The
lead story in today’s EV has some telling data illustrating the genuine explosion in AI. The interesting thing is how this explosion is happening across multiple vectors: the performance of AI systems, the rate with which they are being deployed and the breadth of applications know using them.
In the past 4 years alone, error rates in an image recognition benchmark have dropped from 24% to 4%. Crowdflower reports a 50-fold increase in demand for training data sets in four years. And Google reports that they are now applying deep learning to more than 2,700 different projects within the company - up from fewer than 100 three years ago (27x).
This week Facebook
open-sourced a hardware platform optimised for AI tasks, joining Google and Microsoft in open-source key AI know-how. The rationale is clearly competitive. One way is to establish
de facto standards around methods and processes tied back to these firms. Another is to raise the quality of AI innovation that takes place outside of these firms, and, in turn, improve the quality of potential partnerships or acquisitions.