“Enormous levels of investment are pouring into this technology. The achievements we have seen so far will surely pale against what the coming decades will bring.”
A Look at the Role of Intellectual Property in the Growing AI Space
Who is investing in AI technology and why now?
Artificial intelligence is getting a lot of attention right now. It was one of the key technological themes at the 2016 World Economic Forum and in June of last year, Andrew Ng, chief scientist at the Chinese web services company Baidu called “AI the new Electricity.” As recently as Dec 5, 2016, Google and Elon Musk opened their AI platforms to the public, Uber launched Uber AI Lab and Apple announced that for the first time it will publish their AI research. There is significant interest in the future applications for AI and there are a lot of interwoven technologies advancing alongside like robotics, virtual reality, autonomous vehicles, blockchain, 3-D printing and the IoT. There is fierce competition for ownership and leadership amongst the top companies which is helping to push AI innovations forward and accelerate advancements in its current and future applications.
In 2015, more than $2 billion was invested in 322 companies with AI-like technology. AI has attracted more than $17 billion in investments since 2009. According to a June 2016 TechSci report, in the US alone, the artificial intelligence market is projected to grow at a compound annual growth rate (CAGR) of 75% between 2016 and 2021. Google, IBM, Intel, Baidu, Samsung, Apple, and other technology giants are all racing to acquire AI startups to add to their business portfolios. Industrial giants like Boeing and GE, who recently purchased a handful of AI startups including Wise.io., also see a huge benefit in applying machine learning AI to improve their industry insights through data discovery.
Google’s head of machine learning, John Giannandrea, is calling this the AI Spring in contrast to the AI Winter of the 70s. As research and general interests grow, there is an urgency from leading technology firms to create the best portfolio of products and the underlying platform future AI innovations will be built on. There may still be a chasm between the anticipated future of AI and its current potential, but these investments show the huge prices companies are willing to pay to acquire these technologies and to scoop up relatively scarce AI talent to gain market share.
What do we mean by AI?
Broadly speaking, AI refers to the development of computers that are able to do things normally requiring human intelligence without human intervention. Strong AI refers to creating computers that genuinely simulate human reasoning and can change and adapt. Weak or narrow AI refers to AI designed for specific tasks within a specific program. The AI that is most commonly referenced today (the AI startups that are getting snatched up) would be considered narrow and is achievable thanks to advances in Deep learning, a form of machine learning in which neural networks are fed information that is used by a computer to make decisions and train itself and adjust based on what it has learned within specific parameters. Deep learning is what is currently propelling AI forward and where major investments are happening.
Machine learning AI, as we know it today, is possible thanks to rapid advancements in big data, neural networks, parallel processing, and cloud technology. AI technologies are just now becoming “mainstream” because the hardware and processing technology has caught up with the vision. The critical mass of data needed to “teach” computers exists now and the storage and processing power to execute deep learning are available, fast, and cost-effective.
What can we learn about the business side of AI from analysing patent data?
Based on this race to innovate and acquire startups and talent to dominate the market, it is interesting to look at the patent filing data in this space to better understand who the big and small players are, what kind of patents they are filing, where they are being filed and how that is all progressing over time. Newer technologies grow out of older ones and innovative ideas for applying specific technologies pop up in seemingly unrelated business areas. All of these questions help move the conversation from what the technology is, to what can be done with it and who is actually doing it.
To better understand the innovation landscape for AI, we did a patent search based on terminology and International Patent Classification (IPC) codes associated with neural networks and deep learning since these are the technologies dominating the growth in what is today being referred to as AI.
The chart below (figure 1) illustrates the number of patent families filed relating to these areas since 1965. The first major period of growth in this area begins in the 1980s, an important time in both growth of computer hardware and chip technology, as well as a period of renewed public interest in the potential of AI. While the level of filing dipped during the 1990s, it has since returned to a pronounced period of growth, in particular within the last five years, likely due to the widespread availability of cloud infrastructure.
Figure 1: Patent Filing Rate
Which companies are filing the most AI patents?
After looking at the rate of innovation, we looked at who the leading innovators are. Based on this data, you see that companies with established AI-related programs, such as Qualcomm’s neuromorphic chips, Microsoft’s AI research group, Siemens’ SENN and the most high profile, IBM’s Watson have filed the most “AI” related patents as seen in figure 2 below.
Figure 2: Top AI Patent Filers Between 1965 and 2016
How does that compare to the overall amount of AI patents being filed?
Looking at AI filings as a whole, and not just who is filing the most, but how many have actually been filed, you see a slightly different picture. The innovation landscape is more diverse than would probably be assumed. The number of unique assignees filing per year (figure 3) has continued to grow to the point where in 2016 over 1600 different entities filed AI-related patents, four times the amount of 20 years ago. By comparing the field of operation, the speed of filing and citation networks it would be possible to identify patents and assignees outside of the “usual suspects” that could have a significant impact on the market.
Figure 3: Amount of Unique AI Patent Assignees
Much of this newer research is coming from academic sources, with 40% of assignees filing in 2016 classified as either a university, college or research institute (figure 4). With such a significant number of patents coming from this area, a number that has been growing consistently, the importance of tech transfer to fully realise the value of this pool of IP is clear. Meanwhile, the number of patents filed overall per year by the 50 most common assignees has remained relatively constant, with a much wider field of assignees now filing relevant patents. It’s no longer a case of being able to identify a few key players, and a much broader landscape needs to be considered.
