Wild vs Mild Randomness

Much to the ire of my friends and family, I'm a little obsessed with Nassim Nicholas Taleb. Taleb is an enigmatic combination of erudite thinker and twitter fighter. His uncompromising , irreverent style isn't everyone's cup of tea but Taleb's 'Incerto' series of books which include Fooled by randomness, Black Swan, Antifragile, Bed of Procrustes and Skin in the game are worth reading, and re-reading.

Taleb is most well-known for his work on error in statistical models. Predicting the future by looking at the past is Taleb's biggest bugbear. Data is always backward looking, and projections based on past data is vulnerable to big outliers (Black Swans). Normalised bell curves are shoehorned into many fields, even when decisions made from bell curves can occasionally be egregiously wrong.  

When does 'wild randomness' matter?  

There's an interesting dichotomy we can use. Fields that are 'scalable' (Financial markets, wealth, pandemics, terrorist attacks, climate change) which Taleb's own mentor Benoit Mandelbrot names "wild randomness" and 'non-scalable fields' (Human height, length of a tooth and how fast humans can run) which Mandelbrot dubs "mild randomness". Using data to predict the future is more than reasonable in the latter category. You're never going to see a human grow to 100 feet or run at the speed of light. But predicting the severity of the next terrorist attack by looking at the past 100 attacks is a fools errand.  

The sad and scary thing is predictive, data driven models are pervasive in scalable fields and quite underused in some non-scalable fields. Stock traders use 'chart patterns', environmental analysts use 'long term' projections and economists use every model under the sun to predict our economic future. All these are vulnerable to unexpected shocks. Forecasting what matters is near impossible in these fields.

Regularities

Professions like mine have regularities in its environment. A tooth is never going to be 1000x the norm. Cavities, root canals and salivary flows are never 1000x dissimilar across patients. Dentistry is a perfect field for the use of data driven decisions, but alas there are no strict protocols for dental decisions. Dental decisions are made by the best judgment of individual dentists.  

Problems & Solutions

There are two problems to solve. Firstly, that of accumulating enough dental data to create strict protocols. Dentistry is predominantly performed in small family own clinics. There are thousands of small practices across Australia, and the data has been difficult to aggregate to date. This is about to change in a big way. Secondly, aggregating tens of thousands of research articles and conflicting academic guidelines has been very difficult. Again, AI is an opportunity to aggregate conflicting research without losing nuance.

‍