever the truth of the evidence available to the reasoner does not guarantee the truth of the conclusion being drawn. Despite this, one is entitled to draw the conclusion “by default” on the grounds that we have no information which would make us doubt that the inference should be drawn. It is the type of conclusion we draw in the ordinary world and ordinary situations in which we find ourselves. Formally speaking, ‘nonmonotonic reasoning’ refers to argumentation in (...) which one uses certain information to reach a conclusion, but where it is possible that adding some further information to those very same premises could make one want to retract the original conclusion. It is easily seen that the informal notion of default reasoning manifests a type of nonmonotonic reasoning. Generally speaking, default statements are said to be true about the class of objects they describe, despite the acknowledged existence of “exceptional instances” of the class. In the absence of explicit information that an object is one of the exceptions we are enjoined to apply the default statement to the object. But further information may later tell us that the object is in fact one of the exceptions. So this is one of the points where nonmonotonicity resides in default reasoning. (shrink)

In this paper, the problem of purifying an assumption-based theory KB, i.e., identifying the right extension of KB using knowledge-gathering actions (tests), is addressed. Assumptions are just normal defaults without prerequisite. Each assumption represents all the information conveyed by an agent, and every agent is associated with a (possibly empty) set of tests. Through the execution of tests, the epistemic status of assumptions can change from "plausible" to "certainly true", "certainly false" or "irrelevant", and the KB must be revised so (...) as to incorporate such a change. Because removing all the extensions of an assumption-based theory except one enables both identifying a larger set of plausible pieces of information and renders inference computationally easier, we are specifically interested in finding out sets of tests allowing to purify a KB (whatever their outcomes). We address this problem especially from the point of view of computational complexity. (shrink)

In multicausal abductive tasks a person must explain some findings by assembling a composite hypothesis that consists of one or more elementary hypotheses. If there are n elementary hypotheses, there can be up to 2n composite hypotheses. To constrain the search for hypotheses to explain a new observation, people sometimes use their current explanation—the previous evidence and their present composite hypothesis of that evidence; however, it is unclear when and how the current explanation is used. In addition, although a person's (...) current explanation can narrow the search for a hypothesis, it can also blind the problem solver to alternative, possibly better, explanations. This paper describes a model of multicausal abductive reasoning that makes two predictions regarding the use of the current explanation. The first prediction is that the current explanation is not used to explain new evidence if there is a simple (i.e., nondisjunctive, concrete) hypothesis to account for that evidence. The second prediction is that the current explanation is used when attempting to discriminate among several alternative hypotheses for new evidence. These hypotheses were tested in three experiments. The results are consistent with the second prediction: the current explanation is used when discriminating among alternative hypotheses. However, the first prediction—that the current explanation is not used when a simple hypothesis can account for new data—received only limited support. Participants used the current explanation to constrain their interpretation of new data in 46.5% of all trials. This suggests that context‐independent strategies compete with context‐dependent ones—an interpretation that is consistent with recent work on strategy selection during problem solving. (shrink)

Many recent developments in artificial intelligence research are relevant for traditional issues in the philosophy of science. One of the developments in AI research we want to focus on in this article is diagnostic reasoning, which we consider to be of interest for the theory of explanation in general and for an understanding of explanatory arguments in economic science in particular. Usually, explanation is primarily discussed in terms of deductive inferences in classical logic. However, in recent AI research it is (...) observed that a diagnostic explanation is actually quite different from deductive reasoning. In diagnostic reasoning the emphasis is on restoring consistency rather than on deduction. Intuitively speaking, the problem diagnostic reasoning is concerned with is the following. Consider a description of a system in which the normal behavior of the system is characterized and an observation that conflicts with this normal behavior. The diagnostic problem is to determine which of the components of the system can, when assumed to be functioning abnormally, account for the conflicting observation. A diagnosis is a set of allegedly malfunctioning components that can be used to restore the consistency of the system description and the observation. In this article, this kind of reasoning is formalized and we show its importance for the theory of explanation. We will show how the diagnosis nondeductively explains the discrepancy between the observed and the correct system behavior. The article also shows the relevance of the subject for real scientific arguments by showing that examples of diagnostic reasoning can be found in Friedman's Theory of the Consumption Function. Moreover, it places the philosophical implications of diagnostic reasoning in the context of Mill's aprioristic methodology. (shrink)