Simulating Learning Processes in the Brain With AI/ML
Key Stakeholders: Chief Learning Officers, Chief Human Resource Officers, Learning and Development Directors and Managers, Corporate Trainers, Content and Learning Product Managers.
Why It Matters: The skills necessary for success in the corporate world are varied and include hard skills, people skills and situational awareness. While L&D is embracing the use of AI/ML to analyze learners’ data and to personalize learning paths, curate effective content, and attempt to better engage learners, what L&D has failed to embrace is the application of AI/ML to model each of these distinct learning systems, and their interactions.
Top Takeaway: Corporate learning vendors would be well served to develop AI/ML models that capture the processing characteristics of the three learning systems in the brain known to mediate hard skills, soft skills, and situational awareness learning. A comprehensive AI/ML model that captured the processing characteristics of each of these three distinct learning systems could be used to develop and test products and tools that optimize content curation, learning paths, engagement, and delivery processes that will differ substantially across systems and tasks to be learned.
Vendors with the Skillset and Expertise to Build this AI/ML Tool: Cornerstone, CrossKnowledge, IBM, Infor, LTG, Oracle, Saba, Salesforce, SAP, Workday, and likely many others.
Artificial Intelligence/Machine Learning and L&D
Artificial Intelligence/machine learning (AI/ML) are being applied at an exponential rate in the corporate sector. In many cases, the terms are used more as “buzz words” to generate hype, but in other cases, the applications are truly remarkable from AI/ML systems that extract a person’s emotional state, personality or preferences to robots who move fluidly like a gymnast.
AI/ML is also prevalent in Learning and Development. Measures of engagement whether ratings, time on task, completions, or dynamic gaze paths extracted from immersive environments, such as virtual reality, are driving innovation in L&D. Preference ratings, confidence ratings along with accuracy rates and reaction times are driving content organization, development, and facilitate optimized training paths. Learning paths are being personalized, and content is being recommended to learners that derive from the content that learners engage with. All of these advances rely heavily on AI/ML. These data analytics tools are being applied to the rich set of data coming out of learning experiences. They are driving innovation, offer clear paths for the road ahead, and document the ROI from L&D.
Surprisingly, one area in which AI/ML have not been utilized is to model the learning process itself. This is even more surprising given the extensive body of computational brain science research (i.e., computational cognitive neuroscience) focused on modeling the distinct learning systems in the brain. The last 15 years of my academic career was spent focused on this exciting topic.
AI/ML Models of Distinct Learning Systems in the Brain
It is well established from over 30 years of research that different tasks are learned by different systems in the brain. At least three learning systems are critical to learning in the corporate sector. These include the cognitive skills learning system, the behavioral skills learning system, and the emotional learning system. A schematic of each, along with relevant brain regions is presented below.
Critically, each of these learning systems in the brain has very different operating characteristics and thus requires very different content and delivery tools for optimal training. Briefly, hard skills learning is mediated by the cognitive skills learning system in the brain that encompasses the prefrontal cortex and medial temporal lobes and relies heavily on working memory and attention. People (aka soft) skills learning, on the other hand, is mediated by the behavioral skills learning system in the brain that encompasses the basal ganglia and relies heavily on real-time interactive feedback. Finally, emotional learning (e.g., situational awareness) is mediated by the amygdala and other limbic structures and serves to up or down regulate processing in the cognitive and behavioral skills learning systems.
While L&D is embracing the use of AI/ML to analyze learners’ data and to personalize learning paths, curate effective content, and attempt to better engage learners, what L&D has failed to embrace is the application of AI/ML to model each of these distinct systems, and their interactions. Current applications of AI/ML treat learning as a unitary process when, in fact, distinct learning systems in the brain exist.
Likely AI/ML Architectures for Each Learning System in the Brain
The cognitive skills learning system in the brain has evolved to learn facts, figures, and general knowledge. Information is loaded into short-term memory, is mentally rehearsed using attentional processes with the goal of storage in long-term memory. This is a logical reasoning system that would be effectively modeled with a decision tree architecture, much like a complex database. Memory strength, the influence of rehearsal, working memory capacity, attentional processes would all be captured by parameters in the decision tree architecture. Many of these processes are affected directly by neurotransmitter release into the prefrontal, such as dopamine and serotonin. The processing characteristics of these neurotransmitters would need to be modeled in the system. As just one example, dopamine released into the prefrontal cortex remains active for 10-20 minutes. This is one reason why positive affect, known to release dopamine into the prefrontal cortex, enhances information processing. A system like this learns most effectively when information comes in bite sized chunks with a targeted focus, and is repeated spaced over time. This is why microlearning content is so effective for hard skills learning.
The behavioral skills learning system in the brain has evolved to learn behavior. It links perceptions associated with specific environmental contexts with behaviors by strengthening and weakening neural connections in the basal ganglia of the brain. Learning in this system is gradual and incremental. When a particular behavior in a specific context is followed in real-time by immediate reward, dopamine is release and the neural connections that led to that behavior in that context are incrementally strengthened. Thus, when in that context again, that behavior is more likely to occur. On the other hand, when a particular behavior in a specific context is followed by immediate punishment or lack of reward, dopamine is not release and the neural connections that led to that behavior in that context are incrementally weakened. Thus, when in that context again, that behavior is less likely to occur.
Because this system relies on a mapping from a broad array of sensory representation brain regions with motor regions via the basal ganglia, it would be effectively modeled with deep learning or a multi-layer neural network. Again, the processing characteristics of relevant neurotransmitters would need to be modeled in the system. Interestingly, dopamine is also relevant in this system. However, in the basal ganglia, dopamine that is released remains active for only 1 second or so. This is why immediate feedback is so critical to behavioral skills learning. A system like this learns most effectively when practice is extensive and a broad array of training contexts are included. This is why microlearning is ineffective for soft skills learning.
The emotional learning system in the brain includes the amygdala and related limbic structures and has evolved to up and down regulate the cognitive and behavioral skills learning systems. It has evolved to facilitate emotional aspects of learning such as empathy and trains situational awareness broadly defined. Emotional learning would be effectively modeled with a network of excitatory and inhibitory connections that accentuate and attenuate processing in the cognitive and behavioral systems. Although this is likely the most challenging of the three systems to model, some properties are well understood. For example, it is well known that stress, anxiety and pressure adversely affect processing in the cognitive skills system and have either no or minimal effect on the behavioral skills system. This could be represented in the excitatory and inhibitory connection strengths between the emotional system and the cognitive and behavioral systems. For example, under stress, working memory capacity and attention processes would suffer.
Conclusions
Significant advances in product development will be possible when unique AI/ML systems are constructed that each capture the important processing characteristics associated with each learning system in the brain. These distinct AI/ML algorithms can then be combined into an integrated AI/ML system that more fully captures learning in the human brain. An integrated system like this can then be leveraged to more effectively personalize learning paths, curate content, and engage learners. The tools and best practices that achieve these aims will be distinct depending upon the nature of the training task and the learning system being recruited. A comprehensive AI/ML model of this sort will allow vendors to test products and learning tools through AI/ML computer simulation instead of on the fly with real learners. This will allow more detailed tests of potential roadmaps for the future from a functional and technological perspective.