Charts, Lies, and Data Streams (2)

It’s a double scandal: not only do these glasses make us see the world narratively, but within the stories we could possibly see, they show us only the “normal” ones, and filter out everything else. These glasses cluster everything “normally” in the following loose sense:

• around an average that is representative of the cluster
• dispersed in progressively expanding concentric bands around that average
• inside an outer envelope, allowing for a few outliers beyond.

Now some phenomena are validly normal, and these glasses work great on them. However, the filtered reality in technology and business is full of non-normal phenomena. Among them, power-law phenomena are the most interesting and dangerous.

We’ll come back to power-law/fat-tailed phenomena in their mathematical uniform so to speak, but for now, like “normal” above, “power-law” is wearing civilian clothes. It’s meant in a loose sense that, as we zoom in and out of the data, percentages of inputs and outcomes remain relatively stable. 10% of our green-field investments were responsible for 95% of the total effect (and 10% among that smaller slice, for 100% of my bonus). Or, 20% of code base changes were responsible for 90% of outage hours (and 10% of the faulty code, for 100% of the sleepless weekend in the conference room.) It’s in the sense that both increases and decreases of a metric look similar at different scale levels. (The mention of power in “power law” refers to the exponent of the denominator of the distribution density… see the problem? like in so much terminology, the term’s effective because of the more common connotations of “power”).

Normal glasses hide the percentuality of these metrics, their extreme nonlinearity, and create a habit of vision or of thought. Over long wear we end up potenmyopic (myopic to power-law phenomena, to everything power-law driven, inside past data, inside prediction “data”, and more insidiously…

(…well, look around, look at us, ridiculous, re-scanning our badges, maybe they’ll work now! It feels like a conspiracy).

(If we all see things this way, and deformed to this extent, could it be some “Potenmyopic Bias”, to be added, o horror!, to the stinky toolbox of our bias-biased age? Evolutionary psychology and cognitive science do offer some clues, but it’s hard to tell where Nature ends and where the empirical training begins (and anyway the question fits better into a later discussion of distributions).

The moment to catch ourselves seeing through normal optics is when we’re consuming the data, doing our strategic SWOT analyses and applying case-study methods, or performing our agile-in-the-large ablutions — sure, but we technologists and business leaders wear these normal glasses everywhere, long after the data was consumed, deciding based on it, reaching for the levers to act on those decisions, implementing them. And most of all: when measuring the resulting outcomes. Normalization is always the default setting we start from, which is a problem, but we then leave it on and forget it, which is a scandal. To act in a denormalized world, we must take them off.

This “normality” may or may not involve Humpy the Troubled Siren (technical term we’ll use to refer to the Normal i.e. Gaussian distribution, as in: “Certainly Humpy is a good prototype for normality in the loose sense”). Clear mean, clear deviation bands, and a clear envelope (often drawn at the magic 3 deviations in textbooks, 6 if we feel special).

The term is so much stickier and layered: “normal” in the meaning of “typical” or representative; and normal in the meaning of “norm” as rule, and “normal” in the sense of “expected”… its overloading and memetic charm is just irresistible in this context.

We’re not talking about Humpy the Floppy Hat here, or any other particular distribution, or family thereof, nor of any specific way of averaging, or sampling. Our intuition about compression and clustering stands apart from the mathematical apparatus, with its averages and probability distributions. Going from intuition (with averages and distributions playing groups, clusters and norms) towards mathematics (with averages and distributions playing mathematical objects), from loose to precise meaning, we come across a bridge where we’d like to hold onto our intuition although we should let it go.

In fact the reverse trip, science to intuition, is interesting on its own. We may think that there is something organic about nature perhaps through an evolutionary mechanism creating these glasses, or allowing them to emerge, and then somehow encoding them into our cognition as a default. And this may be true for the mathematically trained/receptive among us; but we know at least since Piaget how oddly this numerical skill sits within other world-building skills. And on our tech planet, Nature lets us do wild things (that we think we “do” virtually), it’s another planet altogether, with constructs like products, branding, projects, operating, algorithms, finite games run infinitely, converging processes, a planet not quite in Nature.

