استدلال غلط بیزی : انگیزش های داخلی و تمایز اعتقادات مذهبی از اعتقادات دیگر

This paper advances the two-axis of information hypothesis. One axis concerns whether one's belief is about the environment as opposed to one's own action (the self). The other axis of information concerns whether the information is about content as opposed to context. Given the two axes, we have four kinds of beliefs: convictions, conceptions, perception, and confidence. The Bayesian fallacy is the failure to observe the differences among the four kinds of beliefs. For instance, convictions are about internal motivations, such as “I can climb this mountain”, which cannot be updated via Bayes's rule as the case with the other three beliefs.

James Heckman has stumbled onto a puzzle. While the expected return on investment in education has risen in the US in the last few decades, people have reduced their investment in education (Heckman, 2007, Heckman and Krueger, 2003 and Heckman and Rubinstein, 2001). To explain this phenomenon, Heckman has proposed that investment in education, “cognitive skills,” is not only a function of expected returns, but also a function of “non-cognitive” skills, where non-cognitive skills consist of perseverance, tenacity, morale, and internal motivations. Heckman explains the decline in education investment by arguing that internal motivations to invest in education must have declined at a steeper rate than the rise in expected returns (which, in turn, can be regarded as the external motivations or stimuli). He has advocated a policy of focusing on early child development to stave off any further decline in morale, given that such non-cognitive skills are nurtured mostly in the first seven years of life. The term “non-cognitive,” however, is problematic. First, in the cognitive psychology literature (Garnham, 1994, p. 167; Lycan, 1999, Reber, 2003 and Still and Costall, 1991), “non-cognition” denotes the emotions, while “cognition” denotes the faculty of knowing, or mental processes undertaken to solve a problem, i.e., to decide on the proper response to a stimulus. But all beliefs, including morale and internal motivations, involve both cognitive and non-cognitive elements. Second, in the same literature, the two terms are opposed, invoking the famous Cartesian dichotomy between the emotions and reason (Greene, 2005). The emotion/reason opposition begs, at a deeper level, bigger questions. Although it is outside the scope of this paper, one such question is how could reason be independent of the emotions, when both are related to the striving of the organism to enhance its wellbeing or fitness? (Khalil, 2010a). Instead of the term “non-cognitive,” this paper employs the term “convictions” to denote beliefs concerning morale or internal motivations. Convictions are beliefs about one's ability with respect to goal commitment such as running in a race, starting an enterprise, managing a business, or persevering through a storm. Convictions can be defined as self-beliefs in relation to the capability of the organism to produce: “I can paint this wall,” “I can finish this project,” or “I rely on God's support in opening my business”. In this proposed usage, convictions include religious beliefs insofar as they sustain the agent's inner motivation or morale. This paper draws a radical distinction between convictions and three other beliefs, called here “confidence,” “perception”, and “conception”. Confidence is also a self-belief—but it is about the assertion of rights of the self vis-à-vis the rights of others and, in the case of temptations, vis-à-vis the rights of the future self: “I will not overeat”. Perception, on the other hand, is a belief about the environment that is empirical: “the storm will incur about $2 billion in damages”. Conception is also an environment-related belief that involves empirical date—but it also involves framing, meaning, or context, which is non-empirical: “the storm will incur about $2 billion in an area that has $15 trillion worth of real estate”. What is the payoff of the proposed four-way distinction of beliefs? The main payoff is the ability to differentiate convictions from the other kinds of beliefs. Convictions include the set of religious beliefs related to one's internal motivation and ability to undertake tasks. Beliefs about internal motivation, as the case with religious convictions, cannot be empirically updated—unlike the case of scientific beliefs such as perceptions that can be empirically updated. The fact that agents cannot use Bayes's rule to update a class of beliefs, viz., convictions, does not mean they are irrational. This payoff should shed light on the findings of behavioral economics: many of the supposed behavioral anomalies, i.e., behavior that seems to deviate from the predictions of standard rationality theory, are not anomalies because they are based on beliefs (convictions) that cannot be updated via Bayes's rule to start with. The four-way distinction of beliefs is grounded on a hypothesis called here the “two-axis information hypothesis”. One axis identifies whether the object of the belief is the agent or his or her environment; the other axis identifies whether the information is about content or context. Using the two axes, convictions concern the self and are about context; while conceptions concern the environment and are about content. And it is shown, corollary, why convictions are non-correctable with regard to truth and, further, are non-warrantable because the object of the belief is not a fixed reality; while conceptions are likewise non-correctable with regard to truth but are warrantable because the object of the belief is a fixed reality. This will become clearer below. But to define the Bayesian fallacy, it is sufficient to clarify that convictions and conceptions are non-correctable in the sense that they cannot be judged, via Bayes's rule, as being true or false because both involve meaning (context), which is non-empirical. Thus, when one asserts “I can paint” or when one conjures the help of God to finish a project, one creates an aspiration that acts as a context and is, hence, non-empirical. On the other hand, warrantability is a weaker criterion of empirical corroboration than correctability. Convictions are non-warrantable because they involve the development of potentiality or capability: as one pursues a goal, the supposed evidence that supports one's belief in achieving the goal, the perceived self-ability, develops, which means that one cannot use self-ability as such evidence. The same cannot be said about conceptions because the evidence, i.e., the environment, is a fixed reality—it does not develop. In this light, we can define the “Bayesian fallacy”: The Bayesian fallacy: One commits the Bayesian fallacy when one treats convictions, which cannot be the subject of Bayes's rule, as no different from other beliefs which are, although in different ways, the subject of Bayes's rule.1 This paper argues that there are, in fact, two degrees of the Bayesian fallacy. To do this, Section 1 explicates the proposed two-axis information hypothesis. Section 2 stresses the importance of convictions. Sections 3, 4, 4.1, 4.2, 4.3 and 5 then advance the core thesis: convictions are non-Bayesian beliefs because, following the two-axis information hypothesis, they involve meaning as well as development.

