{
  "patent_number": "US 11651227",
  "country": "US",
  "title": "How to Force AI to Follow Logical Rules During Training",
  "original_title": "Trusted neural network system",
  "summary": "A system that uses a dual-headed neural network to ensure AI models obey specific logical rules by embedding those rules directly into the training process.",
  "what_it_does": "This system improves AI reliability by preventing models from making decisions that violate predefined logical rules. It uses a 'shared' neural network that feeds into two separate branches: a data head and a logic head. The data head focuses on learning patterns from raw data, while the logic head monitors whether those patterns violate specific logical constraints. If a violation occurs, the logic head sends an error signal back to the shared network, forcing it to adjust its internal parameters until the output satisfies the rules. For example, in a self-driving car application, this could ensure the AI never predicts a path that crosses a solid double-yellow line, regardless of what the training data suggests.",
  "what_it_does_not_cover": [
    "Does not cover standard neural networks that lack a dedicated logic head for constraint enforcement.",
    "Does not cover post-training filtering or 'guardrail' systems that check outputs after the AI has already made a prediction.",
    "Does not cover systems that rely solely on massive datasets to implicitly learn constraints without explicit logical rule integration."
  ],
  "filed": "2018-12-19",
  "granted": "2023-05-16",
  "expires": null,
  "status": "active",
  "holder": "SRI International Inc",
  "holder_url": "https://patentbrief.org/company/sri-international-inc",
  "inventors": [
    {
      "name": "Patrick Lincoln",
      "url": "https://patentbrief.org/inventor/patrick-lincoln"
    },
    {
      "name": "Shalini Ghosh",
      "url": "https://patentbrief.org/inventor/shalini-ghosh"
    },
    {
      "name": "Susmit Jha",
      "url": "https://patentbrief.org/inventor/susmit-jha"
    },
    {
      "name": "Ashish Tiwari",
      "url": "https://patentbrief.org/inventor/ashish-tiwari"
    }
  ],
  "times_cited": 2,
  "tags": [
    "ai_ml",
    "software",
    "telecommunications"
  ],
  "abstract": "In general, the disclosure describes techniques for facilitating trust in neural networks using a trusted neural network system. For example, described herein are multi-headed, trusted neural network systems that can be trained to satisfy one or more constraints as part of the training process, where such constraints may take the form of one or more logical rules and cause the objective function of at least one the heads of the trusted neural network system to steer, during machine learning model training, the overall objective function for the system toward an optimal solution that satisfies the constraints. The constraints may be non-temporal, temporal, or a combination of non-temporal and temporal. The constraints may be directly compiled to a neural network or otherwise used to train the machine learning model.",
  "url": "https://patentbrief.org/patent/us/11651227/muzero",
  "markdown_url": "https://patentbrief.org/patent/us/11651227/muzero/md",
  "google_patents_url": "https://patents.google.com/patent/US11651227",
  "relatedPatents": []
}