Now, to be fair the deformation is smallest, and even beneficial, when we look straight at the well-understood parts of our business. Normal glasses bring out excellent detail in the linear-ish world we know (we think we know), the “factory”:

• well-understood, steady processes
• sustainment operations
• programs of repetitive projects
• long-running outsourcing contracts
• stable pre-production environments
• high-version tools
• our long-running vertical integrations
• our “near neighborhood” in our industry
• our high-level regulatory environment
• most legacy, documented, in-house configured technology

This because there is great efficiency in thinking in shortcuts and working with tools, in the context of a larger image or operation. Compression means speed and clarity of analysis, better reasoning, better connection between the decision and the outcome in this familiar, good leverage in a somewhat predictable world. The communication is easy, many of the games are easy to formalize and simulate a few steps in advance, negotiations are a matter of providing fact and data, using optionality, expressing intention and will. The principles of power and interest group the players around a clear map. The levers and the tools are out there in the open, and we use our craft and science and experience to apply them in optimized sequence along the best possible path. Oh, and every moment is good, sooner is better, we can be proactive, we can embrace change, and fail a lot, and fail fast, and fail early.

But step into networks, echoing and resonating, step into complexity, non-linear dynamics, n-order effects, step into power laws, and it’s chaos. Sudden crushing competition from “free” (we sold the stuff and bought madeleines with our coffee); brand turning toxic overnight (in the coffee line we tuck the badge in the pocket); multi-day production blackouts (the balancers had the latest patch), massive data loss (small drops but networked, they spread like a pretty forest fire).

We sat bleary eyed in that conference room for 3 days straight poring through monitoring data and logs and reports and bugs, “Follow Me”, maybe I’ll see something that the network owner didn’t? As soon as the servers were brought up they’d die and nobody could explain why and no message came back. Objects in memory were calling the mothership to reassure it, some forgotten test code to bypass garbage collection, the whole data access layer was acting like a bunch of digital lemmings, with a cliff at the end of each group of nodes. Irretrievable, irrecoverable.

And yes, supernovas and unicorns when it goes the other way and it’s great when it happens, but let’s not hex things.

It’s not that, with those glasses on, we don’t see these explosive values or their fractal distributions. Rather, it’s that we don’t see them well, or in any kind of useful way. Our peripheral vision catches but a flash, exploding on the horizon, and we say “it’s a once in 10 years unicorn, how could anyone have predicted a meteoric rise like that” (indeed, one can’t). Or: “it was a once in 10 years outage, how can one predict that” (and one can’t). A flash, an outlier, soon banished to the periphery, off the map, where dragons dwell.

This implication — that the unpredictable or unmodelable Black Swan phenomenon is dismissible — that behavior with randomness outside of our models is not actionable — seems to be the grossest misrepresentation of Taleb’s Black Swan idea.

“Actionable” here has a subtractive, negative flavor, as in: stop-doing-that; don’t-even-go-there, a non-action to be un-taken (but examined, observed, and occasionally profitably bet upon). Outside fintech and digital insurance, among entrepreneurs and VC’s, tech in general seems still fairly potenmyopic, or rather, where there is awareness it seems the mentioned misunderstanding, the meteor flash. On news business segments, in tech journals, in training curricula, on quarterly calls., the rare Black Swan narrative seems to mistakenly conflate power-law events with wild phenomena outside of any rational worldview. A shooting star, a flash on the horizon, to be admired or feared, mentioned with awe, then shrugged off.

Which misses altogether the point: there are fundamental asymmetries built into the truly random reality of outcomes and impacts. The idea is to take Taleb’s warning as a starting point and rebuild our strategic stance towards opportunity and risk in technology as a business.

Our technology strategies are first-order. They seem to be affected by problems of strategy representation and strategy heuristics, they rely on selective hindsight, wishful thinking, fake data science, corruption of framing, contingent fact, fake news.

It’s a strategic salad.

In the highly non-linear world that is now ours to own, what we need is a framework for strategy (or several, but if we could scrounge up one, it would be a good start). A Rosetta that helps us sort through what matters; leaving behind how and where our data representations sit within our story, we need a way of producing strategies depending on where our story sits within randomness. We need a meta-strategy.

But we’ve culturally run out of meta’s, and I feel ill at the thought of one more framework. So “Second-order strategy” will have to do:

• it should produce first-order technology strategies
• it should account for the change of their change, dynamics of their dynamics
• it can’t be a static list of first-order technology narratives
• it sees the narratives as framed within first order narratives
• …themselves distinct from first-order business reality

Next