Election results are out! Here is your official 2024-2025 Executive Board. 

President ~ Cassandra Paris

Vice President ~ Kassim Chipkaou

Treasurer ~ Jaden Dunning

Secretary ~ Fikunayomi Adeyemo

Technical Chair ~ Dylan Williamson

Fundraising Chair ~ Addisbrhane Negussie

Volunteer Chair ~ Solomon Brundridge

Public Relations Chair ~ Parth Madaan

Thank you so much to everyone that applied, we appreciate you! New e-board members, please be on the lookout for emails/messages as we make this transition!

The learned behaviors of AI and robot agents should align with the intentions of their human designers. Toward this goal, people must be able to easily specify, inspect, and model agent behaviors. For specifications, we will consider expert-written reward functions for reinforcement learning (RL) and non-expert preferences for reinforcement learning from human feedback (RLHF). I will show evidence that experts are bad at writing reward functions: even in a trivial setting, experts write specifications that are overfit to a particular RL algorithm, and they often write erroneous specifications for agents that fail to encode their true intent. Next, I will show that the common approach to learning a reward function from non-experts in RLHF uses an inductive bias that fails to encode how humans express preferences, and that our proposed bias better encodes human preferences both theoretically and empirically. For inspection, humans must be able to assess the behaviors an agent learns from a given specification. I will discuss a method to find settings that exhibit particular behaviors, like out-of-distribution failures. Lastly, cognitive science theories attempt to show how people build conceptual models that explain agent behaviors. I will show evidence that some of these theories are used in research to support humans, but that we can still build better curricula for modeling. Collectively, my research provides evidence that—even with the best of intentions— current human-AI systems often fail to induce alignment; my research proposes promising directions for how to build better aligned human-AI systems.

Serena Booth received her PhD at MIT CSAIL in 2023. Serena studies how people write specifications for AI systems and how people assess whether AI systems are successful in learning from specifications. While at MIT, Serena served as an inaugural Social and Ethical Responsible Computing Scholar, teaching AI Ethics and developing MIT’s AI ethics curriculum that is also released on MIT OpenCourseWare. Serena is a graduate of Harvard College (2016), after which she worked as an Associate Product Manager at Google to help scale Google’s ARCore augmented reality product to 100 million devices. Serena currently works in the U.S. Senate as a AAAS AI Policy Fellow, where she is working on AI policy questions for the Senate Banking, Housing, and Urban Affairs Committee. Her research has been supported by an MIT Presidential Fellowship and by an NSF GRFP. She is a Rising Star in EECS and an HRI Pioneer.

---

UMBC Center for AI

DocuSign is experiencing a severe degradation today, which has impacted our forms and use of DocuSign envelopes. There's no ETA yet on resolution, and we'll keep you posted. 

The issue only impacts sending to DocuSign, not any of the other Travel functions. 

The learned behaviors of AI and robot agents should align with the intentions of their human designers. Toward this goal, people must be able to easily specify, inspect, and model agent behaviors. For specifications, we will consider expert-written reward functions for reinforcement learning (RL) and non-expert preferences for reinforcement learning from human feedback (RLHF). I will show evidence that experts are bad at writing reward functions: even in a trivial setting, experts write specifications that are overfit to a particular RL algorithm, and they often write erroneous specifications for agents that fail to encode their true intent. Next, I will show that the common approach to learning a reward function from non-experts in RLHF uses an inductive bias that fails to encode how humans express preferences, and that our proposed bias better encodes human preferences both theoretically and empirically. For inspection, humans must be able to assess the behaviors an agent learns from a given specification. I will discuss a method to find settings that exhibit particular behaviors, like out-of-distribution failures. Lastly, cognitive science theories attempt to show how people build conceptual models that explain agent behaviors. I will show evidence that some of these theories are used in research to support humans, but that we can still build better curricula for modeling. Collectively, my research provides evidence that—even with the best of intentions— current human-AI systems often fail to induce alignment; my research proposes promising directions for how to build better aligned human-AI systems.

Serena Booth received her PhD at MIT CSAIL in 2023. Serena studies how people write specifications for AI systems and how people assess whether AI systems are successful in learning from specifications. While at MIT, Serena served as an inaugural Social and Ethical Responsible Computing Scholar, teaching AI Ethics and developing MIT’s AI ethics curriculum that is also released on MIT OpenCourseWare. Serena is a graduate of Harvard College (2016), after which she worked as an Associate Product Manager at Google to help scale Google’s ARCore augmented reality product to 100 million devices. Serena currently works in the U.S. Senate as a AAAS AI Policy Fellow, where she is working on AI policy questions for the Senate Banking, Housing, and Urban Affairs Committee. Her research has been supported by an MIT Presidential Fellowship and by an NSF GRFP. She is a Rising Star in EECS and an HRI Pioneer.

The next CyberPaws CTF challenge has been released! Please submit your flag to the "Prize Pup Bot" on our Discord via DMs, using the `!submit` command! You have until our next meeting to submit. The flag will always be in the following format: Paws{ex4mpl3_fl4g}.

Bb shells for SU2024 are available for all courses (designated as lecture, lab, field or discussion) listed in the UMBC Schedule of Classes (SOC), provided that there is an instructor of record listed. As new course sections and/or instructors are added to the SOC by Department Scheduling Coordinators, corresponding Bb shells are automatically generated overnight. 

Any t-shirts leftover from the DC trip are now free to take for anyone! It's a first come first serve!

Drop by Lydia DeRidder's office, PAHB 451, to grab one! It is suggested to email Lydia at lderidde@umbc.edu prior to stopping by to make sure she's in her office!

