Working as a patent attorney in the heart of Silicon Valley, I have helped protected numerous AI inventions. These days it seems everything I touch is powered by AI. For example, the December 2022 announcement of ChatGPT has threatened Google’s stranglehold over search. I have worked on medical imaging inventions whose innovations are not in the scanner hardware, but in the AI detection of lesions captured by the hardware.
Artificial intelligence is driving innovation in new ways and accelerating technological advances in computing power and data as well as algorithms. Technology advances make it possible to use AI tools in new ways that were previously impossible. As I have observed in my work with startups, these advances have resulted in a recent increase in AI deployments by companies, from startups to established institutions.
Intellectual property (IP), protects and encourages creativity and innovation. Investors and entrepreneurs need to be aware of the key intellectual property considerations in relation to AI innovation. Below we will discuss patents, trademark, trade secrets, and copyright protection for your AI invention.
Overview of AI Technology Protection Techniques
Protecting AI technology can involve a combination of different forms of intellectual property (IP) protection, such as patents, trademarks, and trade secrets. Here are some steps that can be taken to protect AI technology:
- Conduct a freedom-to-operate (FTO) search: Before investing in the development of an AI technology, it’s important to conduct a freedom-to-operate (FTO) search to ensure that the technology does not infringe on any existing patents or other forms of IP.
- Obtain patents: Patents can be used to protect the underlying ideas, methods, or techniques of an AI technology. This can include the algorithms used, the hardware configurations, and other unique aspects of the technology.
- Utilize trade secrets: Trade secrets can be used to protect confidential information such as source code, data sets, and other proprietary information that is used in the development and training of AI models.
- Trademarks: Trademarks can be used to protect the brand name and logo of the AI technology.
- Protect data: Data is often used to train AI models, it is important to protect the data from unauthorized access and use, through secure storage and access controls.
- Limit access to the technology: Only authorized employees and contractors should have access to the AI technology, and they should be required to sign non-disclosure agreements or other confidentiality agreements.
- Monitor and enforce: It’s important to regularly monitor and enforce the protection of the AI technology, and take appropriate action if infringement or misuse is suspected.
Trademarks can be used to protect the brand name and logo of an AI technology. Here are some steps that can be taken to obtain a trademark for an AI technology:
- Conduct a trademark search: Before applying for a trademark, it’s important to conduct a search to ensure that the desired trademark is available and not already in use. This can be done by searching the USPTO’s database or hiring a trademark attorney to conduct a comprehensive search.
- Choose the right trademark class: AI companies will need to choose the right class under which they will register their trademark. The USPTO uses a classification system to categorize goods and services, and it is important to choose the correct class to ensure the trademark registration will be granted.
- File a trademark application: After conducting a search and choosing the appropriate trademark class, the AI company can file a trademark application with the USPTO. This application should include the company’s name and contact information, the desired trademark, and a description of the goods and services for which the trademark will be used.
- Examination of the application: After filing the trademark application, it will be examined by a USPTO examiner. The examiner will review the application to ensure that it meets all legal requirements and that the proposed trademark is not already in use.
- Publication and opposition period: If the application is approved, it will be published in the Official Gazette, which is the USPTO’s weekly publication.
There is considerable debate about copyright protection for AI IP. Some proponents argue that an AI system is not a human author. Others claim that such a system can generate functionally creative works without human input. Several recent rulings have focused on copyright protection for AI IP, and their lessons can be useful for judges and IP administrators in other areas of law.
The most controversial issue in AI copyright law is ownership. One proposed solution would establish a sui generis right to protect AI-generated works. This new right would allow an AI system developer to secure copyright protection for these works for ten years. Its effect would be to encourage more advanced AI and promote the dissemination of AI-generated creations.
Another idea would extend the existing work-made-for-hire doctrine to AI-generated works. However, this approach does not change the scope of exclusive rights protection. In addition, it does not distinguish AI-generated works from conventional works of authorship.
An alternative idea would establish a two-tiered copyright system for AI-generated creations. First, an AI system developer should be identified as the owner of an AI-generated work. Secondly, the work should be registered with the U.S. Copyright Office.
Copyright protection can be used to protect software, including AI software. Here are some steps that can be taken to protect software with copyright:
- Determine copyrightability: Software is generally considered a “literary work” and is therefore eligible for copyright protection. However, it’s important to ensure that the software meets the requirements for copyrightability, such as originality and fixation in a tangible medium.
- Register the copyright: To obtain copyright protection for software, the software developer should register the copyright with the U.S. Copyright Office. This can be done by filling out an application and submitting it along with a non-refundable fee and a copy or deposit of the software.
- Include a copyright notice: Once the copyright is registered, it’s a good practice to include a copyright notice on the software, including the year of first publication and the name of the copyright owner.
- Monitor for infringement: It’s important to monitor for any potential infringement of the software copyright, and take appropriate action if infringement is suspected.
- Legal actions: If a copyright is infringed, the copyright owner can take legal action to enforce their rights, including seeking an injunction to prevent further infringement, and/or monetary damages.
It’s worth noting that copyright protection only applies to the specific expression of an idea, and not to the idea itself. Therefore, it’s important to consider other forms of IP protection such as patents and trade secrets, to protect the underlying ideas, methods or techniques used by the business.
