It has been more than three years since ChatGPT was released on November 30, 2022. Many people believe that generative AI is transforming IP law practices at an increasing pace.
The AIPPI Resolution
In fact, discussions within the international IP community about the potential impact of neural-network-based AI began even before the outbreak of Covid-19. One example can be found in the 2019 AIPPI World Congress held in London, where a highly regarded topic was “Inventiveness and sufficiency of disclosure in AI inventions.” That discussion continued into the Covid years and eventually led to AIPPI Resolution Q276, adopted on October 22, 2021. The key conclusions of this Resolution include the following (highlights added by the writer):
- The definition of the “person skilled in the art” should be the same for inventions made by or using AI technology and inventions made without the use of AI technology.
- The “person skilled in the art” should be a hypothetical natural person (or team of hypothetical natural persons) and would use AI technology only if that use would have been expected from the average person in the field (or fields) to which the invention relates, at the relevant date.
- The definition of common general knowledge should not differentiate between inventions made by or using AI technology and inventions made without the use of AI technology.
Turning back to Taiwan, the local practice is still developing. Questions such as whether and how AI tools should be considered when determining the “person skilled in the art” and “common general knowledge” have not yet been fully tested. While AI is already being used in research and IP-related works (for example, prior art searches), courts do not have many chances to address such issues until they become part of an actual dispute.
Insights from a Medical Dispute Case
However, a recent court decision did touch upon a different yet comparable issue regarding use of AI. Kao v. Yang, et al., 115 Yi-Zai-Yi 1, Taiwan High Court Kaohsiung Branch Court (January 2026).
The issue was: should the “accessibility” of evidence in the context of a retrial request take into account the availability of AI tools?
A retrial is a special remedial proceeding under Taiwan’s Code of Civil Procedure, providing an opportunity to correct a final judgment that is later found to potentially contain major procedural, legal, or evidentiary defects. Of course, the party seeking a retrial must prove the defect it claims.
One recognized form of serious evidentiary defect to warrant a retrial request is “Where a party discovers tangible [and favorable] evidence which has not been considered [in the final judgment], on condition that taking into consideration such tangible evidence will result in a more favorable decision to such party.” (Article 496.1.13 of the Code of Civil Procedure)
Taiwanese courts have traditionally interpreted this as referring to evidence that already existed before the end of oral arguments in the original trial, but was not considered by the court because the party did not know of its existence at the time. In short, the party could not reasonably have been expected to discover such evidence and use it in the original trial.
In this case, which was a medical dispute, the plaintiff argued that the final judgment failed to consider medical information compiled by Google’s AI model, including materials concerning “thrombosis with thrombocytopenia syndrome” and the “AZ vaccine.” He claimed that, had such evidence been considered in the original trial, the Taiwan High Court Kaohsiung Branch would have reached a different conclusion and would not have dismissed his claims against the hospital, which he alleged wrongfully administered a Covid-19 vaccine to his elderly family member and thereby caused his death.
The plaintiff argued that Google’s AI model was not officially available to Taiwanese Traditional Chinese users until October 2025, after the final judgment was rendered in September 2025. Therefore, he claimed that the information should qualify as “evidence which has not been considered” and justify a retrial.
However, the Taiwan High Court Kaohsiung Branch Court dismissed his retrial claims on January 30, 2026. Regarding the AI-generated information, the court reasoned that:
- The online information obtained through Google’s AI model already existed on the internet. The plaintiff could have searched for, analyzed, and presented the information to the court, and he did not prove that it was impossible to locate the information by the end of the original trial.
- As the Google AI model merely simulates human intelligence by rapidly searching, organizing, and summarizing online information, it is not the only way to obtain the evidence in dispute.
- Since the evidence was obtained through AI-based searching and analysis, the accuracy of the AI system’s filtering and interpretation must also be considered. Given that AI-generated information is known to sometimes contain errors, the disputed materials were difficult to treat as reliable enough to overturn other evidence favorable to the defendants.
In short, the court viewed the AI-generated content as prior accessible evidence, rather than newly discovered evidence.
The court’s reasoning—especially its focus on whether the use of AI merely replaces human effort, and on how reliability and accuracy should be evaluated—may offer valuable insights for IP practitioners, for example, in cases involving whether and how AI tools should be considered when determining the “person skilled in the art” and “common general knowledge.”