Ayurveda to Artificial Intelligence: Reassessing the Costs of Unknowability

Artificial intelligence (AI) is all the rage. Or at least a version of it that’s popularly referred to as machine learning (ML). ML is nothing more than the use of computing power to infer patterns from voluminous data, otherwise too difficult to discern by introspection. It’s used today to interpret radiographs, make sense of social media data, interpret voting behavior, recognize faces, and so on. The catch with ML is that it’s often not clear how it arrives at the patterns that are thus identified.  

This unknowability is causing some consternation in legal systems. In US common law for example, judges must explain their reasoning. What facts, precedents, and processes of reasoning lead them to the conclusion that they reach? It should all be there in writing for examination and scrutiny. Such a system makes it hard to rely on AI-driven black-box correlations that try to nudge the legal process one way or another without understanding, and therefore being able to articulate, the underlying reasoning. Should the philosopher’s notion of an epistemic warrant – the notion that precedent, memory, introspection and a priori belief justify a view – suffice more often than it normally does in the legal process?

As an academic social scientist, I too am obsessed with asking ‘why’ something is the way it is? In another avatar though, as an entrepreneur, I’m more preoccupied with finding bits of knowhow that can help me advance whatever venture I’m working on at the moment, without mulling over the issue of exactly ‘why it works.’ So, it prompts me to ask, how important is it to fully understand something before one should actively use that information?  

Modern medicine is full of ‘remedies’ that work without our knowing why. We use these readily, pending a better understanding at some future date.   A review in 2011 of a decade of drug approvals reported in the respected journal Science suggested that less than a fourth of drugs that worked in new ways emanated from a deep understanding of the underlying disease mechanism.

To illustrate, consider diabetes, scourge of the developing world, a disease that originates with problems with the insulin-producing glands of the pancreas.  The treatment is a sense-and-respond, symptomatic one built around suppressing levels of sugar in the blood either by cutting down its availability from the gut, addressing its excretion (loss) in the kidneys or increasing availability of insulin in the blood to mop it up.  Our natural sugar control operates very much like an automobile carburetor, precisely matching fuel supply to the exact demand of internal combustion. Somehow this is also connected to the über-controller organ, the liver, elegantly storing sugar for immediate release and apportioning release of sugar and insulin together, precisely matched to meet metabolic needs while balancing its level in the blood. To do more than sense-and-respond, we need to understand this elaborate symphony of organs much more than we currently do. But we can’t afford to stop improving our existing imperfect treatment – through more accurate and convenient devices for example – pending this fuller understanding.

And what about the medicine of yesteryear? I’m thinking about childhood remedies that my grandmother might have sworn by, and that might still reassure me, the equivalent in ancient cultures of chicken soup for the soul.   How do we know that these work? Have they been subjected to the gold standard of randomized control trials?

In India, widespread reliance on informal remedies coexists with a pithy caveat emptor admonition in the form of a saying or an idiom, a muhavara, that says ‘Neem hakeem khatra jaan’ नीम हक़ीम ख़तरा-ए-जान. Beware of the informally (by today’s standards anyway) trained traditional remedies doctor dispensing remedies from under a ‘neem’ tree. The idiom points to the danger of relying on someone with such half-baked knowledge of medicine. Of course, many ancient cultures rely on traditionally passed-down homilies for sustenance and succor.

But, wait! Let’s not throw the baby out with the bathwater. Remember Tu Youyou, the Chinese doctor who won the Noble prize for medicine (physiology) in 2015? Her contribution was to identify artemisinin which has cured millions of malaria. The inspiration for this discovery came from ancient Chinese texts which Tu Youyou perused when the then Chinese Premier put her in charge of a research effort back in 1967.  

Others are in on this gig too. The Bangalore-based oncologist and painter, Paul Salins, is looking for ideas in Indian traditional texts to uncover naturally occurring substances that can pass regulatory muster as new medications. The São Paulo-based Peruvian entrepreneur, Juan Carlos Castilla-Rubio, hopes to sequence the genomes of all eukaryotic life in the Amazon, to make that bounty more accessible. Neither really understands why nature’s bounty works quite yet.

When I look across the academy, this isn’t unusual. Recently a sociologist colleague described how he works with a computer scientist on complex data sets. ML is used to find patterns. Then they use their differing theoretical paradigms to iterate on possible ways to understand these. That in turns drives more formal testing, and further search for patterns. And so on. They refer to this, evocatively, as ‘data ethnography,’ akin to what anthropologists do in the field.

I’d argue that we should more proactively embrace data-derived candidates for introspection – whether they emanate Ayurveda-like from nature or from computer-driven sleuthing – rather than discount these on account of our not (yet) having fully explanatory theories. Indeed, it would be antithetical to the process of scientific discovery to discard such clues.

Intellectual modesty dictates that we should embrace the unknowability, rather than overstate its costs.