Back in the late 60’s, Edward de Bono introduced a mental tool called “po” as part of his work on lateral thinking and creativity.
Po is a verb, and it gives you permission to make unsupported, illogical, and foolish statements about a problem you are trying to solve. In a sense, it serves like a magic wand that you can wave to jump over constraints and move your thinking forward.
An (imagined) example:
J: We are stuck. We can’t make our laptops any smaller. The DVD drive takes up too much space.
S: Po users don’t need DVD drives.
J: slaps forehead Brilliant! With fast internet and cheap USB thumbdrives, DVDs are obsolete!
And thus was born the Macbook Air.
Po is an incredibly useful tool. Perhaps with a catchier name like “the 80/20 rule” or “the conjoined triangles of success” it might be better known.
This all leads to an observation regarding the recent widespread interest in deep learning. To almost everybody outside of the software development world (and many people inside of it), deep learning is like a magic wand. In a number of conversations about startups and technology, I’ve seen deep learning serve the same purpose as “po”.
I first recognized this during a discussion about a sales CRM. The product team wanted to use deep learning to predict which stage of the sales funnel a perspective customer was in, or when a prospective customer was about to leave, and show it to the salesperson using the software.
After about an hour of strategizing about data and models we finally convinced ourselves that it was possible, and we turned to the discussion of how the application would use the ranking. It turns out that, in all cases, the goal of the ranking was to prod the salesperson toward the same action – call the customer. The real problem was salesperson motivation, a completely different problem that probably didn’t require deep learning.
Without the promise of actually being able to solve the problem through deep learning, the conversation would have ended, and the team never would have realized the deeper truth about the application.
To generalize, humans tend to think linearly, and often get stuck when one of the steps seems impossible. By expanding the domain of what could be done, deep learning helps people see what should be done.
Content © 2006-2021 Rusty Klophaus