Clark offers a review of a view of the brain where the brain processes input information in a way that confirms its priors or its predictions. This does not mean that the brain creates its own reality. The brain, rather, processes input data, but it does so in light of its own priors. The brain is a bidirectional hierarchical structure. While the top layers generate priors, the lower layers process input data. The brain amounts to the dynamics of image-making, where the top-down process generates unified images, while the bottom-up process, which takes data, corrects the images.
Such an iterative cognitive process is not simple. The top-layer generated priors greatly determine the assimilated inputs. But the input data are not fully manipulated by the priors. As such, it is best to characterize the brain as a medium that tries to balance between two competing needs: First, the brain needs to generate a unified, that is, meaningful, image of the real world. The top layers, which generate the priors or the predictions, function to fulfill the need for unity. Second, the brain needs to accommodate raw input data to stay as truthful as possible to the given real world. If the brain performs only the first function, that is, preserving the unity of the image, the brain would generate images that, although unified, are disconnected from reality. On the other hand, if the brain performs only the second function, that is, preserving the details of the world, the brain would generate images that, although detailed, are tremendously messy and meaningless.
As a result of trying to meet these two competing needs, the images that cognitive processes generate are theory-laden. This has long been understood by the emerging new philosophy of science, most epitomized by the contribution of Thomas Kuhn, and can even be traced to Immanuel Kant. This is not the place to review the history of philosophy of science, characterized ultimately as a conflict between rationalism (demanding unity of image) and empiricism (demanding detailed images) (see Khalil Reference Khalil1989). What is germane here is that Clark fails to note two different kinds of theory-laden cognitive processes: the first, which can be called “perception-laden” processes, where one's theory can be ultimately corrected by sensory input; the second, which can be called “conception-laden” processes, where one's theory cannot be ultimately corrected by sensory input.
Perception-laden beliefs, for example, let one predict stormy weather or that the Earth is flat. In light of sensory input, and using Bayes' rule, one may adjust such a prediction and reach the conclusion that the weather will be stable and the Earth is round. Many people may not adjust quickly and insist on “explaining away” the data to justify their priors. But such manipulation can be delineated from the normal course of belief adjustment. When perception-laden processes are at issue, priors must ultimately adjust to correspond to the mounting evidence. The legal system, and everyday science, cannot function without the adherence to the possibility of belief-free grounds that can allow sensory data, in the final analysis, to dominate top-down priors.
Conception-laden beliefs, for example, let one view a picture such as the famous Rubin Vase, where the brain switches between perceiving the vase and perceiving the two profiles. The image depends on what the brain judges to be the background. If the background is judged to be white, the brain sees the two profiles. If the background is judged to be black, the brain sees the vase. No amount of data can compel the top level hierarchy of the brain to abandon its prior. The prior here cannot be confirmed or refuted by evidence because it is not based on evidence as with perception-laden processes. The choice of background, the basis of conception, is similar to the choice of a benchmark, where one can judge a glass to be either half-full or half-empty. Likewise, one judges one's income as satisfactory or non-satisfactory depending on one's benchmark. Happiness seems to depend, at least partially, on the choice of an arbitrary income as the benchmark income.
The conflation of the perception- and conception-laden processes leads to the commitment of a Bayesian fallacy. The fallacy arises from the supposition that all beliefs are perception-laden and, hence, can be corrected by further empirical investigation (Khalil Reference Khalil2010). It is imperative to distinguish conceptions from perceptions. Aside from allowing us to understand happiness, the distinction sheds light on two kinds of stubbornness: intransigence, related to perception-laden beliefs, and dogmatism, related to conception-laden beliefs. Belief in a flat Earth and in conspiracy theories illustrates intransigence. In contrast, to insist on a background, despite the rising evidence to the contrary, illustrates dogmatism. To use the Rubin Vase example, if a person chooses the black as the background and, hence, the image is the vase, but continues to choose the black despite contrary added evidence – such as added eyes and moustache – the person would be dogmatic. While the dogmatic belief cannot be judged as true or false, it can be judged as warranted or unwarranted given the details of the profiles. The choice of background, to remind ourselves, is non-empirical and, hence, cannot be characterized as true or false.
