Last Week in AI A narrowing of AI research? with Juan Mateos-Garcia
Dec 12, 2020
Juan Mateos-Garcia, Director of Data Analytics at Nesta, dives into his recent paper on AI research trends. He discusses the limitations of traditional metrics in evaluating AI innovations and raises concerns over environmental impacts. The importance of diversity in research is emphasized, particularly in the face of deep learning's dominance. Mateos-Garcia warns against the stagnation of ideas despite rising publications, advocating for strategic funding to support varied, innovative approaches in AI.
AI Snips
Chapters
Transcript
Episode notes
Overlooking AI Composition
- AI research often uses aggregate measures like paper counts and investment levels, neglecting the diversity of techniques and applications.
- This aggregation obscures potential risks of over-reliance on dominant techniques like deep learning, despite known limitations.
QWERTY Keyboard Analogy
- The QWERTY keyboard, designed to prevent typewriter jams, persists despite not being optimal for typing speed.
- This illustrates how a technology can become dominant due to inertia, even if better alternatives exist.
Combustion Engine Analogy
- The dominance of combustion engines in automobiles, despite early exploration of electric and steam engines, highlights potential regrets.
- Over-reliance on one technology can lead to societal lock-in, making it difficult to switch even when drawbacks become apparent.