Invited Talks

Carol HamiltonAAAI 2013 · DOI

Introduction

The 2013 Association for the Advancement of Artificial Intelligence (AAAI-13) conference marked a significant pivot in the field’s trajectory. Moving away from the "un-grounded" symbol manipulation that characterized much of early AI research, the invited talks at AAAI-13 showcased a collective move toward practical, real-world applications. This transition, often referred to as "grounding," involves embedding intelligent systems within physical, social, and economic realities. From tracking global pandemics to revolutionizing clinical rehabilitation and high-stakes auctions, the talks highlighted how AI has begun to address complex, messy, and incomplete-information problems.

Grounded Language and Perception

A central theme of the conference was the limitation of traditional Natural Language Processing (NLP). Raymond Mooney argued that treating language as a set of abstract symbols divorced from sensory experience is insufficient for true intelligence. Instead, Mooney proposed "Grounded Learning," where systems acquire language by interacting with a physical environment—much like a child learns through perception and action. This approach seeks to bridge the gap between linguistic symbols and the physical world, though it faces challenges such as the high demand for laboriously annotated data and the inherent ambiguity of real-world environments.

AI in Global Health and Bioinformatics

The application of machine learning to life sciences was another pillar of the invited sessions. Kristin Bennett detailed the use of AI in combating the tuberculosis (TB) pandemic. By analyzing DNA fingerprints and tracking the genetic lineages of Mycobacterium tuberculosis (Mtb), researchers can now identify drug-resistant outbreaks with greater precision. This work has seen direct implementation by the New York City Health Department, demonstrating AI's role in public health infrastructure. Complementing this, Lawrence Hunter discussed the challenges of high-throughput molecular biology, emphasizing the need for bioinformatics tools that can translate massive datasets into actionable biological insights.

Socially Assistive Robotics (SAR)

Maja Mataric introduced the concept of Socially Assistive Robotics (SAR), a field focused on providing personalized, long-term assistance through social rather than physical interaction. Unlike traditional robotics aimed at physical labor, SAR systems use speech, gestures, and movement imitation to motivate patients recovering from strokes or children with autism. Mataric’s research highlights that "embodiment" is not just about physical capability but about social presence. Clinical validations of these systems have shown promise, though the field continues to grapple with the complexities of long-term user adaptation and modeling human personality.

Strategic Reasoning and Economic Scale

Tuomas Sandholm presented a "dramatic scalability leap" in the realm of game theory and strategic decision-making. In environments characterized by incomplete information—such as auctions, negotiations, and poker—traditional search algorithms often fail. Sandholm’s work on automated abstraction and equilibrium-finding algorithms has enabled AI to solve problems of unprecedented scale. These advancements have had massive economic implications, powering generalized combinatorial auctions totaling over $60 billion and optimizing the US-wide UNOS kidney exchange. Despite these successes, Sandholm noted that qualitative knowledge and human-like strategic reasoning remain frontiers for future research.

Explainability and Narrative Generation

As AI systems become more integrated into decision-making, the need for transparency grows. Deborah McGuinness addressed the importance of "explanation and trust" in AI, particularly within eScience and health informatics. By utilizing frameworks for provenance and ontology, developers can ensure that smart systems are understandable to their human users. Parallel to this, Larry Birnbaum discussed the transformation of "big data" into human-readable narratives. Through automated narrative generation, systems can now produce millions of stories from complex data streams in business and sports, often indistinguishable from human-written reports.

Conclusion

The AAAI-13 invited talks collectively illustrated an era where AI is no longer confined to theoretical models. By grounding intelligence in physical environments, social interactions, and large-scale economic systems, the field has achieved measurable real-world impact. However, the transition also reveals ongoing challenges: the need for better data efficiency, the complexity of human-robot dynamics, and the imperative for explainable, trustworthy systems. As AI continues to evolve, the insights from 2013 remain foundational to the development of systems that are both highly capable and deeply integrated into the fabric of human society.

Summary updated

BibTeX
@inproceedings{hamilton2013invited,
  author = {Carol Hamilton},
  title = {{Invited Talks}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2013},
  doi = {10.1609/aaai.v27i1.8515},
}
Figure