Figure 4: Number of AI Patent Filings by Top 50 Companies Compared to Academia and Other
Where is AI technology being applied?
Examining filings made over the last five years (LFY) shows the areas in which AI is currently having the most impact, all of the categories in figure 5 have had over 25% of their AI related filings made in the last five years. Computing, data processing, and sensors make up the bulk of the patents as may be expected, but outside of these, the areas range from agriculture to nanotechnology. The applications of these technologies can be very broad, with a category such as sensors at the forefront of AI research impacting everything from autonomous vehicles and traffic congestion to border and airport security.
Figure 5: Some of the Growth Areas for AI
Gaming – an example of what is old may be new
Gaming, one of the few categories to have AI related filings dating back over 30 years, has been an important part of the evolution of AI technology and continues to be an integral part of its future. AI winning at Checkers, Chess, and Go are all well-cited examples of computers beating humans at games that require strategic thinking, but video games go a step further with expanding the application of AI and creating algorithms for AI that respond to and anticipate the player’s movements as well as strategic actions. Also, the platforms both Google and Elon Musk just released are gaming platforms. Deepmind has open-sourced its game code and Musk’s OpenAI is a computer training ground that helps software play games. Therefore, gaming was a logical place to run a patent search.
By looking at early patents in gaming applications, we are able to trace a line to more modern applications of AI, in particular, those with more commercial applicability. In doing so, we can highlight the fact that research in “model” areas brings value once the supporting technology provides an applicable infrastructure for their use as long as the patents defined are broad enough. In particular, we would expect that areas such as manufacturing, advertising, and navigation/auto have been influenced by these factors, with gaming itself now also being a significant market where this technology is key.
So, what does an Atari patent for simulating vehicle behavior for multidimensional video games have to do with electrophonic musical instruments, text to speech technology, speech synthesis, and speech processing using neural networks? The AI technology described in that Atari patent is highly cited and has helped shape innovations in those other seemingly unrelated areas. This particular patent was selected because it represents the very early stages of gaming AI, in this case, a process to simulate the behaviour of opponents in a racing game, (likely a game in the Sprint series, the first game to include CPU controlled cars – http://hackaday.com/2016/04/28/forty-year-old-arcade-game-reveals-secrets-of-robot-path-planning/ (fast forward to 11 minutes for the patent reference).
Following the trajectory of this single vehicle behavior patent over the past 36 years and looking at its citations, we can examine the impact of this single filing on subsequent technology developments, as any future citations (classified X and Y) indicate a direct link to prior art. In theory, this could be expanded to cover a company’s portfolio, or to look at a wider sphere of influence by looking at ALL citations, however, as a simple example, this single patent was chosen.
For each stage of citation, any X or Y citing patents of the previous stage were added to the examined dataset, with 9 stages and ~250 patents eventually captured as direct “descendants” of the original patent. (For a deeper explanation of our process see footnote below or contact us)
A selection of these technologies is presented two ways below. The first is an interactive timeline graphic allowing you to drag the tab along the timeline and see at what point different offshoots started taking shape. The second is an infographic illustrating in one image the various offshoots of the original Atari patent and the years at which they started to sprout. These visual representations show the range of impact this one AI-related patent has had in the 36 years since it was filed and how quickly its sphere of influence has grown since the mid-90s.
What does the IP data show us?
As exemplified by the Atari patent and all its “children”, key patents for AI may not necessarily be those with the most direct link to current market applications. This will be important information during strategic patent acquisitions and investment due-diligence. However, if the market applications don’t get commercial traction soon, some of the older seminal patents will have expired. So, timing is key for patent value and understanding the market will help dictate what innovations to patent and when. Too late is not good but too early will see the patent expire before the market is mature enough to bring value.
Overall, while the technologies and patents are important, the real reason for giants to acquire startups is to acquire talent as fast as possible rather than build a team from scratch. Seeing how much is coming out of academic institutions, it’s obvious why some of the biggest companies are scooping up top academics in this space.
For each patent in the dataset, the IPC codes of the patents that it cited were analysed and the most common were chosen as a representative of the technology area that had most influenced its development. In reality, the situation is more complex and no single IPC code will have been the only influence, but by restricting it to a single code it is much easier to identify (and visualise) key filing trends.
The IPC codes of the initial patent (“parent codes”) were then linked to any IPC codes (“child codes”) listed as being most directly influenced by them, creating a branching tree of technologies based on areas of influence. For each area the timeframe of filing is also listed, allowing the year in which the initial branch from the parent code was made and at which point that child codes were no longer having a direct influence to be seen. Any future filings in those codes would extend those branches and open them up to influence new technologies.
Other interesting articles to read
- The creator or Atari launches new VR company https://techcrunch.com/2016/10/12/the-creator-of-atari-has-launched-a-new-vr-company-called-modal-vr/
- Why video games are essential for AI http://togelius.blogspot.co.uk/2016/01/why-video-games-are-essential-for.html