My 2019 AI Learnings.....
As we close out 2019, let me share a few fundamentals / thoughts on AI (consider this as random musings coming out of my tired and hyper caffeinated brain after a late night New Year's party).....
1. Designing from First Principles remains a key, AI or no AI. AI is magical, but is not immune to basic and fundamental concepts, such as:
a. Before going down the path of AI, fully understand the business purpose. Determine if it is truly an AI problem.
b. Solid design and work flow remains a key
c. Experimentation in model / algorithm selection is key
d. AI is about experimentation, it is not ‘one and done’. Tuning and iterations are must….
2. Bad data leads to bad predictions, bias in data or model will lead to a biased model
3. It is absolutely worth your time and effort to build a solid data foundation. Clean data, properly labelled (for supervised learning), solid feature engineering, those types of things. Data is supreme, treat it with sanctity.
4. Selection of corporate / enterprise AI projects should be done based on a Venn diagram, with the left set being possible and plausible AI capabilities and the right set being meaningful business objectives. The intersection of your two sets are basically your AI projects.
5. Starting one or a few AI projects does not make you an AI enabled company. AI is a paradigm shift and therefore:
a. Every layer of the org needs to be informed and educated on AI
b. AI is not about a technology, but the right intersection of a technology and a business problem
c. Business analysts are just as important as ML / Data Science engineers
d. Vertical based AI projects (industry specific) and low hanging / standardized horizontal business function focused projects are both needed to kickstart AI
e. Huge emphasis on data standardization needs to happen. Modern Enterprise Data Warehouse is a key foundational thing.
6. Kickstart the AI journey with an impactful lighthouse project, deliver quick, show impact, get buy in from other stakeholders, and then start out broader AI projects. It is a journey, so keep doing and keep improving
7. Along with classic AI roles, don’t forget the role of a good developer. A framework / workflow needs to build around a model to make things happen.
8. When it comes to automation, don’t think of a job, think of the individual tasks that can be automated. Start at the task level and see if automation can help augment the overall workflow.
9. Per Ander Ng, any decision that a human can make in about a second can be automated using AI.
10. AI as it stands today tends to work better on simple concepts (to automate) with lots of data, as opposed to complex concepts with not much data. Humans are better at latter.
11. Decision making in an AI project needs to be pushed down to the ML Engineer / Data Scientist level.
12. Don’t use traditional approaches / thinking for an AI project. Iterations and experimentations are key.
13. Create your AI transformation playbook. This should serve as your methodology to do down the AI path.
14. Remember, AI can be fooled / hacked. GANs (Generative Adversarial Networks) can manipulate pixels with minor perturbations, and with confidence call a cat an ostrich.
15. Keep learning. This is an amazing field and ever changing… Follow industry giants such as Andrew Ng, Lex Fridman and others. Commit to reading and listening every single word coming out of them. (I recommend Lex’s AI podcast series).
16. Be humble. You will never know everything about AI / ML / DL. Things change, keep an open mind and more important a learning attitude.
17. Watch the movie Her. (I have not, but I plan on).
18. For inspiration, listen to 'If', a poem by Rudyard Kipling. (Recited by Lex Fridman, one of my favorite AI guys).
19. Get amazed by the Tesla Autopilot and other AI driven optimizations. AI has arrived, for transportation. I recommend: Andrej Karpathy on PyTorch at Tesla, and Elon Musk Autopilot interviews with Lex at https://www.youtube.com/watch?v=dEv99vxKjVI and https://www.youtube.com/watch?v=smK9dgdTl40
20. Last but not least, use AI ethically, to do more good than harm, to augment and not to replace.
In the comments section, please let me know your top 3 (or 5) that you liked from the list above and why, and what else did I miss?