AI/ML…you cannot afford to not have it!

The Great Corporate AI Arms Race

Marcus Aurelius-esque commitment and judgment is critical to successful deployment. There is no substitute for human leadership.

“[Management does] not know their competitors’ better working models and workflow have led to 2.6% better efficiency per day. 
Given the finite resources in capital and time, a bad implementation can sink the boat, without a second chance. 
You cannot afford to mis-deploy nor under-deploy it; your company could soon become irrelevant.” –Dr WHU


He’s back with us for June, and we’re thrilled. Dr Albert Hu is our storytelling eyes and ears into the brave new world of artificial intelligence (AI) and machine learning (ML). He’s like having one’s own trusty Fodor’s guide to AI/ML in your hip pocket.

“Treat with Utmost Respect Your Power of Forming Opinions…”
The Meditations, III.9
Roman emperor Marcus Aurelius penned the enduring classic The Meditations, which has been used in Western leadership education for more than a thousand years.  Started with the writing above, he went on to reflect how meticulously leaders (and managers) had to be in forming their opinions.  “Ask yourself, ‘What is causing the image or idea now forming in my brain?  What elements compose it?  How long will it last?’…”, III.10.  This emperor, having spent decades fighting on the front line for Rome in daytime, further detailed, presumably at night, how not to make mistakes in decision making.  It reads almost like navel-gazing.

Now we know the “blink”-aspect of opinion formation.  We know our thinking goes through two biological neuro pathways that could be either slow or fast, detailed in Thinking, Fast and Slow, by Nobel economics laureate Dr. Daniel Kahneman

Isn’t there an article on the seven (7) steps to successful deployment of AI/ML, in the latest issue of my B-school’s Business Review? I do not have time to navel-gaze Greco-Roman philosophy thousand years ago.  I am rushing to my MVP meeting after yesterday’s CRISP-DM meeting.

AI/ML implementation is the greatest opportunity and the greatest peril in our generation.  The failure rate is high, despite all management teams having read the “7 Steps” or the “5 Steps.”  The failure rate is even higher if we consider the partial successes and the borderline ones for which the management claims success. 

Furthermore, if your implementation is inferior, the false successes could be as bad or even worse.  It gives the management team false confidence.  They do not know their competitors’ better working models and workflow have led to 2.6% better efficiency per day.  Given the finite resources in capital and time, a bad implementation can sink the boat, without the second chance.  You cannot afford to mis-deploy nor under-deploy it; your company could soon become irrelevant.

For lack of better analogy, AI/ML is like the atomic bomb in the 21st century’s business world.  You cannot afford to not have it.  The decision on how to use it, if one reads the biographies of all involved, rests on a lot of navel-gazing.

Marcus Aurelius-esque commitment and judgment is critical to successful deployment.  There is no substitute for human leadership.

We code in Python and have all the tools: Docker, SQL, Kubernetes, Spark, Tensor Flow, Tableau, Lobe, Curiosity Algorithm; you name it.  Our team is led by a CDO (Chief Data Officer) with experience in all phases of deployment: data wrangling, modeling, prediction, hyperparameter turning, etc.

How to win decisively and continuously in AI/ML deployment?

All self-respecting companies are doing the same.  To grow in market, one needs to do better than others.

Let us look at a little, scrappy company called Toyoda Motor Corporation decades ago.  To Messrs. Sakichi Toyoda and Taiichi Ohno, English is the second language.  We now know that the Toyota Motor Corporation has revolutionized and impacted corporate processes worldwide. MBAs now know Kaizen and Six Sigma, tool kits in the highly revered Toyota Production System.  All this started with, in the words of Mr. Toyoda’s team, the “Five Whys Method” as “the basis of Toyota’s scientific approach by repeating ‘why’ five times [so that] the nature of the problem as well as its solution becomes clear.”  

A bit of navel-gazing.  A lot of global impact.

In Far Eastern countries, 大學 (daxue; “Grand Learning”, “Grand Scholarship”) is the classic that all intelligentsia and civil-servants-to-be have had to read and write essays on, for thousands of years.  Starting from Chapter 5 on the pursuit and application of knowledge, it speaks of the following epistemic steps: first, 格物 (gewu; close first-hand observation of the subject, be it natural objects or societal phenomena); second, 致知 (zhezhi; forming your opinion in disciplined manners), and the next step, 诚意 (chengyi; not deluding oneself).

Assuming you have a small portfolio of five commercial buildings.  One popular guideline in the “7-Steps” or “5-Steps” is that AI/ML is not needed if the data points are few.  The understanding is that if you have a portfolio consisting of 100 buildings, then let us talk about AI/ML.


“Omphaloskepsis” or navel-gazing is contemplation of one’s navel as an aid to meditation.

