Patent Requirements for Algorithm-Based Inventions

** Attention AI Innovators and AI IP Attorneys ** The USPTO just released official AI patent guidelines. Here's what you need to know: The factors and evidence required for enablement and written description of AI inventions are based on the following principles: - The patent disclosure should reasonably convey to those skilled in the art that the inventor had possession of the claimed AI invention as of the filing date. - The patent disclosure should describe the AI invention in such a way that allows one skilled in the art to make and use the AI invention without undue experimentation. Factors and evidence that can be used to demonstrate these principles are: - The predictability and reproducibility of the AI invention, such as whether the AI invention produces consistent and reliable results, and whether the AI invention can be replicated by others using the same or similar data and parameters. - The amount of guidance and direction provided by the specification, such as whether the specification describes the technical features and functionality of the AI invention, the AI system, the training data, the algorithm, the input/output, and the result, and whether the specification provides mathematical formulas, flow charts, pseudocode, or other details that explain how the AI invention works. - The availability of general and specific knowledge in the art, such as whether the AI invention is based on well-known or widely used techniques, or whether the AI invention requires specialized or proprietary knowledge, data, or tools that are not readily accessible or disclosed. Source: Law360 PDF here: https://lnkd.in/emJRMFi2

Can Algorithm-Based Startups Like Fitbit Get Patents in India? Let’s Decode Section 3(k) of the Indian Patents Act? Many startups today especially in AI, wearable tech, fintech, and health analytics rely heavily on algorithms as their core innovation. But when it comes to patenting such ideas in India, Section 3(k) of the Indian Patents Act often becomes the biggest roadblock. ⚖️ What Section 3(k) of the Indian Patents Act Says: “A mathematical or business method or a computer program per se or algorithms are not patentable.” So, if your invention is just an algorithm or software-only method, it’s not patentable in India. If your algorithm is tied to a hardware component or produces a technical effect, there is a room for protection. For instance, Fitbit’s motion-tracking algorithm, patented in the U.S., might not be directly patentable in India as a standalone algorithm. But if claimed as a wearable device that integrates sensors and processors executing the algorithm to improve power efficiency, accuracy, or real-time tracking, it could qualify as a patentable invention under Indian law. In India, the line between an algorithm and a patentable technical innovation lies in how the invention is implemented and claimed. A well-drafted patent application that highlights hardware integration or a technical improvement can overcome Section 3(k) challenges. 💬 So for startups especially those building AI-driven or wearable solutions , the trick is not avoiding Section 3(k), but drafting around it. #IntellectualProperty #Patents #Section3k

𝗘𝗣𝗢 𝗕𝗼𝗮𝗿𝗱 𝗼𝗳 𝗔𝗽𝗽𝗲𝗮𝗹 𝗖𝗹𝗮𝗿𝗶𝗳𝗶𝗲𝘀 𝗗𝗶𝘀𝗰𝗹𝗼𝘀𝘂𝗿𝗲 𝗥𝗲𝗾𝘂𝗶𝗿𝗲𝗺𝗲𝗻𝘁𝘀 𝗳𝗼𝗿 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗣𝗮𝘁𝗲𝗻𝘁𝘀: 𝗔 𝗠𝘂𝘀𝘁-𝗥𝗲𝗮𝗱 𝗳𝗼𝗿 𝗣𝗿𝗮𝗰𝘁𝗶𝘁𝗶𝗼𝗻𝗲𝗿𝘀 The EPO's recent decision in T 1669/21, concerning a method for predicting wear in metallurgical vessels using a "computational model", offers valuable insights into the current EPO approach to examining machine learning patent applications. The case, which was rejected due to insufficient disclosure, provides important lessons for patent practitioners. 𝗪𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝘁𝗵𝗶𝘀 𝗺𝗲𝗮𝗻 𝗳𝗼𝗿 𝗽𝗮𝘁𝗲𝗻𝘁 𝗽𝗿𝗮𝗰𝘁𝗶𝗰𝗲? The decision underscores the need for meticulous attention to detail when drafting patent applications for machine learning inventions. Specificity is paramount. While the EPO accepts broad claims, these must be supported by a commensurately detailed and enabling disclosure. Here's what the EPO expects - 𝗖𝗹𝗲𝗮𝗿 𝗱𝗲𝗳𝗶𝗻𝗶𝘁𝗶𝗼𝗻 𝗼𝗳 𝘁𝗵𝗲 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗺𝗼𝗱𝗲𝗹: The type of model (e.g., neural network, support vector machine), its architecture, and the specific algorithms used must be explicitly stated. Simply referring to a generic "computational model" is insufficient. 𝗗𝗲𝘁𝗮𝗶𝗹𝗲𝗱 𝗽𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿 𝗺𝗮𝗽𝗽𝗶𝗻𝗴: The application must provide clear guidance on how to select, pre-process, and represent input parameters within the model. This includes specifying how to handle time-varying or multi-dimensional parameters. Examples are crucial for illustrating these steps. 𝗧𝗿𝗮𝗻𝘀𝗽𝗮𝗿𝗲𝗻𝘁 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗽𝗿𝗼𝗰𝗲𝗱𝘂𝗿𝗲𝘀: The description should cover the training data used, the training process, and the criteria for evaluating model performance. It should also address potential challenges such as data scarcity and the prevention of artefacts from random correlations. 𝗪𝗼𝗿𝗸𝗶𝗻𝗴 𝗲𝘅𝗮𝗺𝗽𝗹𝗲𝘀: Where possible, include concrete, workable examples demonstrating the implementation of the invention. This could involve providing sample data, model configurations, and training scripts. 𝗪𝗵𝘆 𝗶𝘀 𝘁𝗵𝗶𝘀 𝗶𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝘁? The EPO's focus on specificity stems from the Article 83 EPC requirement for sufficient disclosure. The patent application must enable a skilled person to carry out the invention without undue burden. This is particularly challenging for machine learning inventions, which often involve complex models and data-driven processes. 𝘓𝘪𝘯𝘬𝘴 𝘵𝘰 𝘧𝘶𝘭𝘭 𝘣𝘭𝘰𝘨 𝘱𝘰𝘴𝘵 𝘢𝘯𝘥 𝘵𝘩𝘦 𝘥𝘦𝘤𝘪𝘴𝘪𝘰𝘯 𝘪𝘯 𝘵𝘩𝘦 𝘤𝘰𝘮𝘮𝘦𝘯𝘵𝘴. #EPO #MachineLearning #Patents #SufficiencyOfDisclosure #PatentPractice #CaseLaw