The Learning Equilibrium

Here is a note from one of my 1994 diaries. I had written this on the first Tuesday after taking over as manager of a 216-seater multi-cuisine restaurant and a 30-seater pub in Mumbai.

Record all raw data as per smallest unit of interval available (hourly, by meal times, etc) for a day.
Record all raw data as per smallest unit of interval available (hourly, by meal times, etc) for a week.
Record all raw data as per smallest unit of interval available (hourly, by meal times, etc) for 2 weeks.
Study raw data collected and ascertain as many possible patterns (hourly pattern, meal time pattern, daily pattern, weekday pattern, weekly pattern, two-week pattern).
Analyse various patterns and predict outcome over next same-time interval.
Record all raw data as per smallest unit of interval available (hourly, by meal times, etc) for 2 more weeks.
Study collected data and compare with prediction.

7a. If deviation: Quantify deviation and study pattern of deviation, predict and collect and study (deviation data as well regular data) for a week, correct, and repeat.

7b. If no deviation, study for further 2 weeks and study previous and monthly data, predict, study, compare, and follow previous method.

After 4 months from start:

8a. If no or negligible deviation at end of 4 months from start, you have definite data to define the pattern in Normal State given the set of conditions, benchmarks with defined parameters and negligibility and clear reasons (external which is environment and internal which is process or individual behaviour) for why low or non-pattern deviation might occur in a Normal State.

8b. If at the end of 4 months, there is a clear pattern of stability and also a pattern of more than 10% deviation, then you have definite data to define a Normal State with Exception (Abnormal State) given the set of conditions, benchmarks with defined parameters and negligibility and clear reasons (external which is environment and internal which is process or individual behaviour) for why the pattern deviations might occur in the Abnormal State.

Regular analysis used to be an integral part of any professional’s work, and for managers it was part of core competence; However, the focus of analysis was not the analysis, but the change/improvement in operations that would improve revenue or profitability. Today, the world has changed so much that analysis is done for the sake of analysis itself, and is often divorced from the changes that must be done to improve a product or service or solution. And this most evident when you engage Generative AI platforms - they remind me of people who created work out of nothing, spending countless hours toiling away at tasks that did nothing to improve or change outcomes or take a project closer to successful completion. This is probably where the concept of billable effort emerged from - paid engagement with no purpose but to keep the billing register ticking. Billable effort would make perfect sense if the benefit was passed on to the person doing the work, but when it emerged and for decades thereafter, businesses were the only ones who benefited from it. Which is why the workforce stayed stagnant - skills never improved because there was no benefit to organisations if people moved up the skills ladder. The gig economy of course turned the billable effort play on its head, freeing individuals. Although gig economy has its own problems, but at least it addressed one problem from an individual’s perspective and a few from the viewpoint of organisations.

Illustration comparing analysis methods

While there are several areas of work where Generative AI is clearly demonstrating how it is a benchmark of efficiency, research, analysis, and consultancy are not among those work areas as yet. I had figured that with Generative AI, efficiency would actually amount to getting stuff done expeditiously but the experience so far is far removed from that goal. All that is likely to happen is that workforce will get pared to a bare minimum, with automated systems taking over the onus of inefficiency. Which makes sense in a business perspective - if one has to live with and manage inefficiency, might as well reduce payroll/overhead costs and related legal/engagement headaches.

Generative AI companies are engaging in misadventure by promoting the emerging research and analysis abilities of platforms since there are several drawbacks and pitfalls at every step. It might serve them - and consumers - better if they test and clearly define even 1 task within research or analysis where a consumer can rely on the output of Generative AI.

Illustration symbolising Generative AI effort loop; Generated by author with MidJourney.