Now navel-gaze at your three commercial buildings, ask “whys” five times (or five why-nots).  I know I sound strange.  Bear with me.  With IoT sensors inexpensive and measuring various physical properties, why not install multiple types of real-time sensors?  This will give the owner 24-7 non-stop readings of the temperatures inside and outside the building; the consumption of the electricity, water; the level of ambient noises, vibrations, ventilations, etc. 

Suddenly, you have so much data to analyze for business decisions.  Why not subscribe to an inexpensive AI/ML workbench?  If the owner wants to minimize the energy cost; he/she can look at the inside-outside temperature differences during daytime and night, during summers and winters, and the locations of largest temperature differences over time. 

These help owners decide:  is it an insulation problem?  If so, which windows and/or which entrances?  The owner can then lower the energy cost by upgrading the insulation.  With AI/ML able to read emotion, why not gather additional emotion data from tenants?  The owner can then improve tenants’ experiences with the buildings.

Your portfolio buildings will operate at higher efficiency and at lower cost, just a bit better than your competitors every day; and your customer analytics will keep them a bit happier every day.  Day in and day out, your firm wins because of the compound effect of 1.3% better every day.

How about the cost of implementing AI/ML?  The cost of hardware is low now.  Software-wise, the stack of AI/ML tools and services is low.  In fact, both are so low now, most small- and medium-sized enterprises (SMEs) can afford.  All SMEs should learn to implement just to stay relevant.

So, what are the differentiating factors?

It is YOU, the human who brings in the business expertise, determination, and good judgment.  Formulaic 7-Steps and the likes get the implementation done.  It is only in the spirits of the “Five Whys” that one brings AI/ML to sustained profitability and excellence.  For example, it is estimated that more than 50% of AI/ML deployment time is on raw data preparation; provenance, wrangling, clean up, etc. Human attentiveness to raw data affects the quality of the models; therefore, the success of the deployment.  Other key factors to success, such as talent acquisition and workflow planning, also require good human decisions.

“Five whys” stems from the lenses of Greco-Roman and Chinese-Far Eastern thinking on dedication to direct first-hand observation, discipline in forming one’s opinions, and taking utmost care by rooting out cognitive bias and self-delusion in the process.  These need to be embedded in corporate culture by leaders.

Executives, Charge Forward!!!   (Across the moat of Affordance Gap.  Warriors of S&P 500s and SMEs!)

Executives often do not know what to expect from AI/ML deployment.  Coding alone is not deployment.  In fact, AI/ML is based more on high school probability and statistics.  Do not let fancy terms of AI/ML tools, the “7 – Steps”, threw you off.  Executives and Subject Matter Experts (SMEs) need to be in the trench with the general data professionals.  You share your business insight from the “real world.”  The AI/ML team develops the working models from the data.  Both sides must sit down to compare the business insight and the predictions from the model, and to find creative ways to take advantage of the new capability.

You can steward your team to build the right kind of working models for successful implementation.  You can also be the weakest link.  For the unfolding of insight and continuous model adaptation over time, be judicious in forming business opinion, disciplined in data preparation, and diligent in analyzing epistemic and aleatoric uncertainties.  Understand also the differences between Shannon uncertainty and Knightian uncertainty. 

The working models are never the reality, except that they are useful.  Do not become complacent once the models are in production, that would be deluding yourself and the team.  You need to do continuous iterations and improvements.  Kaizen.

Not Only Your AI/ML Algorithm.

A dynamic digital market is one where better AI/ML algorithms will learn faster and then take food away from weaker algorithms.  Data-driven algorithms are, by design, ones that collect voraciously customers’ every click and stalk other algorithms to strike when and where other algorithms exhibit weakness.  Your company’s optimal strategy in the dynamic market is complex.  Sometimes it is in your best interest to use your AI/ML’s “frenemy” part of algorithm to stay within the ecology of a super-platform, such as Amazon or Google  Sometimes, your algorithm should go full behavioral in pricing and customer acquisition.

Companies now build up their AI/ML arsenal, not unlike atomic weapon build-up in the past. When one launches an asymmetric market share blitz, others need to counter match-by-match, in a game theory manner with limited “God View.”   Regulatory agencies, light touch or not, need to protect our welfare, but they often play catch-up.  “Digital Invisible Hand” does not always protect our welfare.  Humans need to be the key decision makers in ambiguous situations and on uncertain issues, such as protecting our often self-conflicting interests in an age of cyber enterprises.

Since thousand years ago, “in the minds of disciplined and pure man” (The Meditation, III.8), one should: 1. 格物 (close personal first-hand observation), 2. 致知 (forming opinion in disciplined manners), and 3. 诚意 (never deluding oneself). 

This is even more so in this AI/ML-enabled hyper-dynamic world.