AI’s Inconvenient Truth – Part II

Nearly three years ago, here in Epilogues, in the initial post entitled AI’s Inconvenient Truth, we mentioned that something didn’t seem right about IBM’s Watson Health unit. And we weren’t the only ones to notice. There didn’t appear to be any traction.

But that wasn’t the only thing we noticed. If you scanned the entire industry to find all the economic kinetic potential soon to be released by burgeoning AI initiatives, according to the likes of IDC, Gartner and McKinsey, you wouldn’t discover much of anything with respect to realizable returns in the form of positive business value. Whatever that means.

Well, for those of you who didn’t notice, this past week the WSJ ran a series of articles concerning IBM’s move to shop Watson Health in an effort to find a buyer. If IBM was willing to make this public it’s a pretty good bet that they have been shopping Watson Health for more than a year without any luck and the NDA’s were about to expire so, “what the heck?”.

Coincidentally, the WSJ mentioned the same inherent issue that we discussed back in 2018 when it came to realizable returns from a healthcare instance of AI. Data. Or more importantly the sheer volume of data required to produce monetizable insights and the practical and regulatory impediments to getting it. Another important and not inconsequential consideration was the way Watson Health had been positioned, as a content service provider to customer facing healthcare service providers, a queryable question answering facility and epistemological broker of healthcare knowledge. After all, this too was very familiar, didn’t Watson clean up on Jeopardy? So, “what the heck?”.

Only thing is, the value proposition for content repositories had already been well established and it followed a pretty familiar power law (see Social Subsidization and Diminishing Returns – March 2015) and it wasn’t one that augured well for the likes of Watson Health. What IBM might have considered, in lieu of a marginal content play, was a value proposition that put it in direct competition with established healthcare and life science providers like Google’s DeepMind demonstrated with AlphaFold, an AI instance designed to tease out the 3 dimensional structure of proteins, a breakthrough that was fifty years in the making and one that will ultimately restructure the healthcare industry (see Platform Strategies in the Age of AI – August 2016).

But Watson cleaned up on Jeopardy, “So, what the heck?”.

Another conjecture outlined in the original AI’s Inconvenient Truth post was that AI applications would adhere to a power law as well, with only a handful like AlphaFold being responsible for most of the business value created and the rest would end up in the long flat tail of mundane applications like those that might grade eggs and sort out fruit and vegetables.

Well, while IBM was pouring billions into Watson Health, lots of small time innovators were targeting that long flat tail. And their labor is now bearing fruit.

The cover graphic for this post is from the website for Tevel Aerobotics Technologies Ltd and the little bugger in the picture is a Flying Autonomous Robot (FAR) and he’s busy picking apples. What’s more is that the value proposition that results from his engagement is clearly and concisely outlined on the same website, a simple services for labor arbitrage. And while it probably won’t amount to the trillions and trillions in “business value” attributable to AI envisioned by the likes of Gartner and McKinsey, it’s a sensible well grounded proposition predicated on the practical use of AI.

But here’s the thing. This little apple picker has lots and lots of cousins doing lots and lots of labor intensive chores. Like SEeMax grading eggs. And GearBox grading vegetables. Or AMP Cortex sorting recyclables. And although they differ mechanically they are all versed in the primary value proposition of AI, machine based recognition of specific classes of associable artifacts.

And according to some folks, these little guys are about to have a moment.

Not necessarily because of technology but rather economic and exogenous circumstances.

One of the unintended consequences of the recent pandemic was that it made industries of all stripe re-think the labor component of their established value proposition. (see Work in the time of Covid – March 2020) For some, ideas that were only “proof of concept” in nature, cashier-less retail for instance, suddenly took on new and more immediate meaning. Next, the idea of a permanent and proscribed living wage or as it’s referred to in the US a mandated minimum wage, changed the equation for the substitution of capital in low skilled labor markets. According to Investors Business Daily, by the end of this decade, technology will eliminate nearly 17 million, mostly low skilled, jobs, far in excess of the 1.4 million jobs estimated to disappear simply due to wage increases. No wonder that fast food joints from McDonald’s to Steak ‘n Shake to Lee’s Famous Recipe Chicken are piloting cashier-less order processing applications.

So just when it appears that our “gee whiz” AI future isn’t as “gee whiz” as we thought it would be along comes an article entitled “Artificial intelligence predicts the immunogenic landscape of SARS-CoV-2 leading to universal blueprints for vaccine designs” and a glimmer of glamor is restored to our lives.

But as with any technology breakthrough most of the progress we are about to experience in AI will seem like a long slow slog in a landscape both monotonous and mundane, the long, flat tail of the application power law.

And we might not be as sanguine about those prospects as we once thought we would be.

Cover image courtesy of Tevel Aerobotics Technologies Ltd all other images, statistics, illustrations and citations, etc. derived and included under fair use/royalty free provisions.