Not all beliefs are born equal. The proposed four-way distinction of beliefs would, foremost, help us to avoid the Bayesian fallacy. One commits the Bayesian fallacy by treating beliefs as if they are all equally updatable via Bayes's rule. The four-way distinction allows us to identify one kind of beliefs, “convictions,” which proves to be of particular importance. Convictions are about internal motivations, which differ from external motivations or what economists call “incentives” and psychologists call “stimuli”. It is important to understand convictions because they are at the core of entrepreneurship, innovation, and creativity. They are imperative for understanding the vitality of organizations and economic development. As we have seen, convictions are beliefs about one's ability to produce: “I can be a successful hairdresser”, or “I can climb this mountain”. Convictions are at the origin of perseverance, tenacity, or morale. A conviction concerns an object, namely, ability, that cannot be defined precisely even in a world of perfect information. Ability is a potentiality in the sense that it undergoes development and evolution as a result of action. This is not the case with the other three kinds of beliefs. The other three kinds of beliefs – involved in different ways in the formulation of external motivations – are called “confidence,” “perception,” and “conception”. Confidence is also about the self—but concerns the will to withstand temptations: “I will resist cheating on the exam”. Perception concerns beliefs about the environment that are fully based on empirical evidence: “the interest rate rose last year”. Conception is also about the environment—but involves framing, context, or meaning: “the interest rate rose too high last year”. Convictions can neither be corrected, nor made warranted via Bayes's rule. This is because of two reasons. First, convictions are non-correctable because the object, self-ability, involves context, which has no empirical basis. One chooses the context to make sense of one's ability, and it is this context which is at the origin of self-expectation, aspiration, or desire. Secondly, convictions are non-warrantable because the object, self-ability, evolves when one acts to test it. To understand how convictions (internal motivations) are related to the other three beliefs (external motivations), we need to explicate how humans process information. This paper proposes a “two-axis information” hypothesis to categorize information: type of information versus domain of information. When considering the type of information axis, one asks whether the information is about empirical content, such as rise of interest rates, or if it is about the context, such as how to interpret such a rise. Content information forms evidential beliefs, while context information occasions non-evidential beliefs. It is possible to use Bayes's rule to correct evidential beliefs, but one cannot use information to correct non-evidential beliefs. When considering the domain of information axis, one instead asks whether the information is about the self, or if it is about the environment. Self-related information gives rise to internal beliefs, while the environment-related information occasions external beliefs. Thus, when we use these two axes, we end up with four kinds of beliefs: 1. Confidence: internal-and-evidential belief. 2. Conviction: internal-and-nonevidential belief. 3. Perception: external-and-evidential belief. 4. Conception: external-and-nonevidential belief. If evidential, the beliefs (confidence and perception) can be subjected to Bayesian modification, and, hence, should be labeled “Bayesian beliefs”. However, if non-evidential, the beliefs (conviction and conception) cannot, although with one qualification, be subjected to Bayesian modification. If the belief is about the environment (conception), the information can make the belief more (or less) warranted, which makes the belief a “quasi-Bayesian belief”. If the belief is about the self (conviction), however, the information cannot even make the belief more (or less) warranted and, hence, makes the belief a “non-Bayesian belief”. The paper calls the modeling of conviction (non-Bayesian belief) à la confidence or à la perception (Bayesian belief) the “first-degree Bayesian fallacy”. It is first-degree because it amounts to treating conviction no differently from a scientific proposition about mind-independent reality. In contrast, the modeling of conviction à la conception (quasi-Bayesian belief) is the “second-degree Bayesian fallacy”. It is second-degree because it at least acknowledges that the belief involves context and, hence, cannot be corrected as true or false. But it is still a fallacy because it treats the object of conviction as being independent of the belief, as is the case with conceptions. With these fallacies in mind, we need to examine three major views of beliefs/actions in the social sciences: standard rationality theory, normative theory (which is dominant in sociology), and Herbert Simon's procedural rationality theory. But this task is better left for another arena (Khalil, 2010b).