Some reasons why data scientists and behavior analysts can work together
São Paulo | Brazil
Hello guys! First of all, I would like to make it clear that I have two main objectives with the publication of this article. My first objective is to present to data scientists who do not have knowledge in behavioral analysis, the potential of this large field of research, especially when we are talking about of delineating research with a single subject. Data science and big data research currently is focused on analyzing thousands of data from millions of people, I believe there is great potential for improvements in our methods when we also start conducting researches with thousands of data from a single subject. In other words, I believe that data science lacks solutions focused on single subject research.
My second objective is to invite my friends psychologists, especially behavioral analysts, to join this challenging area of data science. Behavior analysts are still unfamiliar with computational and statistical methods in conducting behavioral research. An example of this is: if you search for the keywords: big data, machine learning or deep learning in the two major journals of behavior analysis in the world (JABA and JEAB) you will not find any work (last search: 12/2017).
I believe that this intersection of areas will be mutually beneficial. Behavioral analysis can be consolidated as an effectively pragmatic science, and data science can become more mature to work with human behavioral data.
Big data is already a reality!
In 2013 I published a little article discussing the implications of data mining and artificial intelligence for the science of human behavior (you can access it here — in Portuguese). At that time, I was certain of the impacts that this type of technology would have on our lives, but could not imagine that it would happen so fast.
Now, we are already living in the age of data. Internet of things (IoT), Augmented reality (AR), Virtual reality (VR), Cloud Computing, Artificial intelligence, and a lot of other technologies have left science fiction movies and are already part of our lives. We are using all this technology, and they are accessible at a cheap cost. Anyway, currently, everything we do can already be modeled by using two digits: 0 and 1. That’s what it means when we say we’re in a digital world.
Surely you may have heard that all these technologies are producing a volume of data never seen before in human history. You may also have heard of Big Data and data science. Only in 2013, we produced more information on the internet than all previous years together (I do not have updated data about this but, I believe that today (2017) the number should be much higher.
Before you can think that Big Data will only help the industry marketing to sell you more products I cite a small list of things that Big Data can do for the world in the very near future:
- Cure and prevent cancer: performing more accurate diagnoses, performing imaging, analyzing gigantic medical records to find the most appropriate treatment for a patient.
- Eradicate hunger in the world: the big date can change the way we produce food, eliminating waste, suggesting new methods of distribution
- Predict to warn about natural disasters. Forecasting systems can alert areas of risk a few weeks in advance about the risk for certain types of natural disasters.
- Better security, through computer vision data analysis, security cameras can identify and act on cases of suspected criminal behavior.
- Teach children with intellectual disabilities. Predictive algorithms can create personalized teaching programs that are better adapted to the individual repertoire of any individual.
I’ll try to achieve my goals of showing for data scientists and behavior analysts the importance of working together by highlight points where we are still “crawling” at the intersection of these areas. These are just exemples of how the intersection of these disciplines can produce methodological and technical advances for both areas. I believe that in future, the work of researchers at these intersections can produce technologies never seen before.
Some reasons …
Psychologists, especially behavioral analysts, know the complexity involved in science that we call “behavioral analysis.” As experimental researchers, they are accustomed to dealing with the complexity and multi-determination of behavior. They know the several factors and variables involved in determining a single behavior. They also know that there is still much work to be done and that there are many variables that we still do not know. Most importantly, they can see the granularity involved in determining behavior and that in many contexts this can translate into hundreds of thousands of data to be analyzed. However, most of these researchers still do not know the science behind the “big data” and how much it can contribute to the study of human behavior. As I said before, it is still uncommon today to find behavioral analysts who are familiar with tools that could automate and potentialize their behavioral analysis work. Much of the manual work they do could be done by robots and their research could be made much more accurate.
Across the intersection are software and statistical engineers specializing in Big Data. This second group has the domain of information technologies, information modeling and statistics. However, there is still a certain limitation in restricting research on Big Data and human behavior, comparing individuals with homogenous groups.