The following new OSP procedure presented below is also posted on our OSP website under "Policies and Procedures>Research Procedures>Office of Sponsored Programs:

Purpose/Objective: 

• Per the NSF Proposal and Award Policies and Procedures Guide (NSF PAPPG 24-1), “NSF requires project reports for all assistance awards”.
• Project Reports include Annual Progress Reports, Technical Reports, and Other Reports-both interim and final.
• This procedure aims to provide guidance and instructions to PIs and research administrators on UMBC’s campus in preparing timely submission of reports to NSF and alerting them of the consequences if they do not adhere to the specified deadlines.

Background:

Meeting deliverable requirements is a critical component of sponsored awards. In particular, timely submission of all Reports – and, in particular, Annual Progress Reports, Outcomes Reports, and Final Technical Reports -- are an integral part of the terms and conditions of the Award. The Principal Investigator (PI) is responsible for ensuring all technical deliverables/reports are submitted on time to NSF (with a copy provided to OSP as well) by the specified due date.

Overdue reports have come under greater scrutiny recently and require greater diligence and compliance to ensure that ALL reports are submitted on time.  Failure to submit a Report by the deadline may result in NSF taking any and/or all of the following actions that may have explicit negative consequences and impacts for the PI and Co-Is on the Project, other UMBC Departments, and in extreme cases, the University as whole:

• Non-review of pending proposals
• Non-issuance of new awards or award modifications (withholding funding)
•  Increased risk for Site Visits and Site Reviews by Sponsor
•  Increased Risk of Sponsor Audits

Procedure For All NSF Reports:

Project reports must be submitted by the PI in Research.gov no later than 90-days prior to the end of the current budget period to provide sufficient time for review and approval by the cognizant NSF Program Officer. The report becomes overdue the day after the end of the current budget period if it has not been submitted by the PI and approved by the cognizant NSF Program Officer.

NSF currently sends out automated notifications (from email: reporting@nsf.gov) around the first day of each month.  These notifications alert the PI and OSP of pending due dates for various Reports, including Annual Progress, Outcomes, and Final Technical Reports.  These emails are sent at the 30-, 60-, and 90-Day mark prior to the end of the current budget period.  Notifications are also sent by NSF in the event that your Report is overdue. The Due Date for the Report is one (1) day after the end date of the current budget period.  These automated notifications provide the impetus for the PI to submit their Reports. 

Additionally, a running list of all NSF Reports and upcoming overdue dates are  provided on Research.gov (navigate to the “Awards & Reporting/Project Reports” page) and can be accessed by OSP personnel at any time.  The OSP Manager will follow-up with the PI if they fail to submit the Report by the stated due date.

Please consult and bookmark the following link to the NSF Proposal and Award Policies and Procedures Guide (NSF PAPPG (NSF 24-1) as it pertains to NSF Reporting Requirements (see NSF 24-1  Chapter VII.D). Additionally, please see the Uniform Guidance (2 CFR Part 200) and be compliant with the Federal Annual Reporting Guidelines as specified in 2 CFR 200.328.

Procedure & Consequences for Submitting Late NSF Final Reports:

If a Final Technical Report is due,  the OSP Close-out Manager will coordinate with the PI’s OSP Grants and Contracts Manager to ensure that the PI is not being unnecessarily emailed twice and to clarify any potential extenuating circumstances.

1.      Upon becoming aware of a past due technical report or at 30 days past due, the OSP Grants and Contracts Manager will email the PI with a “cc” to the unit business office that the report is past due, and should be submitted asap.

2.      At 60 days past due, a reminder will be sent to the PI, cc’ing the unit business office and the unit Chair or Director, and the Associate Dean for Research.

3.      At 90 days, the OSP Grants and Contracts Manager will forward the previous PI email and notify the OSP Director, who will send an email to the PI cc’ing the unit business office, the Chair, the Associate Dean for Research, and the Dean as well as the Associate Vice President for Research (AVPR). The OSP Director will indicate that if the report is not submitted within the next 30 days, it may lead to the suspension of OSP processing any award actions for any of the PI’s awards until the overdue report has been cleared.

4.      At 120 days, the ORCA AVPR will email the PI cc’ing the unit business office, the unit Chair or Director, the Associate Dean for Research, and the Dean about enforcement action that may include suspension of award actions by OSP.

A Special Note on the Submission of FINAL Reports:

From 2 CFR 200.344: “All financial, performance, and other reports as required by the terms and conditions of the Federal award must be submitted no later than 120 calendar days after the end date of the period of performance. A subrecipient must submit to the pass-through entity, no later than 90 calendar days (or an earlier date as agreed upon by the pass-through entity and subrecipient) after the end date of the period of performance, all financial, performance, and other reports as required by the terms and conditions of the Federal award”.

As reflected in the Project Report System, the report is considered due during the 120-day period. If the report has not been submitted by the PI and approved by the cognizant NSF Program Officer within the 120-days period, the report becomes overdue.

Submitting late FINAL Reports may have additional implications as specified in 2 CFR 200.344(i),  there is a failure to submit all reports within one year of the period of performance end date, , the Federal awarding agency must report the non-Federal entity's material failure to comply with the terms and conditions of the award with the OMB-designated integrity and performance system (currently FAPIIS). Federal awarding agencies may also pursue other enforcement actions per § 200.339.

It is imperative that ALL FINAL Financial, Technical, and Other Reports are submitted by the specified Due Date!

----------------------------------------------------------------------------------------------------------------

• For instructions and guidance on how OSP manages late FINAL TECHNICAL REPORTS FOR ALL OTHER SPONSORS, please see the instructions here.
• For email templates that OSP can utilize for late FINAL TECHNICAL REPORTS FOR ALL OTHER SPONSORS, please visit this link.
• Please reference the latest NSF PAPPG (NSF 24-1) for all Reporting Requirements.