Big data and AI have revolutionized the manufacturing industry, making factories autonomous as well as smart. With the advent of novel sensors and the Internet of Things, current manufacturing facilities are collecting trillions of bytes of data from almost every tool in a manufacturing operation. AI processes the data for production and throughput optimization, supply chain management, product quality assurance, predictive and/or preventative maintenance, energy reduction and a host of other functions.
Given the complexity of manufacturing operations, one of the biggest challenges is enabling a free flow of information throughout the operational process while simultaneously protecting rights to valuable data and AI-related intellectual property (IP).
The problem is exacerbated because implementing and applying AI is significantly different than that of traditional software systems. As a result, many businesses fail to derive value from data and technology due to a failure to adequately secure and/or enforce their rights.
An AI system is a data-driven iterative process where an algorithm/model is constantly learning and improving.
As visualized above, AI systems use large volumes of training data in ways different from traditional software methods, (e.g., detecting patterns without explicit programming) to create trained AI models capable of outputting previously unknown complex interferences (output) from a received input. The AI model itself is constantly evolving based on the output. It should be noted that the above architecture relies heavily on Big Data that can include the training data, the input data, as well as the output. Such big data requires significant investment for collection, processing and storage, and can independently be a valuable asset for manufacturing companies as both a revenue source through licensing and for generation of AI-driven technologies.
The complexity and constantly evolving nature of the above AI system makes it harder to devise a “one size fits all” IP strategy that conforms to the traditional paradigms of IP protection. For example, a textile manufacturing company may have hundreds of terabytes of training data relating to textile quality and corresponding machine operations (raw or processed data); the AI models generated from that training data (e.g., for assessing textile quality); tools for generating the AI models; etc. For each of these elements, the company must examine one or more IP protections that may be available and best protect its business interests and investments. Moreover, data ownership and protection iare inherently tied to the access control and restrictions imposed on the use of such data.
This article explores different IP protections associated with each of the components of the above AI system (including big data).
Trade secrets are typically one of the most important IP protections available for all components of AI and big data discussed above because of their respective iterative nature, and the lack of registration or government approvals required for trade secret protection. Specifically—unlike other forms of IP protection that are afforded to something that is “frozen” at the time of registration or application filing and do not protect future iterations—trade secrets can be used to protect AI and big data that have notoriously short development cycles.
Additionally, trade secret protection may often be preferable over other forms of IP protection such as when: (a) other types of IP protections are unavailable (e.g., for AI models, for raw data, negative trade secrets, etc.), (b) other IP protection costs outweigh the benefits (e.g., securing patent protection can be expensive when weighed against the difficulty of detecting and proving infringement), or (c) the need for potential IP protection extends beyond the term available for other types of IP (e.g., 20 years for patents).
Best practices: Companies that rely on trade secrets for protecting their assets in AI and big data should take “reasonable measures” to keep their assets a secret. This secrecy requirement may prove difficult for manufacturing industries, given the large number of entities involved in any manufacturing operation. Thus, companies should proactively establish specific security measures—e.g., physical security, access control, proprietary markings and signed non-disclosure agreements (NDAs)—before granting access to trade-secret information.
While maintaining AI as trade secrets can ensure significant protection, unlike patents, trade secrets are susceptible to reverse engineering and do not bestow any exclusive rights. Therefore, patents remain a critical element for protection of certain components of the AI architecture that are eligible to receive patent protection. Given that AI-related patent applications are susceptible to subject matter ineligibity challenges (i.e., abstract ideas), companies should focus on obtaining patent protection for novel technical features in an AI architecture or workflow such as modeling tools, platforms and algorithms; training methods of the AI model; hardware components; overall architecture; practical use of the generated output; etc.
Best Practices: Patent protection requires careful claim drafting that applies potential abstract ideas in a specific way to achieve a practical result. For example, where AI relates to predictive maintenance based on detected vibration patterns, claims may include an active step of performing a specific maintenance action identified based on the AI output. Moreover, since AI systems are often considered “black box” systems for which the inner workings are often discernible, claims should be intelligently drafted to be able to detect and prove infringement. Finally, a patent application must name a “human” inventor, and claims should be drafted to emphasize human involvement.
Copyrights and Database Rights
Copyright protection can be another useful protection tool in a company’s IP arsenal, in particular for the protection of data structures and compilations associated with big data and various software associated with AI. Some European countries also grant sui generis database rights, which provide limited protection to databases if significant investments have been made to obtain, verify or present their contents.
Best Practices: Like patents, copyright protection is only available to works of authorship created by a “human” author, and human involvement should be emphasized. While copyright protection vests without registration, companies should also develop a strategy for timely registering their copyrights (e.g., software) for enforcement of their rights. Finally, “version control” of software is critical because, in litigation, the software version that is registered will need to match the version that the defendant copied.
As the IP landscape for AI and Big Data asset protection continues to evolve, companies should create a comprehensive IP policy to:
- Identify the assets that need protection such as data, algorithms, models, outputs, and hardware
- Identify best IP protection(s) for each asset, taking into consideration, for example, human involvement, availability of different protections, costs involved, desired length of protection, ease of detecting infringement and the competitive advantage provided by each available protection
- Take appropriate and timely steps to avail the identified IP protections
- Close any loopholes using contracts such as employment agreements, assignments NDAs, licenses (data and/or technology) to, for example, secure ownership, allocate risks, and define use restrictions.
A registered patent attorney, Gunjan Agarwal works closely with domestic and international clients to assist in development and management of worldwide patent portfolios.