AIPLA: Being smarter about AI

The role of artificial intelligence (AI) innovations and technologies in business and in the global economy has been a hot topic. This is not surprising. AI-related innovations, like automation and machine learning, have become powerful tools ushering in revolutionary changes in the way people live and work.

Gartner, the global research and advisory firm, calculated in a 2019 press release that in 2021, AI augmentation of traditional jobs will create trillions of US dollars of business value and save billions of person-hours globally, and analysis by McKinsey in its 2018 discussion paper, “Notes from the AI frontier: Modeling the impact of AI on the world economy”, estimated that AI-based data analytics could add around 16% to annual global GDP by 2030.

In order to realise this potential, the challenges associated with AI development need to be addressed. Specifically, as AI becomes more ubiquitous, it will raise an increasing number of questions about the implications for IP policy and law. One of American Intellectual Property Law Association’s (AIPLA) areas of focus for 2020 and beyond is to continue to serve as a leading voice on IP issues raised by the ongoing development of AI, on both the national and the international stages.

In the past several years, the association has laid the groundwork for prudent AI policy by hosting events focused on AI with international IP associations and taking part in conversations with, and providing comments to, the US Patent & Trademark Office, the US Copyright Office, and the World Intellectual Property Organization, among others. In addition, we have provided a number of educational programmes on the IP implications of AI, including a standalone seminar as part of our frequent conferences.

AIPLA is actively seeking to address the following issues in the development of IP policy on AI.

Data

One of the most pressing issues in the development of IP policy related to AI that AIPLA is seeking to address concerns data.

In recent years, the amount of data created, recorded, collected and used all around the world has exploded across very diverse fields (eg, automotive, health, building, banking, marketing, etc).

This phenomenon is due, in large part, to the development and diffusion of technologies that record and process data (eg, sensors, computers) and of electronic communication platforms.

"As AI becomes more ubiquitous, it will raise an increasing number of questions about the implications for IP policy and law."

This massive accumulation of data has, in turn, driven the development of AI and machine learning, because data functions as the raw material for training AI systems through machine learning.

Protection of data

The increasing number of modern data applications gives rise to legal questions about the protection of data, in terms of protecting one’s own rights and investments and avoiding infringement of others’ rights.

For instance, training data for AI can yield extremely valuable AI applications, but often rely on pre-existing data that may be difficult to square with existing copyright limitations and exceptions.

At the moment, there is uncertainty about rights in data, because in most jurisdictions around the world the law does not give clear answers to the following questions: who owns the data (mere data and database)? Who can access the data? Can protected data be used for AI training (eg, health data)? Could the risk of infringement claims or potential violation of privacy rights force companies without clear access to good data to use low quality data, thereby contributing to algorithmic bias?

Rights on mere data

Many industries have been identified as big-data intensive. The automotive industry is one for
which the question of the rights on mere data is squarely raised. For instance, automotive manufacturers assemble each vehicle from many components acquired from subcontractors. These components include data sensors, recorders, and communication units.

Who is the owner of the data produced during vehicle operation? The subcontractor, the automotive manufacturer, the owner of the vehicle (eg, a leaser or employer), or the end user?

In most jurisdictions, questions remain whether mere data are protectable by IP rights other than by maintaining the data as trade secrets. However, as one example, the data from millions of cars, taken together, have immense value.

Control of and access to mere data are often dictated by contract. In most cases, such data are subject to confidentiality obligations, thereby preventing access to that data by competitors and
the public. Particularly in industries with a small number of large players, this may result in an “information monopoly” providing significant economic advantage to a few.

This raises the question of whether control, access, and use of mere data should be subject to a specific legal regime, eg, a new sui generis right, with certain prerogatives for the owner along with specific exceptions and limitations to the monopoly.

Databases

The collection, compilation, and organisation of mere data into a database suitable for training an AI model requires skill and effort and can be extremely time-consuming and expensive.

This raises two issues: (i) the protection of the resulting database by IP rights; and (ii) the access to and use of the mere data stored in the database. When a database is used to train an AI system, the result (eg, invention, work, etc) may be protected by a form of IP, depending on the specific circumstances and jurisdiction.

The question remains whether the owner of the mere data and/or the database could claim protection under IP rights (other than trade secrets) for the resulting output of the AI process.

Copyright issues

Another issue to consider is whether a work produced by an AI algorithm or process, without the intervention of a natural person contributing expression to the resulting work, should qualify as a work of authorship protectable under copyright law.

In some jurisdictions, copyright protection has been afforded to the resulting work under the theory that a natural person was involved in programming the AI algorithm, while in other jurisdictions
the resulting work would be deemed an “authorless” work that is not protectable by copyright
because no human being is involved in generating resulting work.

In those latter jurisdictions, it is less clear how work produced through the combined efforts of a natural person and AI will be treated.

"Training data for AI Can yield extremely valuable AI applications, but often rely on pre-existing data that may be difficult to square with existing copyright limitations and exceptions."

It is an open question how copyright law will address the extent to which AI can learn its functions by ingesting large volumes of copyrighted material. Generally, during the AI learning phase, an AI system relies on copious input information or training data to “learn”. This is part of the necessary development process before the AI system can be expected to produce any valuable outputs.

In an AI system configured to recognise images, for example, the training materials may include a large amount of copyrighted works, which may be (and typically are) obtained from the internet.

Patent issues

Patent issues relating to subject matter eligibility remain for various types of inventions, including those involving AI. What kinds of disclosure meets the enablement and written description requirements for patentability could also be a challenging question, given the reliance on big data as training materials.

Moreover, along the lines of copyright ownership, the questions of inventorship and ownership of inventions involving AI need to be addressed.

Summary

AI is already having profound implications on many aspects of society, raising a multitude of questions on how to balance effective oversight of this emerging technology with still incentivising innovation. At the moment, different regimes appear to allow different forms of protection to achieve different goals based upon different economic rationales.

This article highlights what AIPLA considers to be some of the important questions currently in the debate on AI policy. AIPLA believes that policymakers need to continue engaging with the AI and IP communities to better understand the myriad of IP policy issues, including those highlighted here, and address them in ways to ensure that AI has a positive social impact.

Barbara Fiacco is president of the American Intellectual Property Law Association and co-chair of the patent litigation practice at Foley Hoag. She can be contacted at: bfiacco@foleyhoag.com