“Paint with an airbrush!”
“Yes but all that paint-scatter?”
“Mask the edges!”
An un-blinded Algorithm
We employ methodologies with limited views and with one broad stroke of the brush to wipeout any hidden flaws. The truth seeks but under the colossal weight of dried ink it remains muffled and hidden, like a masterpiece painting it lies submerged under the spray of a modern airbrushed abstract. Eventually the leveling scythe of assumptions and probability science shows its obverse edge.
What if… we had a blind algorithm that went about sniffing through large pieces of data, snaking its way and exploiting each path, determining the cost of each path and all successor events that preceded and followed in each of those paths to arrive at a goal that no one pre-selected, in other words, blind vectoring into a sea of options. The blind algorithm would blind the researcher and would dive down into the sea of amorphous data. The Artificial Intelligence would guide it through the maze just as a modern day Robotic vacuum cleaner dodges and bumps against furniture in a room. This would prevent the ebb and fracture of the scientific culture currently in progress.
O God, I could be bounded in a nutshell, and count myself a
king of infinite space—were it not that I have bad dreams.
Which dreams indeed are ambition, for the very
substance of the ambitious is merely the shadow of a dream.
What would happen?
We might get results that would surprise us all. It might eliminate certain prejudices of thought. It might conflate some, inflate others and deflate concretely held paradigms. It might through aggregation of such information form a shape more akin to reality then prophesied through the false gods of statistical fiat.
Would humanity consider those potential alternatives via a more clear vision rather than the dark lenses of doubt? I am afraid so. The well entrenched busy-bodies would still want to extract their pound of flesh, but their premise would be mortally weakened. Their narcissistic grist would be reduced and their mills dried would seek redemption. To reduce the risk of such posturing or "gaming" the system with the algorithm, one might consider sending a second algorithm-bot which could constantly modify the prime algorithm's behavior when under threat from outside manipulations.
Humans love to seek the truth; beneath the stone, the dark side of the moon, the distance from the moon and the elliptical orbit of the planets. They desire real evidence but often fail in their goals mired in the gutter of soft, pliable and manipulated evidence. Medical literature is bugged with a significant amount of pseudo-science. This form of pseudoscience comes from selected variable sources that attempt to create a reality that is nonexistent. These useless dribblings on our zeitgeist mar the tattered tapestry of our being even further.
A blind algorithm run amuck in a Data-warehouse would produce some sparks of discontent as it bumps against dead ends, ricochets through the “bump and grind” and the stochastic information finds the critical path to a heretofore unrealized solution. This ultimate critical path would use efficiency, expediency and find pecuniary limits to realize the least resistant and best pathway. Or, then it might just prove an existing paradigm. Such a confirmation would equally have immense credibility and usage. The flowering colorful proofs from random analytics would shed more light and do more for humanity than the gunmetal gray of a postured regulatory missive!
Imagine if a large pool of patient data demographics that included age, sex, race etc. and in addition family history, dietary history, exercise habits, thrill seeking or a quiet lifestyle, use of any prescribed and over the counter medication and myriad of other information is floating around in the dark unreflective waters. The blind algorithm would tease outputs that we may not have contemplated. Instead of selected data, univariate analytics or even controlled multivariate analytics, the randomness of “chew and spit” of the esoteric data for the “right taste” could/would reveal outputs of immeasurable weights and benefit…or not. It certainly would be worth the endeavor. variables would then indeed be variables, not chosen and picked like fruit from a tree, the correlations would indeed be obvious and used within the algorithm and the regressions would point to the future with a focused light. The p-values would have meaning and the Confidence Intervals would be tightly bound. Even Fat tails would be obvious and not hidden under the weight of mathematical jujitsu to make “nice.”
Studies of coffee, chocolate and red wine proving goodness for the heart created through selected dataset by those industries to prove the primary aim of the researcher would indeed be a thing of the past. As John Ioannidis points out currently 54% of the studies are not validated or reproducible lends further credence to blinding the algorithm for real “Evidence’ in Evidence Based Medicine!
In the end, the numbers themselves do not deny us our humanity but the calculated decision to stop being human but the complete devotion to reliance on an obtuse and self-motivated junk science.
Let us therefore use the combinatorial force of computer science for the purpose it was designed. Let us start looking for the keys to reality in the dark and not only under the lamplights. Let us use Big Data for all its worth!
We all have dreams…