Patent considerations for AI-related innovations

Especially in recent years, artificial intelligence (AI) has emerged as a transformational and increasingly ubiquitous tool, permeating many innovative developments across a diverse range of industries. As the number of AI-related innovations on the market increases, it’s critical for your business to seek robust protections for its intellectual property (IP).

The intersection of AI and IP protection is evolving on many fronts, particularly with respect to utility patent protection. In an earlier blog post, “Artificial Intelligence In the Innovation Process,” we discussed the patentability of inventions created with assistance from AI. Notably, the USPTO determined that named inventors must be natural persons who have made significant contributions to the invention.

Now, let’s turn our focus toward the question of seeking utility patent protection for innovations that either improve upon AI or rely on using AI in some form as a component piece.

Can AI-related inventions be patented?

The USPTO will grant patents to AI-related inventions so long as they meet the requisite criteria laid out in 35 U.S.C. § 101, § 102, and § 103. Namely, the subject matter must be:

• Novel
• Non-obvious
• Useful
• Patent-eligible subject matter

Regarding the last point, subject matter that cannot be patented includes abstract ideas — a concern that, in the wake of the Supreme Court’s Alice decision, is especially applicable to software or computer-implemented innovations.

(To read more about Alice and its ramifications on patenting software-implemented technology, check out our blog post, “Patenting Software and Beyond: A Guide to Understanding Alice and its Impact.”)

As such, simply mentioning the use of AI in a patent application does not necessarily create an unobstructed path to a granted patent. Rather, it’s important to consider where the innovation lies.

Is patent protection right for my AI-related innovation?

While several different types of IP exist, AI-related innovations with strong commercial potential are most commonly protected via patents or trade secrets:

• Patents give you the right to exclude others from making, selling, using, or importing a particular product or service — in exchange for full public disclosure of your invention.
• Trade secrets are formulas, processes, or other business information that derives its commercial value from being kept secret, and that a company is making reasonable effort to keep secret.

Trade secrets might be more suitable for your AI-related invention if it:

• Does not fulfill any of the patent eligibility requirements laid out in 35 U.S.C. § 101, § 102, or § 103
• Does not need to be publicly disclosed
• Can be kept secret
• Cannot be reverse engineered from your commercial product or is unlikely to be independently developed by a competitor

Ultimately, you should protect your inventions according to what will best serve your business priorities.

We’ve written more extensively about patents vs. trade secrets, and what you should consider when choosing between the two, in our blog post “What’s the Difference Between a Patent and a Trade Secret?”

What types of AI-related innovations exist?

Innovations related to AI can take many forms. In this blog post, we’ll focus on these three:

1. Improvements related to hardware implementations of AI
2. Improvements to AI models
3. Improvements to other areas of technology resulting from incorporating AI in novel ways

In the following sections, we’ll take a closer look at specific IP considerations for each of these three forms.

Patent considerations: Hardware-related improvements to AI

This first category relates broadly to improvements surrounding the physical implementation of AI systems — for example, improvements to the hardware itself, or software-driven techniques for controlling how underlying hardware is used.

This is perhaps the most straightforward category, as it may cover:

• Physical devices or components
• Hardware architectures
• Software for controlling lower-level hardware operations
• Manufacturing techniques that may provide improvements to a computing system or the operation thereof

In addition, hardware-related improvements may include the following measurable outcomes:

• Faster operations
• More efficient operations
• More secure operations
• Cheaper production costs
• Reduced physical dimensions

For an example of a hardware-related improvement for AI, we can look at U.S. Patent No. US11243880B1, assigned to Groq, which relates to a processor having a functional slice architecture. This architecture includes a plurality of “tiles” that each correspond to a functional unit within the processor. These tiles are organized into different physically-abutting “slices” that are used to perform different respective functions.

When protecting improvements to the operation of hardware, patents are preferred for two reasons.

First, these types of inventions are unlikely to encounter subject matter eligibility challenges under 35 U.S.C. § 101, since hardware is not conventionally considered an abstract idea. If anything, the primary hurdles you’ll face will be proving novelty and non-obviousness, which is a risk you can manage by performing a prior art search before drafting and filing your patent application.

Second, it may be possible for a competitor to reverse engineer your invention by inspecting your commercial product. Since trade secret protection can only apply to confidential information, patents will be more effective for maintaining your competitive edge in the market.

Patent considerations: Software-related improvements to AI

Software-related improvements to AI may take different forms, but they share similar patent considerations in the wake of Alice.

Improvements to AI models

At its core, an AI model is a computer program (or set of processes) executed by a computer system that uses collections of data sets to identify patterns. It is then able to draw conclusions, perform actions, or make future predictions without further human intervention.

You may be able to seek patent protection for an innovation related to AI models. Deploying an AI model typically involves three stages, all of which can generate valuable IP:

1. Designing the AI model
2. Training the AI model using data
3. Using the trained AI model in an application

An example of this category of innovation can be found in U.S. Patent No. US11636333B2, assigned to Tesla, which trains neural network models used in embedded systems by iteratively generating untrained models, assessing their performance, and then selecting a successful model to deploy.

Improvements to other areas of technology resulting from incorporating AI in novel ways

Artificial Intelligence as a Service (AIaaS) allows users to access trained and existing AI models, developed by third parties, through the cloud. It increases access to AI technology by offering users relatively low risk and low upfront costs.

If your business is provisioning existing AI tools for specific use cases, or processing the output from existing AI tools in novel ways, you might be able to seek patent protection for those innovations.

That said, if you’re dealing with licensed use of a third party’s AI model, you will want to understand the licensing terms — or consult with a qualified attorney — to ensure you resolve the question of ownership of your work before moving forward.

An example of this category of innovation includes U.S. Patent No. US11263540B2, assigned to Apple, which relates to using a user interface to generate a custom AI model based on user specifications.

Is patent protection the right strategy for software-related improvements to AI?

Software technology directed to AI may be patent eligible, but — on top of the innovation being new and non-obvious in view of the existing state of the art — you will also need to show that the claimed invention has a specific practical application, or recites significantly more than an abstract idea.

What might pass muster with the USPTO post-Alice? It might be helpful to refer to the USPTO’s recent Patent Subject Matter Eligibility Guidance Update, which discusses the patent eligibility of several different technologies that incorporate AI. Examples covered by the guidance include:

• Anomaly detection in data
• Speech processing that includes separating desired speech signals from background speech
• Personalizing medical treatment to a particular patient

An additional factor to consider is that the market for AI technology is evolving extremely quickly — but, as of November 2024, the traditional total pendency for a patent application filed with the USPTO is 26.1 months. Therefore, it is strategic to seek patents on technological features that could be relevant to your business in the longer term.

In view of these challenges, you might be wondering if you should instead pursue trade secret protection (which has no requirement for subject matter eligibility). However, you’ll want to weigh these two important considerations before deciding:

1. Detectability: Since a patent only prevents your competitors from using your technology, the value of patenting your invention depends on whether you can detect infringement. If it will be difficult to tell when your competitor is actually using your AI-related software, trade secret protection might be more appropriate.
2. Reverse engineering: Patent protection may be more suitable for AI-related software that can be easily reverse engineered or cannot easily be kept secret.

Protecting your AI technology: Partnering with a qualified patent attorney

Due to the fast-changing nature of the market for AI technology, we recommend working with an experienced attorney to craft the right plan to protect your long-term business interests. The team at Henry Patent Law Firm has deep expertise with current AI trends; please feel free to contact us with any questions!