Oxford Languages defines “inference” as “a conclusion reached on the basis of evidence and reasoning.” Achieving accurate inferences with a minimum amount of work is of utmost importance in my field of design and analysis of experiments for industrial R&D. So, Amin Vahdat, Google’s Vice President and General Manager of ML, Systems, and Cloud AI, got my full attention by promoting their latest developments in AI as the beginning of the “age of inference” in his April 9th blog on Ironwood —their 7th-generation TPU (tensor processing unit). The blog announces that by “combining the best of Google DeepMind and Google Research with Google Cloud” will “further accelerate scientific breakthroughs, with a mission to become the most capable platform for global research and scientific discovery.”
An April 9 report by VentureBeat offers up an impressive array of statistics on Ironwood with much hyberbolic, high-tech jargon such as “when scaled to 9,216 chips per pod, Ironwood delivers 42.5 exaflops of computing power — dwarfing El Capitan‘s 1.7 exaflops, currently the world’s fastest supercomputer.”
I don’t grasp the units of measure, but it sure sounds great! Perhaps AI will fill in for the ongoing cuts in USA’s funding for institutional research and ripple effects on industrial R&D. I hope so!
However, despite the rapid development of AI, it may be a long while before it gets embraced by researchers. For example, Aidan Toner-Rodgers, an MIT economist, published a paper on Artificial Intelligence, Scientific Discovery and Product Innovation last November that reported 82% of R&D scientists (over 1000 surveyed) being dissatisfied with AI due to it decreasing their creativity and skill utilization.
On a positive note, Toner-Rodgers asserts that the output of “top” researchers nearly doubles due to how they “leverage their domain knowledge to prioritize AI suggestions.” That is the best of both worlds—human intelligence (HI) combined with artificial intelligence (AI).
PS: An April Nature News article summarizes results from a survey of 4,000 researchers that addressed broader questions about AI, not the “what’s in it for me” focus of the Toner-Rodgers’ poll. For example, scientists viewed the glass slightly more than half full for AI whereas nearly all the general public feel it creates more risk than benefit. Interesting!