Why AI Product Managers and Product Leaders should learn to train neural networks
I’ve recently returned from San Francisco, where I’ve been working for the last year. Even before leaving my role there, I decided that I would take some time out - before find a new job or starting a new project - so I could study the technical aspects of AI. While I have previously completed a number of technical AI courses, I’ve been wanting to ‘do it properly’, get my hand dirty and gain a more practical understanding of it all. I want to be able to come up with novel ideas, train neural networks, integrate them into simple applications and then deploy them to the cloud for other people to play with.
When I’ve spoken about this plan, a lot of people have been excited for me / envious, but I haven’t always got a positive response. Amongst the objections I’ve heard are:
- “You have to think about whether it offers a good ROI. It would be much better to focus on learning and writing about the business of AI.”
- “You’ve been working as a Head of Product. Assuming you’re planning to continue along this path, why would you need to do this? It might be useful for a normal Product Manager but a Head of Product doesn’t need to get involved in these details.”
I’m writing this article to justify (mainly to myself) why I believe this is a very valuable way to spend my time. Some of these reasons are specific to me, but many will be relevant for other Product Managers or product leaders who want to work on AI products.
Improve your credibility when applying to AI-focused companies
It goes without saying that the job market has become tougher for candidates ever since the Covid-19 pandemic began. If you’re looking for something specific (e.g. AI Product roles) there’s even less going around. For this reason it makes sense to do everything you can to make yourself stand out to companies hiring the roles you want to go for.
I was previously Head of Product at Jukedeck (an AI-generated music startup, acquired by Bytedance) and this should give me some credibility when applying for AI product roles. Regardless, my priority right now is to build on this further.
If you’re a product professional and want to boost your credibility with AI, there are three ways of going about this:
- Get more career experience as an AI Product Manager or product leader;
- Try to become a thought leader (of sorts) in AI Products;
- Demonstrate that you have a strong understanding of technology of AI.
My plan is to work all three, but the third option is my short-term priority.
Communicate better with ML Scientists / Engineers
Product professionals often spend time learning basic technical skills to be able to communicate better with their technical colleagues. And most AI Product Managers I’ve come across have, like me, done the odd technical machine learning course.
When I was working at Jukedeck I worked very closely with the Machine Learning Researchers and Engineers. Unsurprisingly I found that the more time I spent on technical AI courses, the more I was able to:
- Better understand what they were talking about;
- Understand how to ask them better questions;
- Better grasp the challenges and frustrations they faced;
- Better understand when to encourage or challenge them.
Despite this, my experience at Jukedeck taught me that a high-level understanding of AI technology isn’t enough. To really excel in this space, you need to have an in-depth understanding of the limitations, capabilities and considerations of the large number of threads that make up the rich tapestry of machine learning.
This is important for people working directly with technical machine learning professionals (e.g. AI Product Managers) and it is also important for those who want to lead them effectively (Heads of Product / VPs of Product / Startup CEOs).
Understand what the future has in store
In his TED talk “The single biggest reason why start-ups succeed” Bill Gross concludes that the most important factor for success is timing. ‘Timing’ here refers both to market and technological timing: Good market timing is if you launch a product when the market is most receptive to it (e.g. Airbnb launching during a recession when people needed extra money). Good technological timing is if you launch a much-needed product when technological developments have just made it feasible (e.g. YouTube launching soon after Adobe Flash became available and just when US broadband penetration crossed the 50% mark).
If you’re truly literate with neural networks & machine learning, you’ll be in a much better position to understand how the technology is evolving and which recent developments are ripe to be exploited within your particular vertical.
Be able to build your own AI MVPs
I’ve always been drawn to the idea of running my own startup (or having a successful side-project). Given my fascination with AI, I’d want this to be at the core of what I’m doing.
If you have these aspirations and you get to the stage where you want to build an MVP, there are three main approaches:
- Pay someone to build it - expensive;
- Find a keen, talented and reliable technical co-founder - very difficult;
- Build it yourself - requires you to have some technical skills.
Learning to train your own neural networks, at the very least, opens up the third option to you.
Last-but-not-least: Building and deploying your own machine learning applications is actually pretty fun. If you’re the type of person who wants to be an AI Product Manager, then you’ll probably find it rewarding to be able train models based on ideas you have in your head.