Published Dec 16, 2022

#88 Dr. WALID SABA - Why machines will never rule the world [UNPLUGGED]

Dr. Walid Saba delves into the intrinsic limitations of AI and the mathematical conundrums hindering the realization of artificial general intelligence, while examining the challenges AI faces in fully understanding human language and its unique intricacies.
Episode Highlights
Machine Learning Street Talk (MLST) logo

Popular Clips

Episode Highlights

  • Complexity

    emphasizes the dynamic nature of AI systems, which continuously evolve and adapt based on their environment. He argues that current mathematics cannot model these complex systems, particularly in the context of language understanding, where context and interpretation are unpredictable 1. Walid recommends the book Machines Will Never Rule The World as a sobering read for those who overestimate AI's capabilities, highlighting the intricate complexity of language and the human mind 2.

    Language itself is a beast that we cannot conquer. So imagine the whole mind and the granular thing. They go through it. I mean, it's complex systems. All the way down or all the way up, if you want.

    ---

    He stresses that while narrow AI can achieve impressive feats, the leap to AGI remains far beyond our current reach.

       

    Math Hurdles

    The mathematical hurdles in modeling intelligent behavior are significant, according to . He acknowledges the argument that current mathematics cannot model the dynamic and adaptive nature of mental processes 3. Walid also discusses Montague's work on semantics, which aimed to create a formal system for language, but even this falls short of modeling the full complexity of human language and cognition 4.

    Montague was trying to prove there is a formal system, an algebraic system, and he used lambda calculus, strongly typed system that I can use to compose language, like I do with arithmetic or calculus or anything.

    ---

    He suggests that a new form of mathematics may be required to truly understand and replicate human intelligence.

       

    Human vs. AI

    Walid contrasts human reasoning capabilities with AI's current abilities, particularly in language and reasoning. He acknowledges the impressive achievements of large language models in mastering syntax and coherence, but notes that true language understanding involves much more 5. He remains optimistic about the potential for AGI, but stresses that it will not be human-like and will require new, creative approaches 6.

    These systems have learned non-trivial aspects of language. Now, how do you interpret that? Where do you take it? What do you conclude from it?

    ---

    Walid calls for a diversification of efforts and resources to explore unconventional ideas and hybrid systems.

Related Episodes