Friday, July 21, 2017

Academic versus industry research

Apple just launched the Machine Learning Journal for Apple researchers to release their work to the broader public. They've published one write-up on improving synthetic images for supervised training of neural nets. This comes after Apple has started allowing researchers to publish academic papers. (The write-up on seems to be the same submitted academic paper, only written for a wider audience.)

As an Apple fan with a Ph.D. in the sciences, I'm excited for the glimpse into what Apple is pursuing in artificial intelligence and machine learning. However, after a decade working in the software industry, I'm reminded that industry research and academic research are related, but not the same.

Industry is driven to create value for a few different individual stakeholders, including customers and shareholders. Because those groups are fickle, industry needs to continue innovating, or they risk not getting the next paycheck (or investment). The companies that don't worry about customers and shareholders aren't around anymore. But as described in The Innovator's Dilemma, industry is often incremental, and has trouble making huge leaps.

Whereas academics favor knowledge over delivering value. Individual academics choose autonomy and intellectual curiosity over compensation. (Take a look at salaries in academics versus industry, for positions with roughly the same skill sets, for this stark difference.) To go further, I would posit that academic research is successful specifically because it doesn't have such a focused pursuit as industry.  Research is often a random walk, and to keep it open allows more unexpected results. However, academic research sucks at delivering actual value to the masses. Putting a paper on arXiv so that 17 people can read it is not changing the world.

What I love is seeing industry work in tandem with academic research. So that we can have the momentous discoveries that only come from the random walk of academic research, with the specific payout to customers that only comes from industry. Maxwell discovers the equations of electromagnetism, and Apple uses them to make sure I can watch cat videos on a train.

Apple, unlike most companies, has the scale to support a fully academic research team. (It could fund the entirety of MIT for ~96 years.) Other companies in the past have also supported academic research teams (IBM and AT&T come to mind, with spectacular results.) So, Apple probably doesn't have to worry so much about this dichotomy between academic and industry research. It seems like everyone will be a little happier, though, if they recognize (and capitalize on) the different motivations of different researchers. 

No comments: