- by Elana Duffy, CEO, Pathfinder Labs
Since the ol' Nerd Desk can get a little cluttered with, well, nerd talk, here's a quick rundown of what's in - and why you should read - this month's blog:
We want the decisions we make to be good ones, leading us to the best possible outcome. Every choice has risk, however. How do we know if Option A will make us better, faster, stronger or if we would be better off with Option B? Will I be more suited for job X or job Y? Would I feel more relaxed learning meditation from Provider S, or from joining Provider T's running group? We are asking to know the outcomes of each path we could take before we take it, so we can reduce the risk of making a less than optimal choice. But we can't see the future; we can only calculate the most likely results. We can't predict, but we can figure out what is probable. Knowing the difference between those two concepts - and how to improve the numbers - helps us manage expectations, increase successes, and better handle the occasional disappointment.
Probability, prediction, what's the diff?
We all want predictions to be real. We want someone to tell us which opportunities to seize upon and which to ignore. How will I, or someone around me, behave and what are the outcomes? Sure would be awesome to know what the future holds, because the unknown holds too much risk.
The problem is there is no such thing as prediction; there is only probability.
Prediction is "Path X will lead to Outcome Y." Probability is "Path X is Z percent likely to result in Outcome Y." The important distinction in these words can effectively change how we make decisions: if I believe following Path X will automatically result in Outcome Y, I may do nothing differently but follow the path. If I believe I can increase the probability of Outcome Y, however, I will take actions along the way. "Risk" can then be defined as "1 minus the probability of Outcome Y," and mitigation of risk is whatever is done to change that probability.
So. Many. Math words. What does it actually mean?
Prediction is a definitive statement, while probability is a calculation. Knowing the difference can help manage expectations, both yours and those around you. It can also protect you from making promises you can't keep, and even - in extreme examples - win or lose lawsuits.
If you watched television in the late 1990s, chances are you remember Miss Cleo of the Psychic Readers Network. Her late-night, Caribbean-accented infomercials garnered over an estimated $1 Billion for the hotline, which issued regular promises to predict one's future if they called her now.
Well, the Psychic Readers Network was sued in 2002 for deceptive advertising. Miss Cleo - real name Youree Dell Harris, an actress born in Los Angeles and not Jamaica - was no more able to predict a future than any of the agents answering the phones. False advertising and fraud charges were upheld: you cannot promise a prediction, and the distinction nearly bankrupted the company.
It's no different than how Las Vegas runs sports betting: they do not predict which team will win, they say what the probability is of one team winning by a set number of points. Those probabilities - published as "the odds" - are calculated looking at past team behaviors and game outcomes based on a number of factors. Any number of actions (bye weeks, injuries, trades, you get the picture) changes the factors, altering the odds. Only probability, not prediction, can account for chance and circumstance.
Great, but what does all this have to do with understanding humans?
When Psychic Readers Network went under, one would assume the former callers were angry and felt as though they were duped. Interestingly, however, later reports from Consumer Affairs noted that many callers were satisfied with services received. How could this be, when the future cannot be predicted?
The answer lies in probability. By identifying patterns of behavior in the majority of adults - particularly in those who might call in for a psychic reading - a savvy agent answering the call (which, by the way, was never Miss Cleo) could determine what was a likely path for the individual to follow that was further likely to result in a probable future, and read from the network-provided script.
The scripts were sparse and generally only contained the "future", so the agents filled in holes to figure out how to make it more likely to come true. They could help turn "you'll get a call about that great opportunity" into reality if the agent could subtly assess factors like if the person was comfortable with their resume, had practiced interview techniques, and taken other steps needed to increase the probability they would get the job (and if not, coach them to take those steps). Once the outcome - the job - was realized, the caller was satisfied: the prediction came true. The agents with the highest satisfaction rates were simply the ones who could best understand the individual needs and behaviors of the caller, pair those dynamics with how others behaved in the past to create success, and gently guide the caller towards the "predicted" outcome.
This all comes into play when you are looking at psychometrics, the numbers that help us understand personality and behaviors. If your numbers for assertiveness and aggression are high, for instance, it doesn't mean you will (prediction) lose your temper. However, it might indicate you are more likely (probability) to express your frustrations than someone with lower numbers in similar situations.
Where does the technology come in, nerd?
Natural language processing (NLP) is an advanced probability process, taking the science of how both the psychic and the bookie create their payouts - the most likely outcomes - and mixing in a lot of comparative data to increase the accuracy of results. NLP looks at what you say and how you say it, then compares that to what and how 10,000 other people say things - a collection called a corpus - and their outcomes. Based on how similar some parts of your writing or speech are to those in the corpus, the program calculates how likely you are to reach a particular outcome.
For example, if the desired outcome is a customer service bot giving a relevant response to a user's question, it might look at a corpus of previous questions, answers, and user satisfaction with those answers so it can select the response to give you with the highest probability of satisfaction. If the tech is your phone's "predictive" text NLP, it is looking at your past writing for patterns. It isn't predicting what you will write, it is showing you the most probable next word.
Psychometrics through NLP is similar: statistical analysis is conducted on a corpus of thousands upon thousands of writing samples where the writer's personality was also tested using traditional research methods. This analysis generates scales for things like averages for aggression and anxiety. Then the program compares your writing to the corpus, calculating where you might fall on those scales.
What NLP doesn't do is predict your behavior; it calculates probability compared to the average person, all other things being equal. So if your writing is measured at the 85th percentile for assertiveness, it doesn't mean you are always bossing people around or even that you are always 85% more assertive than someone else. It simply means you are likely to be more assertive than 85% of the random joes you find on the street, if both of you were in the same situation.
I'm not a call receiver at a defunct psychic agency, so is any of this helpful to me?
Determining probability is at the heart of making conscious decisions to increase the likelihood of reaching your desired outcome. Knowing how likely are you to be comparatively higher stress, more resilient, feeling lonely, and so on gives you better understanding of how you might react in situations - and if you don't like what you hear, it gives you something to work on. Are you more prone to worry than average, but maybe less likely to be diligent? You might do well - and feel better - seeking out calming, low-requirements activities like equine or art therapy.
Now if we add machine learning and some additional math (topics for other blogs), we can factor in what the average person with those personality measurements did in a particular situation, and if that behavior led to an outcome that was objectively successful. For instance, if multiple people with similar scores of assertiveness and aggressiveness led to them frequently being successful project managers, it is likely that you would make a successful project manager with similar scores. The knowledge of these numbers can help you decide on jobs, services, and plenty more. It's all about determining where you are likely to feel comfortable - as though you "fit" - and that is done through probability.
The future isn't guaranteed; in fact, it's notoriously unpredictable. Instead, we make the best decisions we can to increase the likelihood we will be successful. NLP is no more psychic than Miss Cleo or the Vegas bookie, though it may at first seem like magic. But when used properly, probability can help you assess your options and empower awesome choices.