AI or artificial intelligence is an increasingly popular buzzword. With figures like Elon Musk weighing in that “A.I. is far more dangerous than nukes.” you could be forgiven for assuming that artificial intelligence is the most important question of our time. Other visionaries like Bill Gates take a much more optimistic stance. Gates is quoted as saying, “AI is just the latest in technologies that allow us to produce a lot more goods and services with less labor.” For an inventor or an innovator, this kind of flashy, fast-paced development can seem particularly attractive. Maybe you’ve been thinking to yourself, “I’d like to get involved in AI”, but it’s worth asking, “Am I too late?”
Markets sometimes exhibit first-mover advantage, meaning that early entrants take over a large section of the market and later entrants are hard-pressed to make up ground. This can occur for a number of reasons. Brand recognition and reputation can lead new consumers over and over again to the same few top competing products; network effects drive consumers to choose products that already have the highest rates of adoption; and economies of scale make it cheaper for large enterprises to produce the same kinds of goods when compared to smaller enterprises and startups. Even if no single firm is able to secure a firm lead, steep barriers to entry can ensure that only huge, established companies can play the game. In the face of these challenges, it’s easy to ask “Why should I bother?” A search across ktMINE’s IP Platform can give insight into whether it’s possible or worthwhile to enter the AI market.
Understanding the Landscape of AI
Because the primary value drivers in big tech tend to be intangible assets, intellectual property (IP) data is a good place to start when you would like to understand the market for new technology, but it isn’t enough to check out a few of Apple’s most recent patents. You need a 360-degree view of IP. For example, ktMINE’s Search App provides access not only to patents but also to news data, royalty rates, trademarks, and patent assignments.
Searching for “artificial intelligence” OR “machine learning” (which are related but not synonymous) and examining the “News” tab within the ktMINE Platform yields the below graph. ktMINE’s news data shows that coverage of machine learning has been growing at an increasing rate over the last two years, which should come as no surprise to anyone who has been following the exploits of Watson and AlphaZero.
Moving over to the patents tab, inventive activity increases in 2016 and 2017. Because patent data tends to lag around 18 months behind, 2018 applications may also exceed what appears on the chart below. What conclusions can we draw about the marketability of new AI technologies from these observations? Based on the hype surrounding AI and the uptick in patent activity, it’s possible that the market for AI is experiencing both rapid innovation and rapid adoption. A 2005 article in the Harvard Business Review suggests that markets like this rarely exhibit first mover advantage. In fact, these markets often exhibit “vintage effects” whereby early market entrants lose ground due to the additional costs of supporting earlier versions of products. Additionally, fast-paced innovation in the AI and machine learning landscape will likely draw probable adopters strongly toward the hot products of the moment, canceling out advantages conferred by networks of consumers that earlier entrants may have tried to build.
The Big Players
One caveat is that a huge firm with lots of money to invest in R&D can still secure the benefits of being a first mover by outspending their competitors. Because big tech seems to dominate the AI news cycle, this may be playing out in the markets for AI technologies. Looking at the top 10 patent applicants over the past five years, we can see that the list is dominated by tech giants like Facebook and Microsoft whose R&D spend last year totaled 7.8 and 12.3 billion respectively.
While it may be unlikely that any single firm could completely dominate the others, they all have huge advantages over the startup or small business. Let’s take Facebook as an example, which according to the New York Times, could predict when you will get married or die based on a 2010 patent application for “Predicting Life Changes of Members of a Social Networking System.”
Assuming that consumers want to know if their friends will be getting married soon (maybe so that they can ration their vacation time) and that you could find a way to engineer around any patents Facebook owns, it would be pretty difficult for you to compete against them. Nevermind that Facebook has pockets far deeper than your average startup or the fact that they have an internal team of elite experts. Facebook is sitting on a treasure trove of your valuable data, and effective machine learning techniques require huge amounts of it to learn trends in the messy free-text that is generated by human users. In June of 2017, Facebook hit 2 billion users. Even if the average user only spent an hour per month on the site, that’s five hundred million man-hours worth of data. The truth is users probably spend close to half an hour per day or 30 billion man-hours per month. A tiny startup could never dream of finding that much data to work with.
A Potential Niche
While research on the cutting edge of AI is probably out of reach to everyone but tech giants, it’s still pretty easy to get your hands dirty with machine learning even for entry level coders. With the proliferation of open data and free libraries like TensorFlow, it is now possible to create machine learning models on hundreds of interesting datasets with only a basic understanding of the underlying math.
The rapid pace of innovation has also left behind underserved markets. Many industries are struggling to incorporate machine learning into their enterprise operations. Others view the whole endeavor as science fiction or something that happens in Silicon Valley, and they dismiss the process altogether. However, there is potential for many jobs in implementation to open up if adoption reaches a tipping point. Furthermore, there is already an untapped value in lightweight models that can be written by non-experts using available free libraries. Algorithms for predicting demand can be carried out on smaller sets of clean, structured data and could benefit almost any company that needs to forecast for hiring or logistics. Given the lack of products and services geared at these non-high tech firms, these gaps are likely avenues for an entrepreneur to capture some of that untapped value.
Still, your time might be better spent trying to snag a position at one of the tech giants. According to the New York Times, AI specialists, even those without a Ph.D. can earn more than $300,000 a year.
The Sword of Damocles
One final risk to consider is the chance a future AI patent war could tear all your entrepreneurial dreams asunder. Some are speculating that the current patenting behavior of large tech firms resembles the era leading up to the smartphone wars. If this trend continues it will be relatively easy to run afoul of patented technologies. According to the ktMINE Patents App, there have been more than 10,000 applications filed since 2015. Google even owns a patent on an industry standard algorithm, which is built into its own open source library and many have likely been using it without knowing that it is patented.
Google Patent US9406017
While companies have expressed a desire to encourage innovation by making much of their research publicly available, this is likely a response to the extreme shortage of qualified experts. If the talent pool catches up with demand, firms may begin to take a less benevolent point of view. If you choose to enter the market for AI as a competitor, you do so at your own risk.