IJCAI-09 Keynote and Invited Speakers

IJCAI 2009

Abstract The 2009 International Joint Conference on Artificial Intelligence (IJCAI-09) marked a significant era in the field's evolution, characterized by a shift from theoretical abstractions to complex, real-world applications. This collection of abstracts from keynote and invited speakers highlights the integration of AI with economics, ecology, education, and the arts. By examining the "knowledge food chain" of sensor data and the evolutionary origins of language, these talks showcase an interdisciplinary AI landscape focused on solving global-scale challenges through statistical modeling, embodied agents, and affective computing.

Introduction

IJCAI-09 provided a fascinating snapshot of the AI landscape during a period of intense transition. The field was moving beyond "toy problems" to address high-impact issues in diverse domains. The invited talks collectively addressed the overarching challenge of integration: how to move AI out of the laboratory and into messy, real-world environments.

Core Challenges and Problem Domains

The speakers identified several critical areas where AI could provide transformative solutions:

  • Economic Efficiency: Hal Varian explored how computer-mediated transactions enable better incentive alignment and personalization, transforming market dynamics.
  • Personalized Education: Cristina Conati addressed the need for systems that understand a student's emotional and metacognitive state, rather than just their domain knowledge.
  • Environmental Sustainability: Thomas Dietterich discussed the "data-to-policy" pipeline in ecosystem informatics, focusing on the management of noisy sensor data for biosphere policy.
  • The Origins of Language: Luc Steels investigated how communication systems can emerge autonomously through robotic interaction, bridging AI with evolutionary biology.
  • Expressive Music: Gerhard Widmer used AI to decode the subtle nuances of human musical performance, revealing insights missed by traditional musicology.
  • Activity Recognition: Qiang Yang focused on bridging the gap between raw sensor data and high-level human goals, a critical step for pervasive computing.

Methodologies and Data Sources

The methodologies presented define the toolkit of modern AI, emphasizing statistical rigor and interdisciplinary integration:

  • Statistical & Probabilistic Modeling: The use of Linear Gaussian Dynamic Bayesian Networks (DBNs) for sensor data cleaning and statistical inference for activity recognition.
  • Embodied AI: Humanoid robots were utilized to play "language games," simulating linguistic selection and evolution in a physical context.
  • Signal Processing: Musical performances were analyzed in "phase space" and "feature space" to identify unique artistic styles.
  • The Knowledge Food Chain: A metaphor for activity recognition where raw sensor data is "digested" through levels of abstraction—from signal to activity to goal—until it becomes actionable knowledge.

Data sources mentioned across these domains include ubiquitous economic transaction logs, automated arthropod population counts, recordings of master concert artists like Artur Rubinstein, and trajectory data from sensor streams.

Key Insights and Results

The talks yielded several profound insights into the future of the field:

  • AI as an Economic Engine: Computers enable entirely new contractual forms and controlled experimentation in the marketplace.
  • Beyond Knowledge Modeling: Effective AI tutors must adapt to affective states (emotions) and metacognitive abilities (how students think about learning).
  • Linguistic Selection: Language evolves through social "selection" during interactions between autonomous agents, rather than being a fixed code.
  • AI in the Arts: AI has moved past toy problems in music to make novel discoveries about expressive performance, impacting the commercial music market.

Qualitative successes highlighted include the successful deployment of intelligent tutoring systems, the development of automated arthropod counting devices, and the creation of musical intelligence systems capable of visualizing master performance styles.

Limitations and Future Work

Despite these successes, significant challenges remain for the next generation of AI researchers:

  • Modeling Complexity: Extending intelligent tutors to support exploration-based learning and complex human traits.
  • The Sensor-to-Policy Gap: Bridging the spectrum from raw data collection to actual biosphere management policy remains an open problem.
  • The Language Puzzle: Understanding the fundamental origins of language remains a major unsolved problem for both AI and biology.
  • Interpretability: Making AI-driven musical and artistic analysis intuitive and interesting for non-experts.

Conclusion

The IJCAI-09 invited talks underscore a field that has successfully expanded into complex real-world systems. By integrating economics, cognitive science, and environmental informatics, AI researchers in 2009 set the stage for the pervasive and interdisciplinary AI applications that define the modern era.

Summary updated

BibTeX
@inproceedings{2009ijcai,
  title = {{IJCAI-09 Keynote and Invited Speakers}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2009},
}
Figure