by Yair Gross
The Boundaries of Math, Science, and Technology
As an engineering student, I take quite a few math courses. It is no secret that math lies at the very core of technological advancement and, some even suggest, of the universe itself. And yet the occasional mathematical application will still come along that reshapes my already-progressive notions of what modern math and science can and cannot do. I recently happened upon one such application of linear algebra and differential equations with fascinating ramifications for the future of artificial intelligence and the ability of AI to understand and predict human behavior and emotions.
Where Science Meets Literature
The aforementioned application was first proposed by mathematician Steven Strogatz as a teaching tool for explaining systems of differential equations. Strogatz suggests modeling a fictional relationship between a fickle Romeo and a reciprocal Juliet, in which Romeo and Juliet’s feeling for one another cycle endlessly between love and hatred. Though Strogatz never intended to realistically model human emotions, he sparked an idea that went on to make regular appearances in journals of both mathematics and psychology. Recent articles have attempted more honest mathematical representations of the classic Romeo and Juliet story. Some researchers have even applied similar ideas to different classic literary relationships, covering the likes of Beauty and the Beast, Pride and Prejudice, and Gone with the Wind. Yet others have extrapolated Strogatz’s idea from literary analysis to more realistic scenarios, modelling the emotional and psychological dynamics of nuclear families, triadic relationships, social networks, and even predator-prey relationships.
Each article considers a different set of emotional drivers and a unique psychological undercurrent, achieving varying degrees of accuracy, complexity, and stability in their results. Despite the wide variety in their approaches, or perhaps because of it, nearly every paper on the subjects admits a degree of crudeness in their analysis. It is a disclaimer that seems to undermine the entire field of study, a testament to mankind’s enduring inability to algorithmically understand its own complex emotional processes. And yet the research goes on, new papers are published, and the subject continues to spark academic interest and conversation.
Math as a Love Language
I believe that the sustained interest in modelling relationships does not lie in the particulars of any one study, story, or relationship. Rather, there is a deeper fascination that the scientific community has with our ability to mathematically and functionally define human emotions. While no individual article can solve the puzzle of the human psyche or universally predict our behaviors, cumulatively, they are building a mathematical language to describe those behaviors and to mathematically express emotional changes even though they cannot yet be accurately predicted. Each new paper describes and quantifies yet another facet of mankind’s emotional composition, or another contributing factor to our behavioral patterns, further expanding and deepening that language.
The ramifications of this linguistic evolution are as far-reaching as they are inevitable. In addition to paving the way for further academic research, the potential impact on artificial intelligence technology is tremendous. Much in the same way that simpler programming languages have enabled people to better understand and interact with computers, a function-based language to describe human emotions could enable computers to better understand and interact with people, giving machines a far higher degree of emotional intelligence than they are currently capable of.
Artificial Emotional Intelligence (So Far)
The concept of artificial emotional intelligence has been a point of contention among academics, executives, and technologists for years. Renowned historian and author Yuval Noah Harari argues in his book Homo Deus that machines can and will eventually mimic every human neurological process, of which emotion is simply a part. Many others, among them Microsoft founder Paul Allen and University of Ireland professor Phil Maguire, have claimed that AI may never be able to fully emulate the physiological complexity of human reason and sentience.
Emotional intelligence in machines could manifest in any of several ways, each requiring different degrees of complexity, learning ability, and computational power. The three primary types of emotional intelligence currently attainable by machines are (1) expression, (2) recognition, and (3) behavior alteration, which is largely a combination of the first two.
- Vocal and visual expression are among the most basic forms of emotional intelligence, enabling a computer to communicate how it ‘feels’. Far from requiring true AI though, mechanized emotional expression depends only on a series of preset alterations to an auditory or visual output. A robot could, for example, show an ‘angry’ face, or change the pitch or rate of its speech to make itself sound happy or sad.
- Automatically identifying different emotions is a somewhat more complicated but similarly attainable process, often using facial or vocal recognition to identify different expressions or body language.
