Enabling a machine to generate real-time intelligence in response to unknown unknowns is one of the biggest challenges in artificial intelligence (AI) research. Dr. Wenxuan Ding, Professor of AI and Business Analytic at the emlyon business school proposed a new dynamic model that offers a novel method for the machine to establish knowledge about an unknown world.

Recently, she presented her paper at one of the most prestigious annual conferences on artificial intelligence - the 28th International Joint Conference on Artificial Intelligence (IJCAI), August 10-16, 2019, Macao, China.

Implicit cognition and understanding unobserved human mind states by machines

It is known that humans often acquire their knowledge through a conscious and deliberative process of concept formation and concept linking with heavy demand on working memory of our brains. For example, people go to various schools to receive education, learn rules on how to play chess, speak a foreign language or use computer-programming languages.  Such a learning procedure is termed as explicit learning because it takes place consciously and results in knowledge that is symbolic in nature and can be represented in an explicit form.

On the other hand, however, humans have the ability to adapt to environmental constraints – to learn in the absence of any knowledge how the adaptation is achieved.  For example, one can speak her/his native language very well without making any grammatical errors even though s/he does not know many of the grammatical rules.  In addition, a person can walk or ride a bike properly although s/he cannot describe the rules of mechanics that human body must follow. Why is it?  Cognitive scientists have found that humans can engage in implicit learning when encountering the world around us.

Implicit learning means that one develops knowledge of a complex stimuli environment without demanding on (1) central attentional resources and (2) analysis involving awareness of the underling abstract rules that govern particular phenomena. It has two key characteristics:

1）the acquired knowledge is fully internalized by the learner, which cannot be observed by other people; and

2）learners themselves cannot verbalize what they have learned (i.e., can only be understood).

As a result, implicit learning yields implicit or tacit knowledge that combines with the explicit learning to make each one of us unique.  Every day, our tacit knowledge about ideas, events, objects or people help determine the way we live and the choices we make.

Therefore, Dr. Ding is interested in whether a machine – a non-living entity – can generate such an ability. Specifically, her research considers a real world situation where humans encounter an environment, and a cognitive machine serves as an observer to watch people’s reactions. Humans may (or may not) engage in implicit learning which are not observed by the cognitive machine.

Dr. Ding investigates the following three questions in the paper:

• Whether the machine can detect if a human engages in implicit learning;
• If implicit learning occurs, can the machine identify what implicit knowledge that the human may generate from the stimuli in the environment;
• Whether the machine itself develops implicit knowledge about unobserved human minds when observing humans’ behaviors.

Currently, artificial intelligence (AI) and machine learning (ML) often focus on explicit learning from existing observed data (e.g., supervised, unsupervised, deep and reinforcement learning) to find potential patterns in the data and ignore unknown unknowns such as the unobserved causal mechanism that generates these data.