STAND: Self-Aware Precondition Induction for Interactive Task Learning

In interactive task learning (ITL), AI agents learn new capabilities from limited human instruction provided during task execution. STAND is a new method of data-efficient rule precondition induction specifically designed for these human-in-the-loop training scenarios. A key feature of STAND is its self-awareness of its own learning—it can provide accurate metrics of training progress back to users. STAND beats popular methods like XGBoost, decision trees, random forests, and version spaces at small-data precondition induction tasks, and is highly accurate at estimating when its performance improves on holdout examples. In our evaluations, we find that STAND shows more monotonic improvement than other models with low rates of error reoccurrence. These features of STAND support a consistent training experience, enabling human instructors to estimate when they have finished training and providing active-learning support by identifying trouble spots that require more training. STAND achieves this by efficiently learning a compact space of greedy classifiers consistent with training data, rather than a finite ensemble of alternatives.

While many kinds of AI are trained on enormous datasets to perform everyday tasks, researchers who study Interactive Task Learning (ITL) design AI systems that can learn interactively on the fly, similar to the way that humans learn. These AI systems learn directly in the moment from people's interactive instruction, forming new knowledge that users can inspect and interpret, and that tends to be reliably correct with repeated use. This is a major advantage of ITL over purely neural AI such as Large Language Models (LLMs), which are often unreliable and have uninterpretable "blackbox" knowledge structures. STAND is an algorithm that helps interactively teachable ITL agents learn preconditions—the "if" part of "if-then" rules—and report back the progress it has made toward learning correct preconditions. Unlike conventional machine learning, STAND is very good at knowing when it is done learning and has achieved reliable knowledge, meaning it can help users know when they are finished training it. In this sense, it is a self-aware learning algorithm.