How to argue with an AI in the patent environment

Our new AI reasoning engine justifies patent evaluation decisions
Our new AI reasoning engine justifies patent evaluation decisions
EconSight has implemented a reasoning engine to better understand and challenge AI decisions. That is, for each patent, our AI justifies why this patent is or is not relevant for a selected question. This is the decisive step towards transparency, traceability, and the qualitative classification of AI-based patent analysis.
Conceptual AI searches – with descriptions formulated by humans – are ideal for cutting-edge technologies where patent classes do not yet exist, where terms are unclear or have varying meanings, and for very small, specific technology domains.
We have been using this for more than a year now, mainly as part of our Precision Investing approach for the financial sector and for broader, difficult, complex or variably worded FTOs.
Until now, however, we have had to use human plausibility checks to verify the highly precise technology definitions, i.e. we have combined human and artificial intelligence. This is often time-consuming, requires subject matter experts and is often one-sided, because while we as experts can evaluate and justify individual cases, the AI simply produces a binary result whose derivation was unclear.
We have now implemented a new approach in which the AI justifies the decision for each patent. Traditional keyword search is «word focussed» and therefore not defined in terms of content. Modern reasoning AI is different, and considers semantic relevance «content focussed» with comprehensive but unspecific expertise, which makes it comparable to the ‘person skilled in the art’ often used in patent legislation.
With the AI reasoning engine, we now have a complete set of operational AI-based patent analysis modules, from search to classification/grouping to relevance assessment and decision.
Using the example of humanoid robots, we show how our reasoning engine evaluated three patent examples that would have been classified as relevant in a classic keyword search. Spoiler: our reasoning engine argues that one patent is relevant, one is partially relevant and one is not relevant. But see for yourself.
Example Humanoid Robots
We have defined humanoid robots with three core principles:
- A bipedal robot that can perform human-like actions.
- The robot can move in human-like environments.
- The robot is able to interact with humans.
The following features result from this:
- Mimic human body in in form and function
- Human-like appearance with head, torso arms and legs
- Human-like movements
- Interaction with human environments
- Navigate in offices and houses
- Capable of mimicking human motion such as grasping, walking, assembling, carrying
- interact and handle objects built for human use
- Human-centered communication
- Mimic human communication with facial expressions, speech, gesture recognition
We provide the AI with these simple terms, which we as experts consider relevant. However, we expect the AI to apply its own knowledge to assess relevancy, based on the claims and the summary of the invention.
As input we provide the AI with a large basket of patent documents compiled by our AI-based patent search engine. Then our AI reasoning engine analyses each patent’s claim and summary of invention according to the defined concept and explains it’s decision.
Below we show how our reasoning engine evaluated three patent examples that would have been classified as relevant in a classic keyword search. Our reasoning engine argues that one patent is relevant, one is partially relevant and one is not relevant.
Example of a fully relevant humanoid robot patent
US2024109612A1, «Systems, devices, and methods for a humanoid robot», Sanctuary AI
The patent pertains to a bipedal humanoid robot. Here’s the reasoning:
- **Bipedal Structure**: The patent describes a robot with a first and second robotic leg, each having a foot lower leg member, and upper leg member. This structure is indicative of a bipedal configuration, as it involves two legs that are mechanically coupled to a torso, which is a common characteristic of bipedal robots.
- **Humanoid Features**: The robot includes a robotic torso and mentions the possibility of having a head, arms, and hands, which are typical features of humanoid robots. The presence of a torso and the potential for humanoid arms and hands suggest a design that mimics human anatomy.
- **Human-like Movements**: The description includes the ability for the robot to bend at various joints (hip, knee, ankle), which are necessary for human-like movements such as walking, bending, and sitting. These capabilities are essential for a robot to mimic human actions and interact with environments designed for humans.
- **Interaction with Human Environments**: The patent discusses the robot’s ability to perform tasks, carry loads, and interact with objects, which implies functionality in human-centered environments. This is a key aspect of humanoid robots, which are often designed to operate in spaces intended for human use.
- **Claims and Descriptions**: The claims and descriptions consistently refer to the robot as having a humanoid body and being capable of various articulations, further supporting the notion that it is designed to resemble and function like a human.
Overall, the combination of a bipedal structure, humanoid features, and the ability to perform human-like movements in human environments strongly indicates that the patent is about a bipedal humanoid robot.
Decision: fully relevant
US2024109612A1, «Systems, devices, and methods for a humanoid robot», Sanctuary AI
Inventor: Gildert Suzanne
Applicant: Sanctuary Cognitive Systems Corp
First Filing Date:09/26/2023
First Publication Date:04/04/2024
A robot includes a robot body having a first robotic leg, a second robotic leg, and a robotic torso. The first robotic leg includes a first foot, a first lower leg member coupled to the first foot, and a first upper leg member coupled to the first lower leg member. The second robotic leg includes a second foot, a second lower leg member coupled to the second foot, and a second upper leg member coupled to the second lower leg member. The robot includes a mobile base having a platform to which the first and second feet of the robot body are fastened.

