Over the last 10 years, I’ve realized that applied research (and more specifically AI) is a double-edged sword. On the one hand, it can help you solve incredibly complex problems, and create an insane Return-On-Investment for companies. But on the other hand, according to a Venturebeat article (based on a Gartner report), only 20% of AI projects ever get deployed to production, and among those, only 40% are profitable! Unfortunately, these figures do not surprise me, as I’ve seen countless algorithmic and machine learning projects fail in my career.

I created this blog in order to analyze the reasons behind such failures, share experiences (mine and others’), and generate discussions that could eventually help maximize the impact of AI research in the real world.

Learning from the Startup World

In this blog, I will often tackle the questions raised above with an innovative angle: adapt lessons from the startup world to the AI conceptual world. Indeed, although we often hear about how startups can leverage AI to improve their business, no one to my knowledge took the opposite angle and explored how to leverage the widely studied and experienced startup world to improve AI’s impact on the world.

This idea may sound very weird at first sight, but anyone who looks closely at those two systems will be stricken by their similarities: a high risk of failure, an uncertain environment, the need for agility and speed, the urge to convince many stakeholders and align their interests, the necessity to find harmony between tech and business, …

Last but not least, the similarities between those 2 worlds particularly resonate with me, since I have been a VC investor and founder in the past, invested myself in a few startups, and generally spend a lot of time discussing and working with entrepreneurs.

Who is this blog for?

At a high level, this blog is for people that - like me - believe that AI’s impact should be much larger than it is today. More specifically, I think 3 types of people can learn from this blog:

Applied researchers like myself, who love reading papers and solving hard technical problems, but who are more motivated by their actual impact than their salary or implementing the latest cool AI paper (who said GAN ;-)?).
Entrepreneurs and leaders, technical or not, on questions such as: “what to look for when recruiting?”, “what kind of ML infrastructure will make my teams most effective?”, “When to know when to stop an AI project”, etc…
Investors or future entrepreneurs, on questions like “what are the problems in the world of AI today”, “who is tackling them and how”, etc…

Terminology

There are many close terminologies in the world of AI, and you will notice that I often use terms such as Artificial Intelligence, Data Science, Algorithms, Applied Research in an interchangeable manner… Some people may tell you that they are very different (I personally don’t think so), and there are many debates about differences between those fields. However, the truth is that in our case, the terminology doesn’t make such a difference, because they all relate to the same problem: how to handle risky projects that can have a huge impact, but whose results are quite unpredictable, and can often also very easily fail. Again, you can see here why I believe startups can teach us a lot on those projects.