SAFRON Opportunity Description

Overview: DARPA/DSO has issued an ARC Opportunity, inviting submissions of abstracts and proposals for innovative exploratory research concepts in the technical domain of artificial intelligence (https://www.darpa.mil/ARC). ARC Opportunities are designed to allow an individual researcher the opportunity and time to focus on nascent, paradigm-shifting ideas for national security applications. This SAFRON ARC Opportunity is issued under the master ARC Exploration Announcement (EA), DARPA-EA-24-01.

Background: Foundation models (FMs) have transformed AI capabilities in many domains by virtue of their large architectures, internet-scale datasets, and unique customization techniques (“fine tuning”). This trend has recently brought transformational capabilities to robots as well. In particular, FMs enable robots that can parse natural-language directions for complex tasks and then contextualize and execute those tasks in unconstrained, open-world environments – to include even “zero-shot” scenarios. This is a dramatic break from existing autonomous systems, which are designed for tailored applications and narrow, precise operating conditions. However, natural-language direction for open-world autonomy presents a critical challenge from a safety and assurance perspective, since current methods to assure learning-enabled systems are inadequate to address FMs operating in this paradigm. For example, FMs are known to exhibit unique (semantically) errant behavior such as hallucination, false confidence in reasoning, and manipulation via “jailbreaking.” Assurances are crucial to deploy FM-enabled robots so that they do not manifest these behaviors, which could prevent execution of a critical task.

Solicitation Question: This SAFRON ARC Opportunity is soliciting ideas to explore the following question: How, and to what extent, can we assure that FM-enabled robots will behave only as directed and intended?

Scope: The SAFRON ARC Opportunity seeks proposals that clearly articulate a novel approach to assurance (or the limits thereof), and that clearly explain how their approach is relevant to current and future state-of-the-art FMs employed in open-world robotics applications.

Special Notice

Event Participation

Registration

There is no registration fee for the webinar. The registration deadline is Monday, October 21, 2024 at 12:00 PM EDT or when attendance capacity is met, whichever comes first. There will be no same-day registration. Advance registration is required for all events and is mandatory for every individual intending to view the webcast either alone or as part of a group.

If interested individuals cannot register because registration is closed or capacity has been reached, instructions for being added to the waitlist will be provided on the website. If slots remain open after registration closes or become available due to cancellations, the slots will be filled on a first-come, first-served basis from the waitlist. Individuals requesting to attend in excess of the capacity limitation will also be added to the waitlist.

Eligibility

Attendance is open to individuals who are U.S. citizens, U.S. permanent residents, and foreign nationals. All registrants who are not U.S. citizens must complete and submit either a DARPA Form 60 (U.S. Permanent Resident and Foreign National Visit Request – e.g., industry or academia) or an Official Visit Request (foreign government personnel, only) through their country's Embassy based in Washington, DC, no later than 5:00 PM EDT on Wednesday, October 16, 2024. Form 60 submission instructions are provided in the registration confirmation email. Contact your Embassy staff for assistance in submitting the Official Visit Request.

Agenda

Webinar Information

This meeting will be held on the ZoomGov webinar platform.

Downloading the Zoom application is suggested but understood that there are affiliation restrictions. Therefore, the alternative would be to join by web browser.