What I Learned at the “Probability & Statistics” Table

Some of the most important lessons I ever learned about probability and statistics did not come from a textbook. They came from people like Warren Mitofsky and from losing money to him at a poker table.

I have been thinking recently about Warren’s life and career and about the extraordinary impact he had on our industry. He was one of the foundational figures in modern survey research, widely credited with pioneering the exit poll and helping shape the public’s trust in election night predictions.

During his long tenure at CBS News, he led the Election and Survey Unit and helped produce the network’s election night broadcasts, bringing statistical discipline to one of the most visible and consequential forms of public decision-making.

“I believe in God, Country, and Warren Mitofsky” – Dan Rather, CBS Nightly News Achor

I had the privilege of knowing Warren during my SPSS days, and like so many others, I learned simply by being around him.

One of my favorite memories takes me back to AAPOR, where some of the best learning happened after hours. There was an evening session track named “Probability & Statistics,” which, for those in the know, was really code for a friendly, but also a socially high-stakes Texas Hold’em game. That was the joke, of course, but it was also the truth hiding in plain sight.

There, I found my thirty-something self, sitting at the table with Warren, Mike Bucuvalas of SRBI, and a few other industry legends, learning in real time that probability and statistics are never just academic ideas when real people have a few chips, real instincts, and perhaps more importantly, real pride on the line.

I was taking it all in like a sponge and, if memory serves, also making a modest but heartfelt financial contribution to my elders. Such is the reality of continuing street education.

That table taught me something I have never forgotten. Probability is never just math when people have something real at stake.

The Formula Is Only the Beginning

That may be the biggest point I want to make here. As computing has become easier, faster, and cheaper, we have gained extraordinary power. But we have also become more casual about the very things that make insight trustworthy in the first place: the definition of the problem, the source of the data, the sample frame, and the human judgment required to know whether the answer deserves to be trusted.

At its core, probability is about uncertainty. Statistics is one way we bring enough discipline to uncertainty to make better decisions.

That principle shapes far more of life than most people realize. It helps us forecast weather, estimate longevity, price insurance risk, assess disease incidence, predict elections, and now build intelligent systems.

The basic idea is simple, even if the math behind it is not. Put an idea against real evidence, make sure the observations are representative, and ask how likely it is that what you are seeing is true rather than random.

We trust this discipline in medicine. In insurance. In economics. In engineering. In election forecasting.

Yet when it comes to understanding actual consumers, real people making real choices in the real world, we still find ourselves debating the value of real human opinion.

That is a contradiction worth sitting with for a minute.

When Computing Became a Gesture

I am reminded of a conversation I had in 1983 with Ron Wood, then CIO at the Emory University Computing Center. Ron said to me, “Computing will become a brief gesture,” while making a subtle, sweeping hand gesture as if waving in the future.

He was right, and he was early.

In our lifetime, computing has become almost frictionless. What once required time, cost, planning, and patience now happens almost instantly.

But what Ron did not say, and what many of us could not fully see at the time, was that easy computing would also tempt us into lazy thinking.

Back then, computing was expensive. Your data had to be right because rerunning the analysis was painful, expensive, and sometimes simply not possible. Scarcity imposed discipline. It forced precision. You thought harder before you hit go because “try again” was not a strategy anyone could afford.

Today, the cost of computing has collapsed, but the cost of bad input has not. If anything, it has grown.

We can generate answers faster than ever, but speed does not create validity. Calculation is not the same thing as understanding. Precision in processing is not the same thing as precision in meaning.

We want our tools to operate with precision and accuracy, yet we seem increasingly willing to feed them incomplete, unexamined, poorly framed, or simply convenient data.

That should concern us.

Before the Model Runs

That is why this matters so much in the age of AI.

We talk a lot about AI in insights and the importance of keeping “the human in the loop.” But I think we need to define much more clearly what that actually means.

It cannot stop at prompt engineering. It cannot stop at guardrails. And it certainly cannot stop at interpreting the output after the model has already done its thing.

In insights, the human role cannot begin after the model speaks. It has to begin before the model runs.

The human role begins in defining the problem. In deciding what data belongs. In choosing the right source material. In determining the sample frame. In recognizing when the wrong input will produce a beautifully written, highly confident, and completely unhelpful answer.

That is where human expertise matters most. Not at the edges of the process, but at the center of it, where decisions about quality, relevance, and truth are actually made.

And No Tells

Today, AI can play poker against an AI adversary, and that is technically impressive for about five minutes. After that, it starts to feel like two calculators bluffing each other.

The game may be mathematically sound, but it is emotionally empty. Neither player has anything real to lose. There is no emotion. No consequence. No memory of what it felt like to win the last hand or lose the one before it. No fear, no instinct, no pride, no smack-talk, no hesitation, shaped by consequence. And, no tells.

In the human world, “tells” are everything. They are the subtle signals that reveal emotion, uncertainty, confidence, contradiction, and intent. They remind us that human behavior is never just calculated. It is felt.

Human beings are different, gloriously so.

Winning and losing in the human world is emotional, consequential, and often final. No two humans are alike, and even the same person changes from day to day, moment to moment, and event to event.

That dimensionality is not noise. It is the signal. It is what makes markets move, brands matter, and decisions difficult.

And that is why synthetic data, however useful it may be in some contexts, cannot fully replace the messy, inconsistent, surprising reality of human life. Humans are not lab conditions. They are living conditions.