Improving Human-in-the-Loop Adaptive Systems Using Brain-Computer Interaction

Although self-adaptive systems try to automate operation and reconfiguration, many such systems can benefit from human involvement in the adaptation process, especially for uncertain circumstances. For such "human-in-the-loop" adaptive systems, a critical challenge is to determine whether, when and how to include the human. Existing research has modeled the benefit of input from human mental states, e.g., variables from the Opportunity-Willingness-Capability (OWC) model, but fell short of operational implementations. The latter require unobtrusive measurement of human mental states. As a proof-of-concept, we conceptualized a self-adaptive system for stock prediction, which is fed with stock data, behavioral and neural input (electroencephalographic neuromarkers) to dynamically assess OWC variables and to predict if a human would likely enhance stock market buy/sell decision. We identified neural activities related to the OWC model and demonstrated enhanced outcome from adaptive involvement of the human. Both human and computer benefit from optimized (at time autonomous or cooperative) interactions. This paper illustrates one step forward in the evolution of software systems: from inception as programs that relied heavily on human intervention, to self-adaptive systems that aimed to function autonomously, and finally now in a third co-adaptive era that involves very close symbiosis of human and computer into cooperative tasks.