Can a company develop an AI-based product that can’t be easily copied?
Advancements in Deep Learning algorithms have emerged at a much more frequent pace in recent years. This creates a situation in which novel improvements become obsolete very quickly. Not only that, these new developments are usually published online as open-source code for anyone to use. Can a company develop an AI-based product that can’t be easily copied?
One of the key components of a startup is the ability to create a moat. The moat is a barrier stopping other companies from developing the same product or service. A good moat is a key competitive advantage that sets a company apart from its competitors. Different types of moats include a technological moat, a branding moat, an operational moat, as well as others. Some moats are better than others but are harder to create.
The messaging application, Whatsapp, has a very good moat. The app benefits from network effects and dominates the smartphone messaging-app space. Competing apps with superior features find it hard to compete with Whatsapp. Everybody gets on Whatsapp because everybody is on Whatsapp. Using a different app will leave you talking to yourself. We usually don’t hear about startups that didn’t have a moat due to survivorship bias. We hear about moats that did their job because the startups that built them are the ones who survived to tell their story. Companies that didn’t have a good moat collapsed and vanished into the startup graveyard.
In Israel (popularly named “startup nation”), companies usually try to develop a technological moat. In other words, to develop sophisticated technology that will be hard or time-consuming to replicate. This is often referred to as a deep tech startup. By the time a competing company develops similar technology, the startup will already have a significant market share and have its technology way ahead. An example of this is the Israeli computer chip company Mellanox that was recently acquired by Nvidia. The giant, Intel, tried to compete with Mellanox for many years but could not create a product of equal performance.
Back to AI. As we discussed in previous posts, the number of startups trying to develop new AI technology is increasing rapidly. Many of these startups are deep tech startups, trying to rely on their novel AI technology in order to create their moat. The difference between the case here and many other tech startups is the pace of developments in the AI community. To demonstrate this, let’s see the performance of different models over AI’s very popular Imagenet challenge. This challenge consists of classifying images into 1000 different categories. Speaking in hindsight, the 2012 win by AlexNet (the first neural network to win) was the event that marked the rise of deep learning. The Imagnet challenge was regarded as a measure for Deep Learning progress for years to come.
Every year since 2012, a new algorithm produced significant improvements in the results. These results made the previous year’s results obsolete. The organizations that create these models write an academic paper describing how they reached these results and how to replicate them. Many times, a code repository with the entire code is released alongside the publication. If the code is not provided, it only takes a few days for enthusiasts of the open-source community to write their own code based on the academic paper. The online code replicates the results and in some cases, even improves them. A company that relied on AlexNet’s great results and developed a product based on it, would find the algorithm obsolete merely one year later!
It is hard to believe that a startup can design an algorithm that will outperform the best of the open-source community when trained on the same data. Does that mean that any deep learning algorithm can be easily copied? Definitely not. Our next post will discuss the best method to defend your algorithm from being copied — the data moat.
Thanks for reading!
— — — — — — — — — — — — — — — — — — — — — — — — — — —
Join my newsletter for updates: The Intersection
Keep in touch on LinkedIn: Noy Shulman
Check us out at The AI School For Leaders