Clark offers a review of a view of the brain where the brain processes input information in a way that confirms its priors or its predictions. This does not mean that the brain creates its own reality. The brain, rather, processes input data, but it does so in light of its own priors. The brain is a bidirectional hierarchical structure. While the top layers generate priors, the lower layers process input data. The brain amounts to the dynamics of image-making, where the top-down process generates unified images, while the bottom-up process, which takes data, corrects the images.
Such an iterative cognitive process is not simple. The top-layer generated priors greatly determine the assimilated inputs. But the input data are not fully manipulated by the priors. As such, it is best to characterize the brain as a medium that tries to balance between two competing needs: First, the brain needs to generate a unified, that is, meaningful, image of the real world. The top layers, which generate the priors or the predictions, function to fulfill the need for unity. Second, the brain needs to accommodate raw input data to stay as truthful as possible to the given real world. If the brain performs only the first function, that is, preserving the unity of the image, the brain would generate images that, although unified, are disconnected from reality. On the other hand, if the brain performs only the second function, that is, preserving the details of the world, the brain would generate images that, although detailed, are tremendously messy and meaningless.
As a result of trying to meet these two competing needs, the images that cognitive processes generate are theory-laden. This has long been understood by the emerging new philosophy of science, most epitomized by the contribution of Thomas Kuhn, and can even be traced to Immanuel Kant. This is not the place to review the history of philosophy of science, characterized ultimately as a conflict between rationalism (demanding unity of image) and empiricism (demanding detailed images) (see Khalil Reference Khalil1989). What is germane here is that Clark fails to note two different kinds of theory-laden cognitive processes: the first, which can be called “perception-laden” processes, where one's theory can be ultimately corrected by sensory input; the second, which can be called “conception-laden” processes, where one's theory cannot be ultimately corrected by sensory input.
Perception-laden beliefs, for example, let one predict stormy weather or that the Earth is flat. In light of sensory input, and using Bayes' rule, one may adjust such a prediction and reach the conclusion that the weather will be stable and the Earth is round. Many people may not adjust quickly and insist on “explaining away” the data to justify their priors. But such manipulation can be delineated from the normal course of belief adjustment. When perception-laden processes are at issue, priors must ultimately adjust to correspond to the mounting evidence. The legal system, and everyday science, cannot function without the adherence to the possibility of belief-free grounds that can allow sensory data, in the final analysis, to dominate top-down priors.
Conception-laden beliefs, for example, let one view a picture such as the famous Rubin Vase, where the brain switches between perceiving the vase and perceiving the two profiles. The image depends on what the brain judges to be the background. If the background is judged to be white, the brain sees the two profiles. If the background is judged to be black, the brain sees the vase. No amount of data can compel the top level hierarchy of the brain to abandon its prior. The prior here cannot be confirmed or refuted by evidence because it is not based on evidence as with perception-laden processes. The choice of background, the basis of conception, is similar to the choice of a benchmark, where one can judge a glass to be either half-full or half-empty. Likewise, one judges one's income as satisfactory or non-satisfactory depending on one's benchmark. Happiness seems to depend, at least partially, on the choice of an arbitrary income as the benchmark income.
The conflation of the perception- and conception-laden processes leads to the commitment of a Bayesian fallacy. The fallacy arises from the supposition that all beliefs are perception-laden and, hence, can be corrected by further empirical investigation (Khalil Reference Khalil2010). It is imperative to distinguish conceptions from perceptions. Aside from allowing us to understand happiness, the distinction sheds light on two kinds of stubbornness: intransigence, related to perception-laden beliefs, and dogmatism, related to conception-laden beliefs. Belief in a flat Earth and in conspiracy theories illustrates intransigence. In contrast, to insist on a background, despite the rising evidence to the contrary, illustrates dogmatism. To use the Rubin Vase example, if a person chooses the black as the background and, hence, the image is the vase, but continues to choose the black despite contrary added evidence – such as added eyes and moustache – the person would be dogmatic. While the dogmatic belief cannot be judged as true or false, it can be judged as warranted or unwarranted given the details of the profiles. The choice of background, to remind ourselves, is non-empirical and, hence, cannot be characterized as true or false.