In addition to protecting the software, there is a raging debate on the protection of AI generated work. Currently, the only legal way to protect AI-generated works is through copyright, but that is subject to debate as to whether a computer creator/inventor qualifies for copyright protection. Despite the growing use of AI in the creation of art, music, and journalism, the issue of AI-generated copyright has not yet been fully addressed. The recent rulings provide some valuable lessons for IP administrators, judges, and policymakers.
While the copyright protection for AI IP debate continues, it is important that a balanced approach is taken. Proponents of this approach should recognize the potential of AI-generated creations and recognize that the law needs to be structured in ways that will benefit humanity.
Trade secret protection
The purpose of a trade secret is to protect a company’s confidential information. These secrets are valuable for a company and cannot be shared with a third party. Often, organizations will require employees and business partners to sign confidentiality agreements in order to protect their secrets. Moreover, using such information without the owner’s permission is deemed unfair and a violation of trade secret protection laws.
Although trade secret protection is not exclusive to AI, it is certainly an effective means of ensuring the privacy of AI. It will protect the data and algorithms used to develop and deploy a product or service. However, trade secret protection does not give the owner an unrestricted monopoly on the subject matter. Rather, the owner is required to take reasonable measures to maintain secrecy.
Trade secret protection should be tailored to deter unlawful acquisition and to maintain the smooth functioning of the internal market for research. These procedures should be in line with the proportionality principle and should not undermine the rights of the public.
Trade secrets can include the structure of a model, formulas used in the model, or proprietary training data. They can also include the way the model output is converted into an end product for the customer. Trade secrets can be compromised in many ways. For example, it is possible for competitors to develop an AI/ML platform that uses the same models.
The first step in protecting your trade secrets is to password-protect the code of any AI/ML product that contains trade secrets. It is also important to ensure that the code is not available as open source. Only employees working on the code should be allowed to access it. Furthermore, this code should never be published or printed.
Another important step in securing your trade secrets is identifying them. Many companies focus their trade secret protection efforts on their most valuable trade secrets. However, trade secret protection is a continuing process that involves multiple parties. Trade secrets often get lost in the shuffle. For this reason, organizations must devote significant resources to identifying and protecting their trade secrets.
As an example, the deep learning AI training process is often protected as trade secrets due to its complexity and the large amounts of data required to train models. Here are some steps that can be taken to protect the deep learning training process as trade secrets:
- Identify the trade secrets: The first step in protecting trade secrets is to identify the information that constitutes a trade secret. This can include the data sets used to train deep learning models, the specific algorithms and configurations used to optimize the performance of the models, and any other proprietary information that gives the company a competitive advantage.
- Implement security measures: Once the trade secrets have been identified, the company should implement security measures to protect the information. This can include physical security measures, such as locked doors and security cameras, as well as digital security measures, such as password-protected databases and encryption.
- Create and implement a trade secret policy: The company should create a trade secret policy that outlines how the information will be protected and who will have access to it. The policy should also include procedures for handling the information and what to do in case of a suspected trade secret breach.
- Limit access to trade secret information: Only authorized employees and contractors should have access to the trade secret information, and they should be required to sign non-disclosure agreements or other confidentiality agreements.
- Train employees and contractors: Employees and contractors should be trained on the trade secret policy and understand their obligations to keep the information confidential.
- Monitor and enforce: It’s important to regularly monitor and enforce the trade secret
Patents offer a limited protection period for an invention. Patents grant the patent owner the exclusive right to use, make and sell their invention. In return for clear disclosure of how the invention is used, the patent owner also receives a patent.
Patents can be used to prevent others from using, making, or selling the patent technology. This may allow companies to gain or retain market share and protect their research and development investments. Patents may be used to gain a competitive advantage and as a negotiation tool. To prevent grants, patent publications may be cited against later filed applications.
If you plan to patent an AI-generated invention, you must ask yourself whether it is machine-generated work, a process, or a device. It should also be clear how it differs from conventional technologies, as well as how it could be generalized. For instance, is it capable of producing tangible results, such as a cheaper copy of a drug?
A patent-eligible AI-based invention is one that produces a tangible improvement to a computer system. This includes inventions that make use of machine learning or AI training models. When developing a patent-worthy AI-based invention, the focus must be on the tangible improvement.
The USPTO has provided guidance on the subject. According to the agency, an artificially intelligent system can be eligible for a patent if it performs an inventive step. This may be achieved by developing a system that can identify novel ideas before a human inventor does.
However, this innovation may be limited to an algorithm or a computer program. As the technology improves, it may eventually be capable of creating entire designs. An artificially intelligent system would need to understand the structure of a design and its functions. Without this information, a machine cannot be granted a patent.
Another strategy is to list a human as the inventor. This might be a good idea in certain cases, especially if the inventor is a professional in the field. A person who instructs an AI will probably qualify as the inventor if the problem is obvious and structured in a way that is a clever design solution.
In sum, patents, along with trade secrets and copyright protection, are critical elements for the security of an AI architecture. While there are other ways to protect your AI, the key is to ensure that you maintain the freedom to operate, while also securing your investment in AI research. If you fail to adhere to the required standards, you could be liable for the damages caused by your technologies.