Current artificial intelligence cannot reliably determine patent essentiality for telecommunications standards because the task is inherently subjective. Unlike precise technical measurements, legal and technical interpretations vary even among human experts, making it impossible to establish a consistent ground truth. Consequently, AI algorithms lack the necessary logical reasoning to justify their determinations in legal contexts, meaning they cannot currently replace human expert judgment or satisfy judicial requirements for accountability. The fundamental barrier to accurate AI modeling is the lack of high-quality training data. Existing datasets are often biased toward patents deemed essential, as patent owners have financial incentives to submit only those with high perceived validity. This systemic bias, combined with inconsistent manual assessments, means that "garbage in, garbage out" principles apply heavily. Without a comprehensive, randomly selected, and thoroughly verified dataset of both essential and non-essential patents, any AI trained on current information will inherit and amplify these inaccuracies. This limitation is critical for open data initiatives because it highlights the necessity of transparent, verifiable, and unbiased data structures. Open data principles cannot succeed if the underlying records are skewed or subjective, as this undermines trust in automated systems. The article implies that for AI to be viable in intellectual property, stakeholders must prioritize the creation of standardized, openly accessible, and rigorously validated data sets that reflect diverse outcomes, rather than relying on flawed or selective disclosures.
Source:Published on 2023-03-17