KM AI KMGN course- Session 20. Today I spoke about Lessons Learned in AI projects.
Lessons Learned debriefing sessions are critical in Artificial Intelligence Machine Learning (AI ML) projects. Yes- I am saying it loud and clear. We all know these projects are not similar to anything else we are familiar with. We all know that AI ML technologies and methodologies are still evolving and unstable. The reason, however, why lessons learned processes are critical is much deeper: Choosing the best algorithms to use, cleansing the data, training the machine to produce decisions or recommendations, fine-tuning before production, and monitoring the real-life work of the AI-machine, are all tasks based on ITERATIONS. And, as one of the students said so accurately (and so sadly) - we will go wrong; not only once; many times. and these iterations of try-err-learn-improve without systematic learning are a waste of time, energy, and money. Multiply it by the times we run the loops, and it's clear why systematic learning is crucial.
Understanding typical potential biases of data and humans will help us in this systematic learning process. Human cognitive biases may lead to reliable machine behavior, imitating our mistakes as humans. Data embedded biases, mainly of under-representation or over-representation of populations in the data, may lead us to un-proper models, sometimes even over-fitting ones, that make us happy, but don't do the job.
I like using AAR. However, it is important to AAR in an efficient manner, being concise focusing on one to two manners each debrief and digging in deep. That enables me to be both efficient and effective.
I believe systematic learning is key to organizational learning, key to successful AI projects. Knowledge managers- here we come! This can be one more contribution of ours to AI.
This post was initially published in LinkedIn
Comments