by Edward Cone
The New York Times has been covering Artificial Intelligence with a passion it usually reserves for real estate you can’t afford and musicals you’ll never see. Good stuff, too, including this article about the race to build an era-defining AI platform – and the limitations of such efforts to date.
As we discussed in this post about the nature of intelligence, artificial and otherwise, AI remains a collection of highly-specialized tools that are very good at one thing but not so much at others.
I asked Harvard Business School professor David Yoffie, trotted out as a skeptic in the Times piece, about the limits that confront platform developers. He said:
The key idea to keep in mind is that AI today falls into two categories: supervised and unsupervised learning. Most successful AI algorithms today are supervised learning, which drives the applications toward particular domains. Over the long-run, the goal is to move towards unsupervised learning, which creates the opportunity for a more independent AI.
That made me wonder about parallels between AI and the human nervous system, a shared platform that evolved over a long period of time and includes several disparate sensory and processing systems that work together and ultimately attained consciousness.
Maybe that’s how it will go for AI, too – one day it’s getting patted on the head by a Harvard professor in the Paper of Record, the next (or one relatively soon after) it has mastered unsupervised learning and unified its multiple specialized intelligences into a powerful multifunction computing platform.
Or, eventually, something beyond that.
Edward Cone heads the technology practice at Oxford Economics