Thursday, August 18, 2016

Managing Patent Rights in the Age of Artificial Intelligence

, Legaltech News


Artificial intelligence and machine learning (collectively referred to as “AI”) are being used to solve complex problems across an ever-increasing variety of industries, emerging in everything from robotics, autonomous vehicles, and consumer devices to health and pharmaceutical technologies. 


Much of the AI in use today is referred to as “soft” AI, where the AI uses computational intelligence to analyze relevant data and attempt to solve a specific problem. As the technology advances, however, the AI will become increasingly able to mimic thought processes of the human mind.

In addition to solving specific problems, AI will be able to improve decision making, perform an ever-increasing variety of tasks, and even develop new products and processes. Eventually, autonomous systems will be able to operate without significant guidance or instruction. Questions abound as to who is then responsible for the results of those actions. While tort and liability issues are at the forefront of people’s minds, questions also will need to be answered as to how these actions impact intellectual property rights.

Who has rights to an AI-developed invention?

U.S. patent law defines the inventor as “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.” Similarly, Congress has stated the Patent Act is intended to “include anything under the sun that is made by man.” As existing laws and precedent appear to rule out a machine or program as an inventor, the question of ownership of AI inventions could leave the many contributors to the AI tussling for rights.

In a relatively simple Example A, a company might develop its own AI software, execute the software on its own servers, and train the software using its own data. As with a laboratory setting, one could argue that the company owns any inventions discovered as a result of the AI algorithm learning and operating in that contained environment. An interesting legal question arises, however, as to who should be listed as an inventor. 

Should it be the person who “discovered” the invention generated as a result of the process? The person who developed the AI? The person who selected and provided the training data? In these situations all employees or contractors might be under an obligation to assign the invention to the company, but there is an interesting legal question as to who actually invented the patentable subject matter. And if none of these people qualify as an inventor, without a change in definition does anyone have the rights to a patent for that invention?

The situation becomes more complicated when there are multiple parties involved. Consider Example B, where a first company, the “Developer,” develops an AI system that it sells to a second company, the “Owner.” The Owner operates that AI on resources owned by a third company, the “Operator,” where those resources may include servers in a cloud computing environment. The Owner might also obtain data from yet another company, the “Teacher,” where the obtained data is used to train the AI. After training, the AI produces an invention.

Who is the inventor in this situation and which, if any, of these companies has ownership rights? To qualify as at least a joint inventor, one must contribute to the conception of the claimed invention in a way that is not insignificant in quality when the contribution is measured against the dimension of the full invention. Arguments could be made that, at least in the setting where the AI operates in a contained environment with specific training data, the Owner who owns and controls the AI should be at least a co-inventor. An argument could also be made that the Teacher, who provided the training material, contributed in a not insignificant way so as to qualify as a co-inventor. Might the Developer and Operator also have an argument if the AI they developed or operate creates an invention as a result of their activities? Or as a result of scanning social media?

Things get murkier in Example C where the AI is able to leave, or operate outside, a contained environment and learn from things that are outside the control of any of the individual companies. What if the AI system is an autonomous robot that is able to leave a confined setting and maneuver in a public setting? What if the information gained from being out in public provides some, most, or all of the training data that results in the invention? Even if not outside a contained location, what if the AI crawls the Internet for training data that comes from multiple different sources?

The Teacher, who provided a significant part of the training data, might have a claim of inventorship if it can be proven that the training data was a sufficient contribution to the conception. 

The Owner might also have a claim as the entity that set the AI in motion with particular guidelines or training. 

If the initial training did not lead to the invention, and the training material from the Teacher turns out to be relatively insignificant to the eventual conception, however, does either have a claim of inventorship? 

If the AI learns on its own from being in a public setting where no individual action is significant with respect to the creation of the invention, who is the inventor? 

In such a situation, where the conception could be said to have been “crowd sourced,” should the invention be owned by anybody? Should such an invention be owned by the public? If Company B brings the AI back onsite every evening for a data dump and analysis, and makes adjustments based thereon, is that enough to make a claim of inventorship?

And what about infringement?

Stephen Hawking stated: “The short-term impact of AI depends on who controls it; the long-term impact depends on whether it can be controlled at all.” As with inventorship, existing laws and precedent appear to rule out a machine or program as infringer. Infringement arises when, “whoever without authority makes, offers to sell, or sells any patented invention.” As the AI learns and modifies its behavior, it is possible that a resulting product, process, or action might infringe one or more patent claims.

Going back to the examples presented above, consider the same situations except that instead of producing an invention, the AI performs an action or produces a product that infringes someone else’s patent. In Example A where everything was developed, provided, or performed by a single company, it seems as if the company should be able to be held accountable for direct infringement, although in this case the AI would arguably be the direct infringer.

For the situation of Example B it seems, given existing law and precedent, that the owner of the AI might be the most obvious target regardless of whether the company had any knowledge of the infringing activity performed on the Operator site. An argument could be made that the Operator is the entity performing the method or producing the infringing item, and thus might also be considered to be an infringer.

An interesting wrinkle involves potential accusations of induced infringement. Under current law, the induced infringer is “whoever actively induces infringement of a patent.” The Federal Circuit has interpreted this to mean the alleged inducer must have knowingly aided another’s direct infringement of a patent. If the Operator is operating the AI under instructions of the Owner, then the Owner might at least be on the hook for inducement.

An argument could also be made, however, that the Teacher who provided the training data induced infringement if the Teacher can be said to have knowingly aided the AI in infringing the patent through the training data or material that was provided. A similar argument could be made for the Developer if the Developer instructed use of the AI in such a way that the Developer knowingly aided the AI in infringing the patent.

What about the situation of Example C? Here, the AI is off on its own and is not operating on the premises of the Operator. Does the Owner now become the direct infringer? This seems like an odd conclusion, as the person responsible for infringement by the AI should not change based on location. 

The issue might turn on the amount of control of operation between the Owner and the Operator. Again, however, if the AI learns from the public or from a variety of publicly accessible sources, is anyone a direct infringer? Should the owner of the AI always be on the hook? If a third party on the street teaches the AI to do something that infringes a patent, should the owner still be responsible? Strict liability for infringement by an autonomous robot seems like the wrong result, but is it? If you are a company developing an autonomous vehicle or robot, or even software that can run anywhere across a network, how do you safeguard against this?

Navigating the Unknown

For now, there are more questions than answers in this new frontier. Moving forward, companies need to think about how these unanswered questions might impact their business, and determine whether to modify their procedures accordingly. What types of new agreements are needed to ensure that inventions resulting from AI are owned by a specific entity? How should joint development agreements be modified to ensure ownership of technology developed by AI? Should access to specific data that could be used by AI be restricted? Do new indemnification agreements need to be put into place? If user data is being used to train the AI, do user agreements or license agreements need to be modified? Time will tell as to how the interpretations of inventors and infringers will change as the result of AI.

If only there were an algorithm trained to solve this complex problem.


No comments:

Post a Comment