Too many Deep Learning researchers overlook the fact that there existed many thinkers from the past who spent considerable energy understanding human cognition. Their unique insights are much more relevant today.
It is also important to realize that DL research demands an experimentalist mindset. Where there's a lot of tinkering and leaderboard chasing is favored over the presentation of unique perspectives. There's a lot of engineering that is involved.
One could also argue that this situation is similar in neuroscience where the innovation in measurement methods is critical.
A lot of science is practiced like its performance art. Research papers are judged on the sophistication of methods used rather than a unique perspective for a difficult question.
Leaderboards are easy because the questions are already served to you on a silver platter. What is hard is the formulation of questions that lead to insight. The first step towards greater insight are good questions. What questions have you discovered lately?
But, we must ask, what makes for a good question? For starters, one cannot frame a good question independent of current knowledge. All questions bridge the gaps with what you know and what you need to discover.
Thus questions are these abstractions from the known to the unknown. Good questions, however, are abstractions that take feasible steps forward to the knowable unknown. The superb questions are the unique ones that are relevant but few are able to realize.
The first question about questions is to understand resolves a question. What makes a good explanation for a question? For that, I channel ideas from @DavidDeutschOxf medium.com/intuitionmachi…