Example of a partially relevant humanoid robot patent
US2016046023, «Control interface for robotic humanoid avatar system and related methods»,
Univ Florida
The patent pertains to a humanoid robot system that is controlled remotely through a master-slave architecture.
The focus is on the control interface and the mapping of a user’s movements to a virtual character, which then controls the humanoid robot. The description and claims emphasize the humanoid nature of the robot, mentioning features such as a head, torso, arms, and hands. The robot is described as having joints that allow for human-like movements, including head and arm motions.
However, the patent does not explicitly focus on bipedal locomotion or the use of legs for walking. While it mentions that legs or other mobility mechanisms (such as wheels or whegs) can be provided, it also states that in some embodiments, the legs are not used, particularly in environments where safety is a concern. This suggests that the primary focus of the patent is not on bipedal movement but rather on the control and interaction capabilities of the humanoid robot.
In conclusion, while the patent describes a humanoid robot with human-like features and capabilities, it does not specifically pertain to a bipedal humanoid robot, as the emphasis is not on two-legged locomotion or walking. The robot’s mobility is not a central aspect of the invention, and alternative mobility mechanisms are considered.
Decision: partly relevant
US2016046023, «Control interface for robotic humanoid avatar system and related methods», Univ Florida
Inventors: Kelly Brian, Nagendran Arjun
Applicant: Univ Central Florida Res Found Inc
First Filing Date:08/17/2015
First Publication Date:02/18/2016
A system and method for controlling a humanoid robot from a remote location are provided. One or more sensing devices are used to sense positions and movements of a user at an inhabiter station. A computing device generates a virtual character based upon the sensed positions and movements of the user and provides a plurality of outputs for controlling the humanoid robot based upon motion of the virtual character. The computing device includes a master controller to generate the virtual character and its behaviors and to transmit instructions to control motion of the humanoid robot to a slave controller located at the humanoid robot, providing smooth and natural motion of the humanoid robot. The method permits switching between multiple virtual characters in a single scene to control and alter the behavior of one or more humanoid robots located in several geographic locations.

Example of a not relevant humanoid robot patent
US2020047329, «Humanoid robotic system and methods», Willow Garage Inc
The patent does not pertain to a bipedal humanoid robot. Here’s the reasoning
- **Mobile Base**: The patent describes a mobile base that is configured for holonomic or semi-holonomic motion, which typically implies a wheeled or omni-directional base rather than a bipedal structure. Bipedal robots would generally have legs for locomotion, which is not mentioned here.
- **Structure and Components**: The patent mentions a spine structure, body structure, and robotic arms, but there is no mention of legs or a bipedal locomotion system. The focus is on a mobile base with a vertically extensible torso and arms, which suggests a wheeled or stationary robot rather than one that walks on two legs.
- **Human-like Features**: While the robot is described as humanoid, the emphasis is on human-scale capabilities and interaction with human environments, not on mimicking human walking or having a full humanoid form with legs.
- **Claims**: The claims focus on a mobile base, a vertically extensible torso, and arms with a gravity compensation mechanism. There is no mention of legs or bipedal movement, which are essential for a robot to be considered bipedal.
- **Comparison with Other Robots**: The description contrasts this robot with other humanoid robots like Asimo and HRP2, which are explicitly mentioned as bipedal. The patent’s robot is more akin to those with wheeled bases, like the STAIR robots or Robutler, which are not bipedal.
In conclusion, the patent describes a humanoid robot with a mobile base and arms, but it lacks the key features of a bipedal humanoid robot, such as legs and the ability to walk on two feet.
Decision: NOT relevant
US2020047329, «Humanoid robotic system and methods», Willow Garage Inc
Inventors: Berger Eric H, Wyrobek Keenan A
Applicant: Berger Eric H, Willow Garage Inc, Willow Garage Lic, Wyrobek Keenan A
First Filing Date:04/11/2011
First Publication Date:03/15/2012
Systems and methods related to construction, configuration, and utilization of humanoid robotic systems and aspects thereof are described. A system may include a mobile base, a spine structure, a body structure, and at least one robotic arm, each of which is movably configured to have significant human-scale capabilities in prescribed environments. The one or more robotic arms may be rotatably coupled to the body structure, which may be mechanically associated with the mobile base, which is preferably configured for holonomic or semi-holonomic motion through human scale travel pathways that are ADA compliant. Aspects of the one or more arms may be counterbalanced with one or more spring-based counterbalancing mechanisms which facilitate backdriveability and payload features.
