Two kinds of AI Solutions... and we need both!
There are (at least!) two kinds of of intelligent solutions. A burning business problem that can be solved using a smart / AI application. For example, why customers are leaving, why business is shrinking in certain geos or with certain product segments. In these kind of problems, we start with an issue, of course, work backwards, collecting evidences (data) and analyzing the data for the root causes and remedies. Typical diagnostics whether it is equipment or human body - fall into this category - reacting to an event or issue.
The second category is more pro-active. Even if we don't create a solution, the sky is not going to come down. However, we may be sitting on hidden magical steps and decisions where we can scale up the operations, improve productivity, prevent outages/catastrophes, make our planning more efficient, etc. Examples of such applications are price optimization, auto-analysis of documents, risk predictions, discovery of new materials - all of which need intelligence that stems from data. Here we start from data - sometimes with an ill-defined or vague objective or sometimes explorative. By the very nature, such applications may not be very appealing to industry, because they can still survive without it today. And, only innovative companies will think about tomorrow and day after. So, if the funding is a crunch, acceptability of such innovative AIs will be lesser than the first category of reactive ones where it helps to solve today's problems.
"Succeed with AI doing it backwards" is relevant when we are thinking of immediate benefits. Creative applications will take time to mature and achieve acceptance. Compare Yahoo and Google. Who believed in Google back then. Look at the stories of autonomous driving. Did we really need it? Did it solve any immediate business problem? No. However, now, it became part of all "modern auto software in one form or another" and exponentially influencing the value.
So, in short, we need both these approaches.
- doing backwards by finding solutions to burning problems and resolving the pain points which will be quicker successes
- identifying innovative applications that make life better which often need blind searches for (mining) the unknown jewels in the data, Successes here may not immediate, failures my be more prevalent, however, we need to focus on this category too to advance the technology, to achieve orders of magnitude leaps in productivity, efficiency, and quality.
Glad to discuss this further
