Back to Bloodletting!
Can statistical models — “big data” — tell you what we should do about COVID-19? Or about climate trends?
Can statistics possibly justify shutting down the economy to fight a virus or change the climate? Or would bloodletting, as performed and tested by surgeons for millennia on their patients, be a far better course of action?
Artificial intelligence (AI) and “machine learning” — based on “big data” modeling — are the foundations of a new generation of information tools that are being used and abused widely by policymakers.
Jeopardizing their usefulness are ideas of their capabilities no more realistic than 18th century illusions about the effects of leeches on disease.
These days, we hear constantly of the need to respect what is called science, whether of climate change or of coronavirus effects. But when the so-called science is scrutinized, it nearly always turns out to mean more leeches, a statistical model.
A Telic AI Explanation for the Ages
In previous prophecies, you have met statistical philosopher William Briggs, now blogging incandescently on the COVID viral new mania. Read his blog by clicking here.
In his book, Uncertainty, he proves the inexorable limits of big data. Statistics simply cannot prove cause. If you don’t know the cause, you cannot differentiate noise from signal in the statistics.
The original authority on cause was the ancient Greek philosopher Aristotle, who divided causality into formal, material, efficient, and telic (purpose-based) explanations.
For an example, take a vase. The formal cause is its shape, but the shape cannot produce the vase without appropriate material such as clay or silica (glass). The material and the shape together are immediate causes of the vase. But they are impotent to produce it without an efficient cause, the hands of a potter or the breath of the glass blower — and an external heat source such as a kiln.
However, the final or telic cause is the mind of the potter himself, who conceives the purpose of making a vase rather than an ashtray or a bowl or a pipe. Without reading his mind, which a machine can never do, your understanding of the vase is trivial and almost irrelevant.
Suitably programmed and structured by human minds, AI can seem to reveal formal, material, and efficient causes. But actual purposes are beyond it. Thus, AI is intrinsically a rearview mirror. It can predict the future only by assuming continuity with the past and disregarding the power of human free will to reverse a trend.
For example, to address a more topical set of causes, we might return to COVID-19, the disease that is deranging the world in 2020. The form is the ostensible aspect — people getting sick with the flu; the material is the invisible substance involved — a novel coronavirus; the efficient cause is normally some kind of mechanism — the spread of the virus from raw bat markets in Wuhan to the world through human contact; the telos is the rub — the purpose, the essence of viruses. The telos reveals not what, but why an event happened.
Where does artificial intelligence fit in this scheme? Using a structured and filtered assemblage of big data from hospitals, a machine learning algorithm could resolve on a pattern of symptoms that represents the form of the disease. Through masses of carefully structured testing data, AI could identify a particular material pathogen that is present in all these cases. By poring through records from cellphones, tele-thermometers and other tracking devices or hospital reports, AI might show that personal interactions with infected parties account efficiently for the spread of the disease.
Aha! a policy-maker might declare. Stop all personal interactions and the disease will die!
The AI algorithms, appropriately deployed, arguably could contribute significantly to a description of “what happened.” Why the disease occurred and how it can be appropriately remedied, however, presents a conundrum beyond the data.
Knowledge of cause is above and beyond and deeper than knowledge of “what happens.” It reflects the grasping of essences through induction of universals beyond the material world. It entails grasping the biology and informatics of viruses, the way they function and communicate, how their RNA codes, peptides, and proteases can be generated through Polymerase Chain Reaction (PCR) machines. It requires knowledge of immunology and herd immunity and vaccines and hygiene and mutation of DNA and RNA. It requires a transcendent multidisciplinary understanding of the world utterly absent from AI.
The Pursuit of Deep Learning
Induction is the cognitive mechanism that produces the “leap of insight” necessary for all understanding. It moves from the finite toward the infinite, from particularities collected by the senses or by sensors, to unobservable “insensible” generalities or universals.
Induction stems from the consciousness of a human “interpretant” of such specific but indeterminate features as knowledge of a thing’s essence. It mediates between symbols in an algorithm and meanings in a mind — maps and territories. Only rational creatures can accomplish it.
As a form of reason by analogy and intuition, induction is superior to mere deduction, a mechanical and algorithmic process that can be performed by a computer. AI is deductive and mechanical.
Without understanding causes and essences through induction and experience, a model can be no better than deterministic. It can tell what happens but not why. Without grasping why — the essences and purposes involved —an algorithmic model cannot reliably predict anything or prescribe any relevant remedies.
A model is analogical. Many models can explain a particular set of facts, but there can be only one true understanding of cause. As Briggs remarks, “If we do grasp the cause of something we need no model or experiment. After all, we have the cause!”
No full, universal efficient deterministic model of nature exists: no “theory of everything.” At the time of Newton, the alchemists thought they were close to grasping a final model in chemistry. Quantum mechanics declares that such a model is impossible. But that does not stop local models that adequately explain everyday physical phenomena and their efficient causes.
However, as Briggs says, the pursuit of “deep learning” by machines leads to a search for a new alchemy, a “Philosopher’s Stone” that can turn lead into gold. Briggs dubs it a “Statistician’s Stone,” that can yield caused revelations from “big data.” Computers churning the oceans of language or literature, so it is believed, can ultimately distill a “natural language” interpretation indistinguishable from a human understanding.
This belief implicitly denies all the necessary decisions about what data is relevant and is included in the database. These decisions all come by inductive human judgement, as does the “deep learning” algorithm itself.
Briggs’ target is the issue of the efficient cause of disease. Long before the coronavirus madness, he explored the case of sub-2.5 micron particles in the atmosphere. Found to cause cancer by the Environmental Protection Agency on the basis of a statistical model, emission of the particles was widely banned and restricted.
The data showed up to a 200% increase in the incidence of cancer in areas with high levels of the sub-2.5 micron pollution compared to areas with low levels. Wow! It causes cancer, said the EPA.
Yet the totals involved were infinitesimal compared the exposed populations. A 200% increase in the incidence of cancer represented a rise from one case to two cases in a population of hundreds of thousands. It was noise not signal.
Briggs shows that the EPA perpetrated grave malpractice in its attribution of causality. Either the particles “cause” cancer in everyone sufficiently exposed or they “cause” it in no one. Cause is not mere correlation.
Today, governments around the world are closing down economies on the basis of statistical trends in deaths ascribed to the virus. They might do better by conducting bloodletting or bleeding campaigns as physicians commonly did for some 2,000 years until the early 1800’s.
There is no difference at all between bloodletting patients to fight a virus and bleeding the economy, except that the bloodletting remedy was more rational, and far less outrageously destructive to the patients.
Editor, Gilder’s Daily Prophecy