AI’s “dementia moment”…coming soon to a supply chain near you

During the last year the field of AI let out a silent scream, or maybe it was a yawn, either way the market seems to have missed it entirely.

We could chalk that up to the Covid distraction of 2020 perhaps, but regardless there are implications for all data intensive industries, in particular Supply Chain.

Face Value

The tectonic shift began mid-year, when out of nowhere IBM announced it was basically abandoning its facial recognition program, quickly followed by Microsoft, and then finally Amazon, who stated their offering known as Rekognition, would no longer be commercially available. (Incidentally, just this week, almost 12 months on, AMZN reconfirmed that it’s suspending sales to law enforcement indefinitely).

In all the noise surrounding the use of facial recognition technology and the sudden collapse with the “big three”, what’s been somewhat overlooked is this... Facial recognition’s failure is a failure of AI.

The common thread in all of the commentary on algorithmic bias etc., that hasn’t been spelled out enough, is that this is a significant failure at the core of AI itself.

Arduous debate and discussions about training data sets, and how AI may be amplifying existing bias, (in this case, racial bias) are still only scratching at the surface.

Put simply, that these incredibly sophisticated and complex arrays of machine learning and AI models can even develop serious bias at all- and then not detect it within their own computations and outputs- is bordering on comedic.

A major blind spot and strike one for AI.

Roiled Markets

Another front where AI limitations quickly emerged was that of wildly see-sawing financial markets.

The free fall across all major bourses globally, once the pandemic took hold, was already frightening. But far worse was the inability of the plethora of “Quant” and algorithmic based trading/analytics systems to begin to make sense of what was happening, or much less predict where things would go and what might happen next.

This was captured nicely by Wired (among others) in its July 2020 headline piece “Even the Best AI Models Are No Match for the Coronavirus”.

That’s not to say that AI has no place in financial market analytics, but rather that its capability to deal with anomalous events and meaningfully support human decision making, was found wanting.

Strike two. (Arguably).

Pandemics and Supply Chains

The main game with AI’s shortcomings throughout this period has been within the overall pandemic response itself.

On reflection,AI’s contribution has been greatly conspicuous… by its absence.

In several key areas where we might have expected AI systems to save the day or at least make a significant difference, somehow it seems to have fallen short.

Areas such as transmission tracking and predicting spread, automating contact tracing, predictive diagnosis and predicting requirements for medical supplies, have all so far failed to generate meaningful results when coupled with AI.

In fact some have suggested that this should have been AI’s big chance to shine, and if we can’t leverage AI in these times, then when can we and what’s even the point?

Nowhere will the question of AI and algorithmic bias be more crucial, moving forward, than within Supply Chain. As manufacturers, producers and distributors of medical equipment, vital supplies and even vaccine doses themselves, scramble to meet the volatile and urgent demand, supply chain planning and execution is even more critical.

Production, capacity planning, demand and forecasting, plus even shipping, are all supported to some extent, by predictive based modelling and are greatly needed as part of recovery efforts.

But there is a major catch…

Right now, the aggregate global supply chain data set (for want of a better term) is seriously skewed and therefore fundamentally flawed. The sum total of data globally for any geography and in any category is monumentally tilted and unreliable, because of Covid impacts. This will also apply within any subset of the data. (Case in point, capacity supply and demand within Shipping).

All of which means that more than ever, our models are vulnerable to ‘impaired reasoning’, a techno-dementia so to speak, in the clinical sense of the term.

An enormous amount of data correction and remediation over a much longer timeframe will be required to even begin to achieve equilibrium of data following the pandemic “spikes”.

We can’t simply “roll back” to the pre-Covid data set, because we’re in a new normal, (which is still evolving). But at the same time we can’t just jump to the post Covid data because it’s vastly noisy and unreliable.

This is strike three unfortunately for our current crop of AI “miracle-ware”.

Taken together this all points to what might (in years to come) turn out to be an inflection point with industrial grade AI. This may well be the ‘cross roads’ moment, when we shift our thinking toward the field of AI overall, and return to a more pragmatic position.

AI is good when it’s good and can make a difference in some areas (think pattern recognition), but the implicit and now commonplace assumption that it can readily replace human beings or “do it better” or “be turned loose” in a given area, is increasingly dubious, as this past year has shown.

For those of us working in Supply Chain during these trying times, afar better perspective might be that AI can augment and support human efforts rather than replace them.

This is certainly a theme we’ll return to in future…