Home Technology range AI and machine learning
AI and machine learning patents
AI inventions can be patented, but they are easy to claim badly. The claims have to describe a concrete technical improvement, not an abstract idea, and they have to cover something a competitor’s product would visibly do.
What the work covers
- Models and training. Model architectures, training methods, data pipelines and techniques that make training faster, cheaper or more accurate.
- Inference and deployment. Serving, compression, quantization, on-device inference and edge deployment.
- Generative AI and LLM applications. Retrieval, agents, tool use, evaluation and the systems built around large language models.
- Computer vision. Perception, image and video analysis, and their use in robotics, vehicles and devices.
- AI in products. Inventions where machine learning improves a device, a network or a business system.
Issues that decide the value of an AI patent
- Patent eligibility. Under the Supreme Court’s Alice decision, claims directed to an abstract idea, such as math or a mental process carried out on a computer, are not patentable unless they add something more. Claims that tie the model to a specific technical improvement, such as lower latency, less memory or better accuracy on a defined task, fare better.
- Detectability. A claim to how a model is trained may be impossible to prove against a competitor. Claims are drafted at several points, from training to the observable behavior of the deployed product, so at least some can be enforced.
- Patent or trade secret. Training data, weights and some training methods may be better protected as trade secrets, especially when they cannot be seen in the product.
- Inventorship. Only people can be named as inventors; the Federal Circuit confirmed in Thaler v. Vidal (2022) that an AI system cannot be. When engineers use AI tools, recording who contributed what matters.
The foundation
Alex holds a B.S. in electrical engineering from Columbia, worked six years as an Intel product engineer on the i486 and Pentium® processors and then drafted and prosecuted Intel’s applications in processor architecture, hardware and software. His client patents include algorithm-and-system claims in video encoding, search and network routing, the same kind of claims many AI inventions require. See Selected patents for work obtained to date.
Talk it through
Alex leads every matter and is your point of contact. Most work is a fixed fee, quoted in writing before it starts.
Need advice this week? The one-hour consultation can be booked and paid online. See how to get started. Please do not include confidential technical details in a first message.