- Once a machine can identify and express emotions, it is just a small step away from understanding emotions well enough to alter its behavior accordingly. As early as 2002 researchers were experimenting with emotionally responsive, artificially sympathetic computer programs and finding that they helped to mitigate user frustration.
Evolution of Artificial Emotional Intelligence (Looking Ahead)
As impressive as emotionally intelligent technology already is, its potential is practically boundless. More advanced emotion-identification technology could one day detect depression and other mental illnesses. A bot that truly understands or ‘empathizes’ with human emotions may eventually be able to provide emotional support in real time. Imagine an automated virtual assistant, perhaps a futuristic version of Siri or Alexa, that could help users through anxiety attacks, depression, and other moments of personal crisis. Built on a deep understanding of human psychology, emotional and behavioral patterns, and available information about the user, such an assistant could offer realistic, automated, sympathetic responses, calculated to achieve maximum positive change to the user’s emotional state.
Of course, there is an obvious dark side to any technology capable of pointedly manipulating a person’s emotions. It is with these risks in mind that organizations such as the IEEE set guidelines around the use of automation, and that the U.S. government has made a continued attempt to better understand and regulate artificial intelligence. That dangerous degree of advancement is still a long way into the future and the ways in which we interact with technology, along with the regulation surrounding it, will likely have evolved by then to account for and protect against the potential risks. These ideas, and the attention that they have been receiving from the governing bodies of technological progress, only further speak to the incredible potential that lies in artificially and emotionally intelligent bots.
As another, perhaps more immediate example, online dating platforms like OkCupid algorithmically approximate users’ personalities to try to determine matches that are most likely to be successful. They generally do this by simply asking questions and matching up users according their overall similarity. But imagine if their algorithms could accurately simulate the emotional progression of each partner in a hypothetical relationship spanning several years, checking the projected growth, decay, and limits of each partner’s feelings towards the other with respect to time. This would enable dating platforms to effectively predict the stability and outcome of any potential relationship.
Language is Key
The key to unlocking artificial emotional intelligence at such a high level is translating human emotion into a language that computers can understand and analyze. Mathematics is one such language. By creating a mathematical language to describe emotional behavior and relationships, researchers are building the dataset on which the next generation’s emotionally intelligent machines will train.
Machine learning algorithms are incredibly good at finding, learning from, and applying patterns in data sets. They have been used to identify linguistic patterns, mathematical patterns, and soon, with the recent and ongoing translation of human behavior into a series of equations, they may learn to identify psychological patterns. So, even though the individual results of each study do not contribute tangibly to our scientific understanding of psychology, math, or literature, this new language that they are creating and the training samples that they are providing may one day be the foundation upon which the first emotional AI is built. This in turn will eventually reveal new, interesting, and actionable insights into the human psyche.
From College Math to Emotional Intelligent AI
In addition to being a fascinating subject on its own, artificial emotional intelligence demonstrates the degree to which even the most advanced and futuristic technological developments trace back to the same core mathematical principles. There is a valuable lesson in that realization, both for me as an engineer, as well as for anyone who develops, depends on, or in any way interacts with technology in daily life. Technological advancement and innovation are not strictly matters of knowledge or training, but of creativity and ingenuity.
Further advancement of linear algebra, calculus, and other math and physics methods are, without a doubt, tremendously important. And yet the foundation of progress, and the eventual key to emotionally intelligent AI and other coming technological attractions, lies equally in extending classic, “old-school” methods to new fields and imaginative applications. At their cores, artificial intelligence, driverless cars, radios, the Hubble Space Telescope, the typewriter, this jumping Disney robot, and nearly every invention in modern history were created simply (or not so simply) by reapplying the same well-known fundamental truths in creative and original ways.
The creation of a dynamic, computer-accessible language for human emotions and behaviors, and the ongoing development of emotionally intelligent AI are just the tip of what can be accomplished with artificial intelligence and machine learning, by creatively applying classical mathematics to new facets of life and technology.