By Kirsten Rulf, Harvard Kennedy School Autonomous Vehicles Policy Initiative
As I have argued in a previous blog post, patent documents and the ability to analyze them with machine learning have three advantages that directly address some of the information barriers that policy decision makers and investors wrestle with, when it comes to Autonomous Vehicles: Patents provide a measurable indicator for the genuine state and the progress of the technology. A patent document is legally required to describe the actual invention. It is by nature a marketing free zone. The economic value of a patent, for example, as defined in economic theory, is a measurable indicator for the importance of a technology component and simultaneously for the ability of an AV company.
Perhaps the most important example for the advantages of machine learning-driven patent analysis: patent documents clear the fog of anonymity in the AV realm, a critical information barrier for policymakers. In a 2017 Harvard Kennedy School survey, close to 90 percent of a sample of key policymakers in the leading AV test states lamented that they had no access or notion of the actual inventors of the Autonomous Vehicles that cruise their streets. Outside the industry expert circle, next to nothing is known about the talent in AV engineering. Sometimes it is not even clear inside the industry, who the star inventors are, a fact that is also problematic for investors. Policymakers in key test states and municipalities, for example in San Francisco with currently 27 testing companies, claim that the anonymity of the industry is a problem that hinders their decision-making. “We absolutely want to know the person or persons that program the car. After all it eventually makes decisions about life and death, so we want to know about the ethical background and values of the people that write the software”, demanded Kurt J. Myers from PennDot in the 2017 survey. “In the end, we will be held accountable for safety by the public, so we do want to know the engineers on the other side”, adds Jason D. Sharp, Executive Deputy Chief Counsel of the Department of Transportation in Pennsylvania. Besides, all surveyed policymakers expressed that they would prefer a direct contact for technical questions that arise from AV testing, in particular when they are making long-term decisions about infrastructure. “We recently had to make a decision about a big building project that included a bridge. We wanted to make sure that it is built AV friendly, but we could not reach out to any expert, and the management staff of AV companies that are testing in Pennsylvania were unhelpful”, reports Mark C. Kopko from PennDOT.
Patent analysis with machine learning can alleviate some of these concerns. Patents provide the names of the actual engineers (the inventors), their contact information, and their company affiliation. Additionally, the documents give an insight into the leading companies in the field, if we accept that the principal engineers in the field work for the top companies. Lastly, with the right prediction algorithm, patent documents can become a leading (versus a lagging) indicator and predict future technology developments.
For my analysis, I have developed two algorithms that analyze a patent data set of more than 100,000 from the Autonomous Vehicles sector. The data set was generously provided for my final capstone client project at the Harvard Kennedy School by the German Max Planck Institute for Innovation and Competition the startup Octimine Technologies.
Algorithm #1 first finds the economically most valuable patents in Autonomous Vehicle technology. It automatically calculates and then lists these patents by their internationally assigned ‘Family ID’ number. Furthermore, it uncovers what inventions are patented under these ‘Family IDs’, which company has applied for the patent, and also who has invented the patented technology (the engineer or engineering team). Another, more complex algorithm sketches inventor networks, uncovering teams and leading experts.
Below are four sample results and visualizations that were surfaced by algorithm #1 alone. While they are not perfect, they demonstrate the type of analysis and knowledge that machine learning driven patent data analysis can surface.
To begin with, a list that describes the five most valuable patented inventions for AV technology to date. Policymakers emphasized as one concern that they need to make decisions on infrastructure and do not know which AV technology will emerge as the dominant one. This sample list gives clues, for instance on sensor technology.
What are the top 5 most valuable inventions in AV technology?
1) A remote transaction interface system for vehicles, that has a transmitter to modulate signals from an input device and can transmit signals to a remote transaction machine in order to engage in transaction with a transaction unit outside the vehicle
2) A “presentation slots utilization method”: a method that enables a vehicle to “sense” and alert its driver to an unforeseen event.
3) A “torque vectoring device” that increases and decreases the torque of the vehicle independently.
4) A rearview mirror assembly device with microphones.
5) A thick-film paste for electronic devices that forms a conductive path for devices, such as solar panels, heaters, and windshield panels
What is the timeframe from invention to market?
Next, a timeline shows when these inventions where patented. This can be informative for policymakers and investors, and it can serve as a starting point to build a predictive analytics algorithm that can predict which AV technology may evolve when. Furthermore, this figure can more generally be used to gauge the speed of AV technology development by looking at the age of the most valuable inventions and extrapolating.
Which engineers or engineering teams have invented the top 250 most valuable technologies for Autonomous Vehicles?
Another visualization importantly addresses the problem of anonymity in the AV industry. It gives a hint as to which engineers or engineering teams have invented the top 250 most valuable technologies for Autonomous Vehicles.
Which Companies Hold What Percentage of The 250 Most Valuable Patents in AV Technology?
Lastly, both algorithms combined can calculate companies that either a) hold the most valuable patents and can therefore be considered to be eminent in the technology field (Figure below); or b) employ the thought leader engineers in the field and therefore have the greatest talent (not pictured here for reason of propriety).
There are many ways to improve this methodology. However, the examples above show that it for sure has the potential to close information gaps and help shape the AV regulation process. What is more, we can use it as a template for analyzing other emerging technology fields that share the same characteristics, like complexity and speed, that make Autonomous Vehicle technology such a tough field for policymakers.
The Most Valuable Patents and Engineers in Autonomous Vehicles was originally published in Harvard Kennedy School Autonomous Vehicles Policy Initiative on Medium, where people are continuing the conversation by highlighting and responding to this story.