• The evaluation function of academia and business are different.
  • It seems to be related to the post just before (Difficult to manage research department), but it’s a completely different trigger, but academia and business have different evaluation functions.
  • In business, it is better to solve a problem with a simple, dead tired technique for ease of maintenance, etc. afterwards, but in academia, that is hardly a thesis!
  • In business, the strategy of “using catchy technology to hit the press and thereby lure customers and talented people” is often used because it is rather profitable.
  • The research for those presses is more convenient for the prerequisites than for the actual business case.
    • For example, a difficult problem (Go) can be done by deep reinforcement learning! This does not mean that any other difficult problem can be solved by deep reinforcement learning, but rather that Go has been studied by a large number of people, that a large amount of game record data, the result of that study, is available in a neat format, that simulation is easy because it is a human-made game and not a phenomenon of the natural world, and that a large amount of data can be generated by the simulation of self-games. The simulation is not a natural phenomenon but a game created by humans, so it is easy to simulate, and a large amount of data can be created by simulation of self-game.
  • In business, customer attraction is important. By working on actual projects from customers, the company accumulates know-how in the domain and acts as a “barriers to entry” against other competitors.
  • and publication bias.
    • In academia, “I tried using a complex model to solve a problem, but the learning didn’t work and I was defeated” is not published as a paper.
    • If a simple logistic regression solves a problem that could not be solved, or if it is very accurate, it could be a paper, but it is not very catchy and is not likely to become a topic of conversation.
    • In business, there is a positive disclosure bias for demonstrations using catchy technology, whereas there is a negative disclosure bias for information on how the solution to a real case was implemented to prevent competitors from entering the market.
      • If a customer’s case that came to Watson as an inducement to win a quiz show was solved with a decision tree, such a thing would never make it onto a press release.

https://www.facebook.com/nishiohirokazu/posts/10214062787340259

This page is auto-translated from /nishio/アカデミアとビジネスの評価関数が違う using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.