Explore the evolution of Artificial Intelligence from theory-driven to data-driven approaches in this thought-provoking conference talk. Delve into the history of AI, starting with small data and rich semantic theories, and trace its progression to modern data-driven methods. Examine the challenges of knowledge acquisition and the shift towards statistical learning techniques. Learn about the development of Watson, its success in factoid question-answering, and its descendant WatsonPaths. Consider the limitations of current AI systems in reasoning, dialogue, and understanding meaning. Discover the speaker's vision for advancing AI by combining human cognition, massive data, and logical theory formation. Gain insights into the importance of bootstrapping collaboration between humans and machines to enable fluent engagement with logic, language, and learning. Understand the potential for machines to learn how to learn and ultimately fulfill the promise of AI.
AI: A Return to Meaning - From Theory-Driven to Data-Driven Approaches
Overview
Syllabus
Intro
Outline
Artificial Intelligence
MEANING IS NOT EXPLICIT IN LANGUAGE
A practical perspective on Al...
Theory-Driven Beginnings
Data-Driven Success
A Painfully Simple Decision
Galoshes: Theory Version 1.0
Galoshes Data Version 1.0
Galoshes Data Version 1.1
Is Intelligence simply finding the functions that map input data to the output data
Data, Reason and Expected Value
Jeopardyl: A great challenge for advancing Al. Specifically in the areas of natural language understanding
What It Takes to Win!
Decompose and Synthesize
Multi-Step Challenges
The Jeopardy Contest: Human vs. Machine
Reflections on Watson
We need machines to go beyond the words and their patterns. We need them to UNDERSTAND.
Taught by
ChariotSolutions
