Sam Underhill’s Post

METR provided a public comment on the U.S. AI Safety Institute’s valuable draft document “Managing Misuse Risk for Dual-Use Foundation Models.” In our comment, we offer several key recommendations: 1. Discuss additional non-misuse risks of AI, including loss of control. 2. Provide security recommendations to prevent model theft by advanced adversaries. 3. Expand guidance on model capability evaluations to include mid-training assessments, full capability elicitation, and third-party evaluations. 4. Provide additional guidelines for testing safeguard robustness, including automated methods. 5. Offer actionable suggestions for managing risks from deployment modalities, such as fine-tuning APIs and model weight access. 6. Include more detailed technical suggestions for implementing safeguards. 7. Suggest releasing AI safety frameworks with public commitments to evaluate and manage risks from dual-use foundation models. Read our full comment: https://lnkd.in/gPYRGcPR

📢 National Institute of Standards and Technology (NIST) has released the Initial Public Draft of the Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1) 🤖 This framework is a voluntary resource that helps organizations manage risks associated with the design, development, use, and evaluation of #AIProducts, #AIServices, and #AISystems. The Generative AI Profile is a specific part of this framework that deals with risks unique to or exacerbated by #GenerativeAI. It provides actions to manage these risks and includes primary #GAI considerations. The framework was released on January 26, 2023, after a consensus-driven, open, transparent, and collaborative process that included public comments, multiple workshops, and other opportunities for input. NIST is seeking public comments on this draft document, and the deadline for comments is June 2, 2024, at 11:59 PM Eastern Time. Please share your comments electronically with NIST-AI-600-1@nist.gov with “NIST AI 600-1” in the subject line or submit via www.regulations.gov (enter NIST-2024-0001 in the search field.) #AIRisks #GenAI #AIStandards #AIRiskManagement #AICertification #NIST #GenerativeAIProfile #AIDataPrivacy #AIConfabulation https://lnkd.in/eZ5FtK3m https://lnkd.in/ePBzWkgs https://lnkd.in/enGq2hnH https://lnkd.in/ecFFPjFW

I am glad to represent Morningstar Inc. in in its participation alongside many other institutions who responded thoughtfully to the #Treasury’s #RFI on #Uses, #Opportunities, and #Risks of #ArtificialIntelligence in the #Financial Services Sector including #Microsoft, #BlackRock, #CapitalGroup, #ICI, and others. In the comment letter we address many of the Treasury’s questions about how #AI is used in the #FinancialServices industry, discussing how #FinancialInstitutions can #mitigate AI risks and how AI can #benefit those institutions. We advocate for #regulators to take a principles-based approach to AI #regulation commensurate with the risks, outlining how #GenerativeModels and #LargeLanguageModels pose a greater risk than #DeterministicModels. Further, we discuss why AI regulation should be consistent across jurisdictions and point to the recent #EU AI Act’s definition of AI as a good model to follow. If you would like to learn more about how regulators ought to manage AI risks you can read the comment letter here: https://lnkd.in/dHmdUB7p

A team of professors has this warning about the direction of AI and related service providers, loosely known as Big Tech. "....if current trends continue and this emerging model for Internet traffic exchange becomes even more dominant, it will both greatly reduce the traditional model of Internet transit, and it will undermine the availability of a global, neutral platform for communications and innovation...." from https://lnkd.in/eX45HHUs Worth considering.

Interesting, the USPTO is kicking off a consultation on the proliferation of AI on prior art and for the person skilled in the art. Some of the questions are really interesting - AI generated art should be considered or not as prior art as they are generated without any human input, review or validation ? influence on the concept of person skilled in the art of the availability of AI technologies? and plenty of others interesting questions !!!

