A Discussion on The Scientific Method as “Bayesian”

Jan 24, 2020, 9:01 AM


This is a good start. Now I know the paradigm your working from and its limits. We have a long way to go. But smart folk will get there.

My job is to (a) get you to operationalize bayesian learning instead of using it as a ‘magic word’. And once we are there, (b) to forgo the discipline’s self congratulatory presumption of its innocence, and demarcate between science, pseudoscience, and deceit. (c) and then to articulate the means by which we warranty against pseudoscience and deceit. The rather (humorous) framing is the presumption of innocence and positive intention or positive bias of the fields. And while the grammars (rules of continuous recursive disambiguation) of the physical world are relatively simple, as we move from physical to biological, to economic, to legal, to cognitive and rational the problem increases in complexity, as well as opportunity for malfeasance.

—“Bayesian updating is done by individuals. Market competition takes place in groups of individuals. So these are disjoint processes from the off, and there is no contradiction to both of them operating to some extent, and both of them being necessary.”—


(a)Personal learning, and personal survival of hypotheses.
(b)And the market competition in personal learning, and the market survival of hypotheses under frictions of the investments and malinvestments of the individuals in advancing and defending their propositions.

Now, what is the difference between “bayesian updating” and ‘learning’? Or is this terminology an attempt (pseudoscientific) to use an analogy as an appeal to scientific authority? For example, in computer science the value of bayesian systems is that they are better at ‘accounting for’ a large number of very small changes.

So just as double entry accounting increased our ability to perceive and measure otherwise incommensurable aggregates to measure profit and loss, Bayesian learning is yet another improvement upon accounting that does not require we limit ourselves to the commensurability provided by money and prices. We can use any categories (identities) that we want. In the publication of scientific papers (which doesn’t have a good record) and books (which do have a good record) we are clearly within the limits of human introspection. But we rapidly lose the possibility of introspection in neural networks.

So you just mean ‘learning’, where learning is always and everywhere produced by the process of exposure to information > free association > hypothesis > survival (or return) > weak theory > survival(or return) > strong theory or ‘law’ (or return) > revision(falsification)

—“Bayesian updating is necessary so individuals are able to change their plausibilities consistently based on new evidence, without which knowledge creation would be impossible.”—

So we account for change both positively reinforcing and negatively falsifying.

—” secondary effects of the “market competition” “—
—” nothing to do with the evidence”—

And in case of the replication crisis in the social sciences; or in the case of the entire field of economics despite its financial literacy producing income statement outputs without the context of balance sheet constants; or in the case of physics losing generations to Bohr’s and Cantor’s (somewhat Einstein) re-primitivism of mathematics in physics. And of the fields only physical science grasping and integrating the Operationalist revolution that failed to take hold in every other science during the twentieth.

Mathematical sciences have produced harmful externalities. The physical sciences are not innocent but other than physics produced the fewest externalities. The psychological and social sciences are arguably harmful if not reversing the gains made by the physical sciences. And economics is all but a desert of failure only another generation of mathematics will solve.

So. The foundation of my criticism is that the presumption of innocence in the sciences is a falsehood. And that the harm by externality has cost us at least a hundred million dead. And the harm of sociology, psychology, economics and mathematics together have created a crisis as vast as that of the late empire. Conversely that the principle problem with the physical sciences is underinvestment given the increasing cost of experiments given the increasing differences between inquiry and human scale of sense perception.

So, yes, scientists learn by attempting to discover that which is not known by use of logical and physical instrumentation in tests of trial and error. The question is, when you make your claims what due diligence do you use? When taking and publishing (testifying to) measurements you only report facts. When publishing theories what are you testifying to? And what demarcates what is testifiable (publishable) and not? And as complexity increases (as we move from continuous physical relations where state cannot be stored and no ‘choice’ can be made, to continuous behavioral relations where state can be stored in memory and choice can be made) what criteria do you use to determine whether your statement (promise) is not the result of ignorance, error, bias, wishful thinking, loading and framing, suggestion and obscurantism, the fictionalisms, or outright denial and deceit?

Consistency, completeness, parsimony:

Meaning: Realism, Naturalism, Operationalism, Parsimony, and categorical, logical, empirical, operational consistency, within stated limits – and where psychology is involved, rational choice by incentive to acquire, maintain, or prevent loss, within the limits of knowledge and bounded rationality.

This is the other side – the via-negativa defense against the immoral – of the scientific coin so to speak, where exploration – the via-positiva – is the moral.

So what is universal in the scientific method? Well, that we learn through free association and incremental survival (darwinian falsification) by continuous exposure to competition, resulting in the positive reinforcement of explanatory power, and the negative reinforcement of failures of explanatory power (or deceit, or fraud, or incompetence), at the individual, interpersonal, and market levels of complexity, where our careers are punished for false promise of truthful testimony that we contribute to that market, and our careers are rewarded (somewhat) for promises of truthful testimony that we contribute to the market, that provide either new opportunity for investigation, new explanatory power, or new falsification of priors.

So. the lesson?

COMPLETENESS – FULL ACCOUNTING provides a very different understanding of science than does cherry picking and virtue signaling.


Leave a Reply