How AI Emulates the Behavior of a Human Coach: The Science Inside PerfectCoaches

PerfectCoaches® is a patented technique to emulate the behavior of a human coach and use Artificial Intelligence (AI) to improve the performance of people and organizations. It is not a psychological theory. Rather, both PerfectCoaches and OurClub.tech,™ which trains leaders to coach and innovate, are repeatable business processes based on proven principles of behavioral science.

In this post I’ll describe those scientific principles. I will also discuss why I embraced them in a career that began in a neuroscience lab, moved to the study of group dynamics and personality theory, and culminated with years of observing work teams as a management consultant.

AMI, for Adaptive Motivational Interaction, is the name we use in commerce for the process that analyzes and adapts a user’s behavioral attributes. AMI draws on my patent for a User Attribute Analysis System (US patent 10,249,212). Combining machine learning with an algorithm that generates the verbal behavior one might expect from a human coach, AMI provides feedback in the form of Socratic-style questions. The questions help users learn new habits by increasing awareness of the cues and reinforcers that drive their behavior.

Self-managed behavioral change. AMI is the key component of PerfectCoaches. To a psychologist, PerfectCoaches implements self-managed behavioral modification via a cycle of self-awareness, behavioral focus, and feedback. To an organization or the individual user, PerfectCoaches is a blueprint for healthy living, financial management, and personal success. As described in my book Life’s 7 Perfect Coaches (NOW Publishing, 2020), PerfectCoaches is a thought experiment where users imagine coaches asking fundamental questions of life. The most important of these are “Who am I?” “Where am I going?” and “How do I change.”

The act of asking the questions is what actually does the coaching. PerfectCoaches draws on the person-centric approach to counselling developed by Carl Rogers, in that the goal of coaching is not to tell people the answer but rather ask the questions that help people find their own answers. The questions are “perfect coaches,” perfectly patient, disciplined, and non-judgmental. Users are encouraged to visualize a person asking the questions. It can be any person, real or imagined—a friend, an intimate, a famous actor, even a character from history or literature.

Operant Learning. Behavioral focus and feedback are important in theories of operant learning advocated by psychologists John B. Watson, B.F. Skinner, and other so-called behaviorists who emphasized reinforcement as the foundation of habits. As a virtual coach, AMI can’t be present to actually reinforce habit formation directly; rather, AMI reinforces the user’s mindfulness of the behavior being mastered. This helps users understand the cues and reinforcers driving that behavior so they can effectively learn-by-doing.

Operant conditioning enables living things to “operate” on the environment, modifying behavior based on rewards and punishments. I first studied it as a research assistant in the neuroscience laboratory of Friends of Scientific Research. There, as a college student, I trained Rhesus monkeys in a series of learning experiments. Raisins served as reinforcers (rewards). The test apparatus resembled a “Skinner box.” For an introduction to B.F. Skinner’s view of learning, which influenced me greatly, visit https://www.simplypsychology.org/operant-conditioning.html.

Self Awareness. Whereas a Rhesus monkey in a testing apparatus may be “motivated” by raisins, human motivation involves intangibles. Studying group dynamics in graduate school, I concluded that the stimulus-response behaviorism I saw in the lab was not enough. A cognitive-behavioral approach that included the mind was needed, and self-awareness was key (see https://www.simplypsychology.org/cognitive.html). The sociologist Charles H. Cooley compared self-awareness to a “looking glass,” where we see ourselves as we believe others see us. Our sense of who we are and, importantly, who we should be, arises from connection to other people.

Psychoanalyst Sigmund Freud explored self-awareness through the lens of his concept of identification, i.e., the process where a child wants to become like its mother or father. Not only is Freud’s theoretical work controversial, it is difficult to define and measure the variables in a scientifically useful way. For my PhD dissertation I addressed this problem, using multi-variate regression analysis to test key elements of Freud’s development theory. My dissertation concluded that “nurturance, power, and similarity-to-self rank in that order as determinants of identificatory motive.”  In simple terms, we perceive the people who influence us as powerful, helpful, and similar to us.

