Thinking Fast and Slow in AI: the Role of Metacognition

Theories into human learning and cognition have led to much research into new methods and structures for Artificial Intelligence (AI) and Artificially Intelligent Systems (AIS) to learn and reason like humans. As we move toward completely autonomous AIS, the ability to provide metacognitive capabilities becomes important [Crowder and Friess 2011b] in order for the AIS to deal with entirely new situations within the environment it may find itself (e.g., deep space, deep undersea). Presented here are theories and methodologies for Constructivist Learning (CL) processes that provide the methodologies to allow completely autonomous AIS to understand, evaluate, and evolve its "Locus of Control [Watts 2003]." Presented will be the a discussion of how the use of AI learning systems, like Occam [Crowder and Carbone 2011a] and PAC learning can be combined with Cognitive Economy concepts to provide this constructivist learning process to allow a Locus of Control evolution within the...

American Journal of Intelligent Systems, 2013

This paper focuses role based architecture for complex agent and the idea towards the developmental approach of model for metacognition in comp lex agents with the principles of cognitive architecture. This model exp lains the mechanis m o f metacognition in a co mplex agent. Ho wever fro m arch itecture it is observed that complex agent under metacognition princip le works well with less time in an environment. Refinement of the decisions is being made by the complex agent under metacognition environment. Fro m the graph it is observed that the correctness and accuracy of decision gets increased in case of complex agent which is inevitable in nature.

2013

We describe ongoing work toward automating human-level behavior that pulls together much of traditional artificial intelligence in a real-time robotic setting. Natural-language dialog, planning, perception, locomotion, commonsense reasoning, memory, and learning all have key roles in this; and metareasoning is a sort of glue to guide the robot through rough spots.

2008

Abstract As automated systems become more complex, their propensity to fail in unexpected ways increases. As humans, we often notice their failures with the same ease that we recognize our own plans going awry. Yet the systems themselves are frequently oblivious that the function they are designed to perform is no longer being performed. This is because humans have explicit expectations–about both the system's behavior and our own behaviors–that allow us to notice an unexpected event.

At higher levels of abstraction, there are finite and limited ways in which agents can fail. Therefore, it is possible to create a general purpose agent that notes anomalies, assesses them and guides responses. The metacognitive loop (MCL) agent aims to achieve this objective using three ontologies; indications, failures and responses. Continued applications of the MCL agent to increasingly more complex situations requires that the MCL agent be equipped with a flexible interface that enables integration with a variety of cognitive agents without programming, and, with a persistence capability that independently and actively monitors expectations over time as well as generate expectations dynamically. This paper describes the work undertaken to produce MCL3, a general purpose metacognitive agent that builds upon the existing MCL2 implementation.

Cognitive 2013, 2013

Agents situated in dynamic environments have limited time to deliberate before performing their actions. Cautious agents that deliberate for too long may miss deadlines to accomplish tasks whereas bold agents that deliberate for too little time may behave rashly or miss opportunities. There are several approaches discussed in the literature that rely on meta-level mechanisms to monitor and control the deliberation time. These approaches seem to follow the view that the metalevel mechanism is an external component not constrained by the same resource limitations as the underlying agents deliberation mechanism. In this paper, we present an approach to time-bounded metacognition wherein an agent monitors and controls its deliberation and metacognition within the uniform framework of Active Logic.

Humans confront the unexpected every day, deal with it, and often learn from it. AI agents, on the other hand, are typically brittle-they tend to break down as soon as something happens for which their creators did not explicitly anticipate. The central focus of our research project is this problem of brittleness which may also be the single most important problem facing AI research. Our approach to brittleness is to model a common method that humans use to deal with the unexpected, namely to note occurrences of the unexpected (i.e., anomalies), to assess any problem signaled by the anomaly, and then to guide a response or solution that resolves it. The result is the Note-Assess-Guide procedure of what we call the Metacognitive Loop or MCL. To do this, we have implemented MCLbased systems that enable agents to help themselves; they must establish expectations and monitor them, note failed expectations, assess their causes, and then choose appropriate responses. Activities for this project have developed and refined a human-dialog agent and a robot navigation system to test the generality of this approach.

2012

Humans confront the unexpected every day, deal with it, and often learn from it. AI agents, on the other hand, are typically brittle-they tend to break down as soon as something happens for which their creators did not explicitly anticipate. The central focus of our research project is this problem of brittleness which may also be the single most important problem facing AI research. Our approach to brittleness is to model a common method that humans use to deal with the unexpected, namely to note occurrences of the unexpected (i.e., anomalies), to assess any problem signaled by the anomaly, and then to guide a response or solution that resolves it. The result is the Note-Assess-Guide procedure of what we call the Metacognitive Loop or MCL. To do this, we have implemented MCLbased systems that enable agents to help themselves; they must establish expectations and monitor them, note failed expectations, assess their causes, and then choose appropriate responses. Activities for this project have developed and refined a human-dialog agent and a robot navigation system to test the generality of this approach.

Proceedings of the AAAI Conference on Artificial Intelligence

This paper proposes a research direction to advance AI which draws inspiration from cognitive theories of human decision making. The premise is that if we gain insights about the causes of some human capabilities that are still lacking in AI (for instance, adaptability, generalizability, common sense, and causal reasoning), we may obtain similar capabilities in an AI system by embedding these causal components. We hope that the high-level description of our vision included in this paper, as well as the several research questions that we propose to consider, can stimulate the AI research community to define, try and evaluate new methodologies, frameworks, and evaluation metrics, in the spirit of achieving a better understanding of both human and machine intelligence.

Nature & Biologically …

Evidence suggests that metacognition-the ability to monitor the cognitive processes and regulate them-exists not only in humans but also in some animals. In nature, humans and animals use metacognition to self-regulate their learning process. This paper gathers evidence of metacognition in nature from research in various disciplines. It also shows how metacognition can be modeled in artificial systems and how the model is applied in an Air Traffic Control Simulator system.