This is the last post written by BT_Bear. Click here to follow him on Twitter for more insights.
Previous Parts
Part 1 - Is it just Hype?
Part 2 - Things to Pay Attention to
Part 3 - Finding Places to Grow
The Detached AI Lab
Preferably at some fancy location away from the rest of the organization with little or no interaction? This setup makes sense if you want to cover some far-out topics that may bring business value in the future. This has been very popular in recent years but is the wrong answer for 90% of the cases. A place to quickly test new ideas with strong involvement from functional teams?
Can make a lot more sense, especially if there is a clear pipeline to develop successful prototypes into full products.
Here’s a nice useful video on AI research:
AI as The Solution For Everything
I would recommend avoiding anybody that sees it this way. For any tool in the world. There should be many cases where the clear recommendation is to go with something different, often a less sexy and more pragmatic solution. There is a playful grey area here if you play with the definition of AI – Imagine you have leadership in front of you that absolutely wants to see innovative technology, but your cases today are better solved with simple linear regression models? You can bend the definitions a bit to fit it, solve the case and make your leaders happy. The world is full of many much worse cases, but please make sure you can still look at yourself in the mirror
Misaligned Incentives
Even good models will only be used if incentives are aligned. Imagine the following situation: You developed a targeting mode that allows a 50% cut in advertising costs without loss in revenue. Brilliant for the company, what will most likely happen next? You have a fair chance to end up in a fierce trench ware with whoever is responsible for the advertising budget. Why? 50% waste makes them look pretty bad plus they will likely loose benefits like conference invites and “entertainment”. IYKYK. This is something you can rarely resolve from a junior level, so better take some time and avoid if possible.
BT_Raptor Note: If you are making a ML model to basically replace some of your human analytics team, DO NOT DO THE PRESENTATION for it. You will be grilled by them, and they will pick apart every single nook and cranny. Instead, pass it off to your boss or something, and say some BS about how he has more knowledge on the subject and can do a better job or something.
If your AI model is legitimately successful, you’ve just proven to the management why the human analytics team should no longer be employed….
Final Remarks
So, this is a nice point to end my musings on the state of Machine learning in the world and some ideas to navigate a career path. I picked points that I believe beginners should be aware of and are typically not – would be curious to hear how this resonates with you. Let me know if you have questions, I may dig a bit deeper on them in the future. Until then, happy training!