AI, Big Data and Decisions

In our daily lives and in the workplace, we make many kinds of decisions. Some are habitual, others impulsive and still others are carefully calculated.

We humans spend a lot of time making decisions. Our decisions are often made collectively and…no one is wholly satisfied. Sometimes the decision is not to decide or to persuade others to make the decision for us. Some decisions meet expectation, others exceed or fall below what we’d aimed for. How decisions are appraised can be quite systematic or highly emotional, and in everyday circumstances the latter may be the more appropriate way (“I loved that movie!”).

Evidence plays many roles in decision making. Sometimes it is ignored. Sometimes it is manufactured and thus not true evidence. Humans can also honestly misconstrue conjecture as evidence – this is commonplace. Evidence may also be cherry-picked to justify decisions that have already been made. Evidence that doesn’t tell us what we want to hear often winds up in the trashcan. Since perfect data and perfect analyses do not exist in the real world, it’s quite easy to find fault with any data or analysis, however sophisticated. Moreover, different data, or the same data meticulously analyzed in different ways, may suggest very different courses of action. Statisticians encounter this practically every day.

At its heart, a decision is a form of causal analysis, though we seldom think of it as such. “If we do A then B will happen” or, retrospectively, “If we’d done A then B would have happened.” A and B can be “good” or “bad.” Causal analysis is very problematic and causation is usually difficult or impossible to prove. Here’s an example of when it’s easy: My lawnmower won’t start, and I think this is because it’s out of gasoline. I take a look and, lo and behold, I was right – no gas in the tank. So, I put in some gas and it starts right up. In this case, causation was deterministic and the result of immutable physical laws.

Management decisions are seldom that straightforward, however. We may be in a situation in which we don’t understand enough about a problem to know what to do. Sometimes we’re trying to solve the wrong problem but don’t realize this. Or, we simply may have no relevant data. For instance, a New Product Development person has a rough idea for a radically new kind of product, and we must decide whether to pursue it or not. In this case we are ignorant, in the parlance of decision theorists, because we have no applicable data or benchmarks at this juncture. All we have is our gut, our unconscious intelligence, in the words of psychologist Gerd Gigerenzer. We may decide to commission marketing research or shelve the idea. Decisions such as these may be described as instinctual.

Evolution has enabled humans and other animals to make quick decisions, some involving life or death, with little or no evidence. Until fairly recently, kings and queens had limited data of any kind and no statisticians to consult. Interestingly, Gigerenzer and his colleagues have found that gut instinct often beats sophisticated analytics, and this has implications for the future of AI. Furthermore, judgment and experience play a crucial role in nearly any statistical analysis, from the very simple to the highly complex. Statisticians don’t just plug numbers into formulas. I’ll return to this point in a moment.

There is also a sort of Twilight Zone that lies between ignorance and determinism that you may have already sensed. This is the land of probabilistic decision making, the natural habitat of statisticians. Here we have the data and analytics tools we need to make good, if imperfect, predictions, e.g., If we do A, what is the probability of B occurring? We may conclude our prediction will be accurate enough to use in our decision making. Not perfect, but better than the proverbial coin toss by a wide-enough margin to pay for our efforts.

Actuarial science was an early application of probabilistic decision-making, as was gambling. Though we don’t normally think of it this way, statistical science is really a form of legalized gambling. Other examples include credit scoring, used by banks for quite some time now, targeted ads and recommender systems. The trendy term for these sorts of methods is predictive analytics, though when I began my marketing research career they were simply considered applications of statistics.

As you’ll have already surmised, many decisions are multifaceted and combine two or more of these decision processes. Extending our earlier example, say we’ve now done some marketing research on the new product concept and the results strongly suggest the product would fail if launched. We don’t know for certain, but our gut tells us to trust the recommendations of our marketing research agency. So, the total decision process was gut + research + gut.

There are also circumstances in which the outcome itself is a matter of dispute. Is this what we really want? Would it be right? Opinions might clash as to what’s the best thing to do and ethical considerations may weigh heavily in our decision. There may be no absolute right or wrong answer, in other words, and decisions such as these are inherently based on human judgement.

Big data, artificial intelligence, machine learning and real-time decisions grab many headlines these days. How will they fit into all this? I’ll have to go mostly with my gut to give you an answer since the historical data I have is limited to say the least, and I also am not an authority on this topic. There are two brief articles I can recommend, though, which are relevant to our discussion. In Neuroscience and Marketing, Professor Nicole Lazar, a prominent statistician working in neuroscience, spells out in layperson’s terms what neuroscience can do as well as its limitations. In Text Analytics: A Primer, Professor Bing Liu, a noted authority on the use of AI for text analytics, gives us a snapshot of text mining in plain English.

Replicating our unconscious intelligence is the biggest challenge AI and machine learning face, and this may prove insurmountable. However, decisions that are highly repetitive and grounded in clear decision rules are increasingly being automated, and this trend will surely accelerate. Though I doubt very much robots will take over the world in the near term, we’re going to see a lot more of them, including very humanlike androids. Will anyone be driving their own car in 2050? I don’t know, but I wouldn’t be surprised if the answer is almost no one. True AI remains elusive, and may always remain so, but by all indications we’re getting closer and closer to this dream (or nightmare). My bet is that machines will be doing a lot of what humans now do in 2050, perhaps most of it. Humans may be doing other kinds of work, since work creates work and new technology creates new kinds of jobs.

Many of you reading this are marketing researchers. How will marketing research be affected? Again, I’ll have to go with my intuition. Historically, marketing research has mostly been upstream from the point-of-sale and usually more concerned with strategic than tactical decisions. It requires a lot of thinking and experience, and aspects of it will be very hard to automate. Many components of marketing research have already been standardized and automated, however, and I see no reason why this will not continue. Costs and benefits will have to be weighed, of course, and quality and flexibility will remain concerns for some clients. Curiosity about what’s inside the black boxes will not vanish overnight, as many black boxes will compete against each other. Sales and client handling skills will continue to be an asset for some time, I suspect.

Similarly, those with strong customized research and analytical skills, and a solid knowledge of marketing, will be in demand for a long while yet. There are many aspects of quantitative and qualitative research that will be out of reach for AI for years to come. The explosion in the number and variety of statistical methods in recent years, for instance, makes automation harder, not easier. Moreover, AI cannot find and analyze data that do not exist – something that has been holding back its progress in fields such as medical research – and the need for competent primary research will never disappear.

How will decision-making be affected? I don’t know for certain. And neither do the machines.

Arrange a Conversation