I was happy to partner with Morningstar Inc. on their comment letter to the recent #Department of the #Treasury #RFI on #Uses, #Opportunities, and #Risks of #Artificial Intelligence in the #Financial #Services Sector. We outlined how the #AI definition used in the RFI was overly broad and inclusive, threatening to overregulate long-established, well-regulated #deterministic models and how any new regulation should focus on #generative AI models and #large language models. We also outline the myriad #benefits of AI and were glad to utilize the survey research of Saifr, another partner of Sethi Clarity Advisers, highlighting how 85% of #marketing and #compliance leaders in #financial institutions believe that implementation of AI would #save them #money. You can read the comment letter and learn more about the benefits of AI here: https://lnkd.in/dHmdUB7p

AI's rise presents both opportunities and challenges for intellectual property (IP) policy. The USPTO aims to understand how AI affects what qualifies as prior art, the knowledge of a person skilled in the art (PHOSITA), and patentability determinations. Link: https://lnkd.in/dTVjzWNP #AI #USPTO #Patentability #Priorart

AI has many important benefits and drawbacks in society at large. In terms of accessibility, it can often be a great thing in many cases. However, we must have a say as patients how AI is developed and what it is used for to truly have equitable AI systems. The Department of Transportation is seeking comments from the community about AI in transportation systems and looking for ideas on what could be a possible use case scenario for different forms of AI. You don't have to be an expert in AI to comment on this. They're looking for all ideas and information. As patients and caregivers it's important that we have a say at such an important crossroad in our nation's history as this will be shaping transit systems and all forms of transportation including personal vehicles. Please take some time and share a comment through Regulations.gov to help out with this and to share your concerns as well as ideas. https://lnkd.in/epUERNQD

It's nearly impossible to have a conversation these days without someone bringing up AI... if you are one of those someones, let the Department know about your work! ARPA-I (it's DARPA for Infrastructure!) is asking for industry perspectives on the use, promise, and challenges of AI in transportation. This is an opportunity to shape the Departments research and innovation agenda, which feeds directly into it's policy and funding priorities. The RFI is open for the next 60 days: https://lnkd.in/dayBKygp And if you haven't seen it and are a fast writer, the Complete Streets AI opportunity is open for another week! https://its.dot.gov/csai/

USPTO’s New Subject Matter Eligibility Guidance The USPTO has released updated guidance on patent subject matter eligibility, specifically targeting innovations in critical and emerging technologies, including artificial intelligence (AI). Effective from July 17, 2024, this update aims to assist USPTO personnel and stakeholders in determining the subject matter eligibility of AI inventions under patent law (35 § U.S.C. 101). This update builds on previous guidelines to provide further clarity and consistency in evaluating the eligibility of claims in patent applications and patents involving AI technology. The update also includes three new examples illustrating the application of this guidance across various technologies. Key Highlights Effective Date: July 17, 2024 Objective: To foster and protect innovation in AI and other emerging technologies by providing clear guidelines for subject matter eligibility New Examples: The update includes three new examples that provide detailed analyses under 35 § U.S.C. 101 to address specific inquiries such as whether a claim recites an abstract idea or integrates it into a practical application Public Feedback: The USPTO invites public comments on the guidance update and examples until September 16, 2024 Detailed Examples of AI Subject Matter Eligibility The document “July 2024 Subject Matter Eligibility Examples” includes hypothetical scenarios illustrating the analysis of patent subject matter eligibility, specifically for AI applications. Below are summaries of the three key examples provided: Example 47: Anomaly Detection Claim 1: Eligible, as it recites a physical circuit (an application-specific integrated circuit for an artificial neural network). Claim 2: Ineligible, as it recites an abstract idea without integrating it into a practical application. Claim 3: Eligible, as it integrates the judicial exception into a practical application, improving network security. Analysis: Claim 1 focuses on hardware implementation, which is statutory. Claim 2 involves mental processes and mathematical concepts without a sufficient inventive concept. Claim 3 improves network security through specific technical implementations. Example 48: Speech Separation Claim 1: Ineligible, as it recites a mathematical concept without integrating it into a practical application. Claim 2: Eligible, as it improves speech-separation technology. Claim 3: Eligible, as it integrates the abstract idea into a practical application, improving speech-to-text transcription. Analysis: Claim 1 lacks details on how the deep neural network (DNN) operates, only reciting abstract ideas. Claim 2 involves specific steps that improve the functioning of the speech separation system, integrating the abstract idea into a practical application. Claim 3 focuses on the use of embeddings in a feature space to improve the transcription process. Check out the detailed examples https://wp.me/p4KrL7-2TM