Mercier and Sperber's (M&S's) argumentative theory of reasoning (ATR) claims that reasoning is aimed not at improving knowledge, but at persuading. According to ATR, an argumentative context (actual or proactive) is a necessary condition for reasoning and reasoning is biased toward winning the argument. It will be argued that the very logic of ATR is problematic and that it cannot deal with a large majority of reasoning contexts in which agents (also) reason to determine a true or valid conclusion. I propose that a theory of reasoning should incorporate utilities of reasoning outcomes, to explain various motives for reasoning, including argumentative motives.
Although M&S discuss in detail the function of reasoning, it is not always clear how reasoning is defined. If we assume reasoning to be about opinions and preferences for which no normative standard is available, an argumentative view is quite straightforward. People indeed may argue about political opinions and preferences with no other goal than to convince. However, if reasoning is the treatment of information about some (yet unknown) true or valid inference, with the objective to reduce inference errors, then the argumentative theory is puzzling. Indeed, the ATR theory of reasoning disregards this inherent concern of approximating the valid conclusion, going against what reasoning is defined to be.
The uneasy coexistence of finding out the truth and seeking confirmation for one's belief in ATR is apparent in M&S's analysis of the confirmation bias in hypothesis testing studies (Wason Reference Wason1960). On the one hand, M&S acknowledge, in line with classical critical analyses of the confirmation bias (Klayman & Ha Reference Klayman and Ha1987; Poletiek Reference Poletiek1996; Reference Poletiek2001; Wetherick Reference Wetherick1962), that participants' behavior in these studies is not indicative of a biased search of supporting test outcomes, but that it reflects a sound heuristic. This heuristic is testing hypotheses with positive predictions. Using M&S's example, suppose I believe that my keys are in my jacket because that is where I remember putting them. I look for them in my jacket (positive testing) and not in my purse (negative testing). Hence, as opposed to the interpretation of positive testing as a tendency to confirm (confirmation bias) (Cowley & Byrne Reference Cowley, Byrne, Bara, Barsalou and Buchiarelli2005; Wason Reference Wason1960), positive testing may occur with no other goal than finding out the truth, as the real-life example suggests. According to ATR, positive testing is a default heuristic that involves no reasoning proper.
However, as M&S further argue, reasoning can be triggered in hypothesis testing situations if an argumentative context is provided. Moreover, in such a context, reasoning is directed at falsification, according to ATR: Falsification is accessible provided that the situation encourages participants to argue against a hypothesis that is not their own, as M&S propose This logic reveals the old misinterpretation that test choice (positive or negative) is tuned at the emergence of favorable test result. In fact, putting one's idea to a (either positive or negative) test assumes the objective to find out the truth and is therefore at odds with testing in order to save our idea from confirmations or disconfirmations. Poletiek (Reference Poletiek1996) showed that participants in a hypothesis testing experiment are aware of the incompatibility of finding out the truth and coming up with confirming test outcomes only. Participants felt that they were looking for information about the validity of a hypothesis, and that they could not control the test outcomes by choosing a test strategy. It was only their expectation of the outcome that changed when they tested a hypothesis they did not believe to be true (falsification was expected). When testing their best guess, they expected a confirming outcome.
Besides determining the truth, many considerations may affect how we reason and whether we reason. These considerations can be modeled as a cost-benefit analysis of making the right inference. In Wason's (Reference Wason1960) task, students participating in the experiment might not care much about making reasoning errors. The key seeker might perform a negative test (looking for the keys in some place other than the jacket) because the costs of a positive one are too high: The jacket was left behind in a friend's house and the effort too great to walk back to check the jacket. Alternatively, we might be in a hurry and have time to check just one location, enhancing the benefits of good reasoning.
To predict reasoning behavior, we need a model with utilities of reasoning outcomes. A suitable tool is signal-detection theory. Motives can be expressed in the utilities of inference errors and in reasoning behavior predicted on the basis of the risk we are prepared to take with regard to particular erroneous inferences (Poletiek & Berndsen Reference Poletiek and Berndsen2000). For example, as shown in Table 1, a pure epistemic motive would be modeled with (A) low utilities of making any false inferences. A pure argumentative motive would be expressed in (B) a very high willingness to make the false inference that our favorite hypothesis is true; and (C) reasoning with both motives (searching for a valid inference within both practical or argumentative constraints) is reflected with some in-between utilities with regard to making a false or a correct inference about our favorite hypothesis.
Table 1.
Utilities of reasoning outcomes about H (the reasoner's subjective belief), assuming an epistemic motive (A), an argumentative motive (B), and both types of motives (C).
In this manner, reasoning in a variety of contexts and with a variety of goals can be modeled, offering an elegant alternative to the paradox of the ATR that we start up a reasoning trajectory about a prior belief if and only if the end of the route leads us to inferring that belief again.
