REM: A Battle Anthem of Deep Learning
Courtesy of R.E.M.
Tickle the synthesizer — start the drum — queue the guitar
In 1988, R.E.M. released their Green album, including the song World Leader Pretend. Whatever the band may have intended, I have always viewed this songs as a battle anthem for artificial intelligence and machine learning. My first ringtone came from this song:
Let my machines talk to me.
As you read the lyrics, you can just as easily picture the lone analyst sitting in his cube, raging against his data. There is an inevitable frustration that erupts with each new obstacle and the realization that aspects of his/her model are breaking apart.
It is a ballad that applies most to deep learning, where failure and mistakes provide the necessary learning to make things right. A technique where walls and barrier must be overcome. This includes cognitive bias and other personal ‘defenses’ that often get in an analysts way.
Unstated claims, inconclusive data, retests, and causality are not a big reach here. Data is often weaponized in arguments Techniques like curve fitting are classic areas of weakness. Most good analysts are quite aware of their models flaws & weakness.
In the analytic process, analysis is the system for tearing down walls. Often these walls are the very structures we initially created. Modeling often cycles through periods of destruction (analysis) and construction (synthesis). It is a process of identify our own mistakes and rebuilding our systems.
In the end, we arrange smaller elements creating connections with the aspiration that these connections will create something larger, something more complex, and something insightful. The greatest models create a component harmony. But should they fail, we knock them down and start again.
