In a recent New York Times op-ed, Gary Marcus, a professor of psychology and neural science at New York University, commented: “AI is colossally hyped these days, but the dirty little secret is that it still has a long, long way to go.” Though Marcus conceded that “AI is exploding with microdiscoveries,” he remained critical of how AI research is currently funded, and how “progress toward the robustness and flexibility of human cognition remains elusive.”
Kyndi CEO Ryan Welsh sent the following letter to New York Times editors to address Marcus’ critique of AI development. Rather than looking at funding as a core research issue, Welsh argues that targeting only one AI principle is truly limiting potential advancements. He concludes that acknowledging how individual AI paradigms are incomplete can facilitate integrated solutions across paradigms, which in turn will begin moving AI technology progress forward. Read Welsh’s full response below.
To the editor:
I read with great interest the op-ed commentary by Gary Marcus, “Artificial Intelligence is Stuck. Here’s How to Move it Forward,” that ran in the July 29th edition of the NY Times. While there is much to agree with in Mr. Marcus’ comments about Artificial Intelligence, I don’t share his view that AI is stuck because of current approaches to funding AI research. Rather, AI is stuck because researchers and practitioners alike forget the teachings of one of its pioneers, Marvin Minsky. “What magical trick makes us intelligent? The trick is that there is no trick. The power of intelligence stems from our vast diversity, not from any single, perfect principle.”
In that spirit, I would argue that the enormous power and promise of AI is not predicated on perfecting or championing just one principle. The major AI principles, or techniques, can be scored on four competencies: perceiving, learning, abstracting and reasoning. For example, deep learning and other statistical approaches score high on perceiving and learning, but score low on abstracting and reasoning. Expert systems score high on reasoning, but register low on perceiving, learning and abstracting.
Despite the deficiencies of each area of AI, researchers and practitioners tout their chosen focus as the “single, perfect principle.” Until we accept the incompleteness of each paradigm, and seek new ways of seamlessly interoperating between multiple paradigms, we are likely to remain stuck in realizing the vast potential of AI as a historically transformational technology.
Founder & CEO
Image Credit: Jun Cen