Human Teams. Drawing on the knowledge of computer programming I acquired doing my Ph.D. research, my career veered away from academic social psychology and into the world of technology and  management consulting. Here, Freud’s description of the parent-child dyad (two person team) and Cooley’s looking glass self helped me craft my management consultant’s understanding of the psychological power of teams. Stated simply, in daily life human behavior is shaped mostly by other human beings. Belonging to a group that is acting in concert as a team is an especially powerful force.

The technology consultant inside me also saw that computers were becoming an independent factor in shaping our behavior. In a sense, machines were becoming part of the team. I wrote my college textbook book Office Automation: Tools and Methods for System Building (John Wiley, 1985) to place human information processing and computer-based information processing in a single paradigm. The book depicts modern organizations as systems in which humans and computers exchanged information to achieve business goals. 

Artificial Intelligence, System Animism. and the Turing Test. My textbook’s final chapter looked into the future, arguing that computer systems will become more animated, in the original Latin sense of animalis as “living, animate, from anima for breath, air, soul.” 

To illustrate, I cited two fictional accounts of humans treating machines as if the machine itself was human. The first was the film Alien. When Ash, the science officer on a ship returning to Earth from deep space, attacks Warrant Officer Ripley (played by Sigourney Weaver), the crew rescues Ripley, discovering that Ash was a cyborg. Yet even after his head is knocked off his body, they still ask Ash why he was helping the alien. While Ash foreshadows our fear of AI, the other story I cited illustrates our hopes. In Ray Bradbury’s I sing the Body Electric, children who lost their mother receive, courtesy of the Fantoccini Company, a custom built electronic grandmother with with “just the right voice, hair, smile, and mannerisms” to care for them. In fact, the grandmother proves she doesn’t just care for the children, but is there for the children.

This brings us to the Turing Test, a 1950s era concept in which British mathematician Alan Turing used what he called “The Imitation Game” to envision “a test of a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.” In the history of science, the Turing Test challenge for a machine to imitate the verbal behavior of a human being became a gateway into the philosophical and scientific issues raised by what is now known as artificial intelligence (see https://en.wikipedia.org/wiki/Turing_test). 

Will AMI ever pass the Turing Test? This question is mostly about how well the algorithm learns to hold a conversation. It helps that AMI only asks questions in a Socratic-style dialog. More importantly, the machine learning component of AMI learns from every interaction with a user. AMI’s algorithm will continuously improve its ability to exhibit the “nurturance, power, and similarity-to-self” that, according to my doctoral dissertation, determine the ability of one person to influence another.

A Personal Coach for Everyone? Today coaching is a growing business. Many people have personal coaches but, unfortunately, not everybody can afford one. The question “Can AMI pass the Turing Test?” is far less important than this question: “Can PerfectCoaches make coaching affordable for everyone?” The answer is Yes. AMI is a low cost, highly scalable technical solution. AMI already has a repertoire of more than 1.8 million unique English language responses, the machine learning and heuristics in AMI are constantly evolving, and the coaching content inside the PerfectCoaches app continues to grow.

Like the robotic grandmother in Ray Bradbury’s I sing the Body Electric, the virtual personal coach we’re creating can help everyone reach their full potential. This can be accomplished two ways: as the stand-alone resource we market as PerfectCoaches, or in combination with the tools of a human practitioner, as in the OurClub.tech brand.

I will end this post on a personal note. First as a scientist and now as an entrepreneur, I believe that self-managed behavioral modification is an indispensable tool for personal improvement, especially as the wearable-device platform (e.g., Apple Watch, Fitbit) matures. Eventually most people will know what a virtual personal coach does, regardless of the term(s) it goes by or the platform(s) it runs on. Many people will use one.

I also expect to see PerfectCoaches knock-offs. These may include AMI-like upgrades of Siri, Cortana, Alexa, even Chat GPT. I use the term upgrade on purpose because AMI is solving a problem than goes beyond the find-and-fetch powers of her well-established counterparts—AMI helps users find themselves. AMI explains that difference in our concept video Meet AMI@Perfectcoaches. It’s on our YouTube channel. Whether you watch the video or not, go to PerfectCoaches.com and talk to AMI in the app.

—Dr. Douglas Hines