This article is more than 1 year old
Tech industry stuck over patent problems with AI algorithms
Lawyers at Google are unsure if they can patent chip floorplans created by machines
The question of whether AI-generated outputs can be patented is impacting how technology companies can protect their intellectual property.
Some of the most hyped up AI technologies are systems that can produce surprisingly creative outputs. Uncanny poems, short stories, and striking digital art have all been generated by machines. The human effort required to initiate these processes are often trivial: a few clicks or typing a text description can guide the machine towards producing something useful.
Similar generative AI models are also being applied in scientific and technological applications. Machine learning algorithms can, for example, spit out molecule combinations in the hunt for new drugs, map out schematics for novel chip designs, and even write code.
Under current US laws, intellectual property is only recognized and protected if it is created by "natural persons". Humans build these models but, after training, their outputs are often automatically generated with little assistance. Which raises the question of whether the human developer of an AI system should be considered the inventor, or if the machine can claim the credit? The question is also a legal puzzler: is it permissible to patent chemical compounds, or software created by these systems?
Laura Sheridan, senior patent counsel at Google, asked the US Patent and Trademark Office for guidance on the issue. She said Google had filed multiple patents describing a machine-learning technique used internally to automatically design and map out components in the company's custom AI accelerator TPU chips currently used in its servers.
"We have filed patent applications on the novel machine learning models. As for the floor plans themselves generated by the models, we have not pursued patents on these," she said at the inaugural meeting of the AI and Emerging Technologies (ET) Partnership Series held by the US Patent and Trademark Office.
- AI-friendly patent law needed 'as a matter of national security', ex-USPTO boss says
- AI inventors may find it difficult to patent their tech under today's laws
- AI really can't copyright the art it generates – US officials
- Meta materials: Facebook using AI to design green concrete
The uncertainty surrounding whether it's possible or how best to apply for patents protecting IP produced by algorithms can sometimes be a roadblock in developing new products, especially in the pharmaceutical and biotech industries.
Companies that rely on using AI software to create new drugs or antibodies, for example, often need to secure patents before they can kick start clinical trials. The stage is critical for the approval to get new treatments and medicines to market for patients.
"We at Google are definitely giving a lot of thought to the inventorship question overall… We're thinking through inventive contribution issues throughout the AI development process," Sheridan concluded.
"It's an area that we've given a lot of thought to and I think it would be very useful for [USPTO] to also offer patent applicants guidance to just help make these assessments. They're very complicated." ®