Search Google for “Big Data and Human Behavior”. Most likely almost all the material you will find will be related to group studies for pattern identification.
My personal hypothesis is that this is because it is the most economically easy to do. The cost-benefit ratio in this scenario is very clear. The fact is that you can tell a lot about a population just by looking at a sample, but you can not say it all.
Most importantly, there are questions you will only be able to answer if you compare the behavior of the individual with his own, as well as compare it with that of groups.
Unlike traditional statistical models, where we tend to look at tens of thousands of data from groups of individuals to predict a person’s behavior, behavioral psychologists usually analyze the behavior of a single person.
I’m not saying that comparing individuals in the traditional way is a problem, I’m saying that we did this because we were limited, today we have technologies for conducting individual behavior level too.
Predicting and explaining the causes of human behavior are related actions, but they are not the same things. We can predict that a child will learn more quickly a topic x of math if she learns before y looking at how thousands of children with the same characteristics learned, but we can not say that a child learned the topic faster y just looking at this fact. To really explain why a particular child learned y, after learning y, we need to look at the behavioral multidimensions of this single individual’s behavior. This is an expensive procedure and quite difficult to do by non-specialist professionals. The results of this type of research, although valuable, do not serve the interests of the market, therefore, little has been invested in this type of approach in behavioral research.
Certainly, the most important point where behavioral analysts can help data scientists with behavioral research is to understand human behavior as a dynamic variable rather than as a static attribute.
Unless you are a student of this area, you probably do not know that all scientific approaches in the area of psychology derive from one of two great schools: the structuralist school and the functionalist school. Explaining the differences between both approaches goes beyond the scope of this text, but for educational purposes only, it is important for you to understand that the methods of research in psychology derived from the functionalist approach tend to be more empirical, and close to the natural sciences (in contrast to the structuralist approaches that deal with the human mind as a structure (eg: Freudian psychoanalysis and Jungian analytic psychology).The analysis of behavior founded by BF Skinner is an approach to psychology heavily influenced by functionalism. Behavioral analysts are psychologists who work with this approach.
B.F Skinner was a Harvard researcher and one of the most important and influential psychologists of all time. Without any doubt, Skinner’s most important contribution to science was the concept of operant behavior. Ironically, this is also the most difficult concept to be understood from all Skinnerian theory, because it reverses the notion of causality of human behavior.
When we speak of human behavior, it is very common to think that what determines or affects the probability of a determining behavior occurring is something that happens before the behavior and not afterward.
The notion of operant behavior reverses this logic because from empirical studies shows that what causes a determinant behavior, is the event that comes after, in other words, the consequence.
The consequence (the technical term for anything that happens after a given behavior occurs) may affect the future probability of this particular behavior occurring again. The form, intensity, and other attributes that determine how probability is affected has been studied by behavioral psychologists. If you are not familiar with this subject, I strongly recommend reading this book: Link to amazon.
I often read papers by brilliant statisticians and computer scientists who fail categorically at this point, that is, they make basic errors in interpreting data related to the determination of human behavior. I do not blame them, that would be the same as demanding that psychologists know how to perform complex mathematical calculations. I just argue that good data scientists, need to understand human behavior minimally if it wants to perform more coherent and accurate analyzes, especially if its task is to explain a particular variable related to human behavior and not just to predict it.
To conclude, I return to the following point: predicting and explaining human behavior are very different tasks. In my conception, very soon applications that involve data science and human behavior will require a lot beyond predicting as a particular behavior, they will need to understand them and then act intelligently in front of them. When this happens, I believe we will have two great benefits for science: (1) in the field of psychology we will have a much broader knowledge on the variables involved in the acquisition of human behavior (2) in the area of data science we will have much more intelligent, capable applications complex tasks. To make its possible, I believe that psychologists and data scientists can already start working together to produce breakthroughs for both sciences.