Mercier and Sperber's (M&S's) argumentative theory of reasoning (ATR) claims that reasoning is aimed not at improving knowledge, but at persuading. According to ATR, an argumentative context (actual or proactive) is a necessary condition for reasoning and reasoning is biased toward winning the argument. It will be argued that the very logic of ATR is problematic and that it cannot deal with a large majority of reasoning contexts in which agents (also) reason to determine a true or valid conclusion. I propose that a theory of reasoning should incorporate utilities of reasoning outcomes, to explain various motives for reasoning, including argumentative motives.
Although M&S discuss in detail the function of reasoning, it is not always clear how reasoning is defined. If we assume reasoning to be about opinions and preferences for which no normative standard is available, an argumentative view is quite straightforward. People indeed may argue about political opinions and preferences with no other goal than to convince. However, if reasoning is the treatment of information about some (yet unknown) true or valid inference, with the objective to reduce inference errors, then the argumentative theory is puzzling. Indeed, the ATR theory of reasoning disregards this inherent concern of approximating the valid conclusion, going against what reasoning is defined to be.
The uneasy coexistence of finding out the truth and seeking confirmation for one's belief in ATR is apparent in M&S's analysis of the confirmation bias in hypothesis testing studies (Wason Reference Wason1960). On the one hand, M&S acknowledge, in line with classical critical analyses of the confirmation bias (Klayman & Ha Reference Klayman and Ha1987; Poletiek Reference Poletiek1996; Reference Poletiek2001; Wetherick Reference Wetherick1962), that participants' behavior in these studies is not indicative of a biased search of supporting test outcomes, but that it reflects a sound heuristic. This heuristic is testing hypotheses with positive predictions. Using M&S's example, suppose I believe that my keys are in my jacket because that is where I remember putting them. I look for them in my jacket (positive testing) and not in my purse (negative testing). Hence, as opposed to the interpretation of positive testing as a tendency to confirm (confirmation bias) (Cowley & Byrne Reference Cowley, Byrne, Bara, Barsalou and Buchiarelli2005; Wason Reference Wason1960), positive testing may occur with no other goal than finding out the truth, as the real-life example suggests. According to ATR, positive testing is a default heuristic that involves no reasoning proper.
However, as M&S further argue, reasoning can be triggered in hypothesis testing situations if an argumentative context is provided. Moreover, in such a context, reasoning is directed at falsification, according to ATR: Falsification is accessible provided that the situation encourages participants to argue against a hypothesis that is not their own, as M&S propose This logic reveals the old misinterpretation that test choice (positive or negative) is tuned at the emergence of favorable test result. In fact, putting one's idea to a (either positive or negative) test assumes the objective to find out the truth and is therefore at odds with testing in order to save our idea from confirmations or disconfirmations. Poletiek (Reference Poletiek1996) showed that participants in a hypothesis testing experiment are aware of the incompatibility of finding out the truth and coming up with confirming test outcomes only. Participants felt that they were looking for information about the validity of a hypothesis, and that they could not control the test outcomes by choosing a test strategy. It was only their expectation of the outcome that changed when they tested a hypothesis they did not believe to be true (falsification was expected). When testing their best guess, they expected a confirming outcome.
Besides determining the truth, many considerations may affect how we reason and whether we reason. These considerations can be modeled as a cost-benefit analysis of making the right inference. In Wason's (Reference Wason1960) task, students participating in the experiment might not care much about making reasoning errors. The key seeker might perform a negative test (looking for the keys in some place other than the jacket) because the costs of a positive one are too high: The jacket was left behind in a friend's house and the effort too great to walk back to check the jacket. Alternatively, we might be in a hurry and have time to check just one location, enhancing the benefits of good reasoning.
To predict reasoning behavior, we need a model with utilities of reasoning outcomes. A suitable tool is signal-detection theory. Motives can be expressed in the utilities of inference errors and in reasoning behavior predicted on the basis of the risk we are prepared to take with regard to particular erroneous inferences (Poletiek & Berndsen Reference Poletiek and Berndsen2000). For example, as shown in Table 1, a pure epistemic motive would be modeled with (A) low utilities of making any false inferences. A pure argumentative motive would be expressed in (B) a very high willingness to make the false inference that our favorite hypothesis is true; and (C) reasoning with both motives (searching for a valid inference within both practical or argumentative constraints) is reflected with some in-between utilities with regard to making a false or a correct inference about our favorite hypothesis.
Utilities of reasoning outcomes about H (the reasoner's subjective belief), assuming an epistemic motive (A), an argumentative motive (B), and both types of motives (C).
In this manner, reasoning in a variety of contexts and with a variety of goals can be modeled, offering an elegant alternative to the paradox of the ATR that we start up a reasoning trajectory about a prior belief if and only if the end of the route leads us to inferring that belief again.