CENTER FOR ADVANCING SAFETY
OF MACHINE INTELLIGENCEA collaboration between Northwestern University and UL Research Institutes

CASMI Mission and Program Background

The Center for Advancing Safety of Machine Intelligence (CASMI) was formed to address the problem of operationalizing the design, development, and evaluation of AI systems that are safe for individuals and society. A collaboration with the Digital Safety Research Institute (DSRI) at UL Research Institutes, CASMI builds on Northwestern’s research expertise and UL’s 128-year mission to create a safer, more secure, and sustainable future. 

Achieving the mission of operationalizing safety of AI requires work and expertise from many perspectives, disciplines, and fields. The breadth and depth of the skills needed to address the existing research gaps is larger than any one organization can confront alone. Because of this, CASMI is structured as a research hub to guide, support, and harness critical work done across sectors, institutions, disciplines, and geographies.  

To execute this vision, CASMI seeks to fund the initiation of projects aligned to its Research Roadmap. CASMI already has awarded and executed agreements to fund crucial research projects at six universities. As many vital research gaps and opportunities remain, CASMI is issuing this call for proposals to identify additional research partners to fund for critical work over the next two years. 

While the research work may be done anywhere around the world, it is unified in the CASMI vision of developing mechanisms for the evaluation, design, and development of machine intelligence that is safe, equitable, and beneficial for all who are impacted by its use.  

Award Information and Project Parameters

CASMI expects to make 6-8 new awards for individual research projects, contingent on receipt of competitive proposals. Each research project is expected to be funded for $250,000 up to a maximum of $275,000 over the term of the project. 

Awards for projects to be conducted at eligible external organizations will be executed as subawards of the master agreement for CASMI between UL Research Institutes and Northwestern University. Awards also will be made to fund specific research projects at Northwestern University under the existing master agreement.  

Researchers should anticipate starting their projects in January 2023 and completing the work within two years (no later than December 2024). Projects anticipated to be longer than one year in duration will be expected to demonstrate sufficient progress toward the projects’ objectives in the first year in order to continue to receive funding support for the remaining duration of the award. 

Research Scope and Priority Focus Areas

CASMI's Research Focus

CASMI’s research mission is to support and foster work on the mechanisms needed to design, develop, deploy, and evaluate AI systems that are safe for both humans and society. Our goal is to amplify the positive impact of these systems by developing the knowledge and techniques needed to reduce their negative effects. 

CASMI shares with many other organizations the core values of the responsible and ethical development of AI. Building on those core values, CASMI's goal is to operationalize these ideas into best practices, design patterns, tools and guidance that can be used by those building AI systems and those evaluating them. This work on the impact of decisions throughout the pipeline of AI – while developing data, crafting algorithms, and designing interactions – is at the center of CASMI’s approach to developing the science of safety with respect to AI.  

While this work includes a clear focus on technical issues within Artificial intelligence, Machine Learning and Computer Science more broadly, CASMI’s mission scopes beyond the systems themselves to how they are used and who they impact. As such, CASMI’s scope includes work on how humans interact with and are impacted by machine intelligence, drawing on fields such as Sociology, Cognitive Psychology, Anthropology and Behavioral Economics. Equally critical to establishing the science of safety in intelligent systems is the science of evaluation. Understanding the real-world impacts of intelligent systems necessitates understanding the ways in which these impacts materialize with respect to human life, society in general, and communities that are impacted in different ways. Therefore, the CASMI research mission includes research into the metrics, tradeoffs, and long- and short-term effects of AI deployments. 

Vital to any CASMI research work is the ability to further operationalize solutions to genuine problems in AI development and use that negatively impact human life and society. CASMI supports projects focused on outcomes – projects that will deliver results to be put into practice. Outcomes can take many forms, such as publications, datasets, open-source code sets, methods, training or teaching materials, tools, best practices, guidelines, metrics, benchmarks, compendia, etc. Deliverables and outcomes of CASMI projects are intended to be made publicly available to the extent possible in order to further their adoption. Regardless of the type of deliverables that may be produced, CASMI projects must have impact on the practice of developing and evaluating AI from the perspective of safety, through the development of new techniques or advancing safety science in this space.

Priority Areas

While the spectrum of potential research is broad, particular research areas of focus may be prioritized for this funding opportunity. Some examples of those priority areas, in alignment with the research structure defined in the Evaluation Framework, are:  

• Data: research into the causes of problems in the data pipeline that lead to downstream issues such as model bias, imbalanced representation, misalignment of data and task, and lack of transparency; and mechanisms for identifying and remediating them. 
• Algorithms: work on the impact of choices made and mechanisms used during model training such as choice of algorithm, feature selection, approaches to training/testing, and models of explanation to further the development of safe AI systems.  
• Human-Computer Interaction: studies of the ways in which interaction design can impact outcomes, work in the detection of dark patterns, work harnessing tenets of behavioral economics in system building, management of decision support systems, and impacts of mental models of machine performance. 
• Evaluation: models defining the impact of intelligent technologies on both individuals and stakeholder communities; research identifying short- and long-term impacts, managing trade-offs, and developing metrics and standards. We are particularly interested in work that explores how to integrate stakeholder and affected community input and evaluation for systems that have impacts beyond the reach of the individual user. 

This list of examples is not meant to be exhaustive. CASMI's research vision and roadmap are intended to be agile and to evolve with important ideas and critical work from many perspectives and points of view. CASMI seeks to support the most promising new and novel research that is aimed at advancing the mission of operationalizing the safety of machine intelligence.  

Many issues and negative consequences have resulted from AI and ML systems that perpetuate and exacerbate societal inequities, or that were developed without representation from or consideration for all impacted communities. Seeing this as a critical area for research, CASMI encourages proposals aimed at solutions to diversity issues that arise in the development of AI systems and approaches that further inclusion of all affected communities. We are looking for proposals that address key problem areas and needs for diversity and inclusion in the development of systems as well as projects that will incorporate considerations of diversity in their functioning or proposed work. We also are interested in proposals that focus on how to bring communities that are impacted by technologies into the development and evaluation processes to help shape positive impacts. 

Some research proposals, while potentially important to AI as a field, may be less impactful to the specific mission of CASMI and therefore less likely to be successful for funding. Projects that are less likely to be selected would include those that focus on building specific ML systems or instances, theoretical research in ethics or responsibility, theoretical research into formal characteristics of ML systems, or work in building or cleaning use-specific datasets. 

Proposer Eligibility

Proposals may be submitted by an eligible principal investigator at an institution of higher education or an organization with a primary research focus. As part of the proposal package, the institution must provide a letter of support signed by an authorized institutional official which demonstrates support for completion of the proposed project should it be awarded. The letter of support should confirm that the named principal investigator is eligible and authorized to commit the designated effort to the proposed research initiative. Awarded institutions will be expected to provide necessary documentation in order to enter into a subcontract agreement with Northwestern University under the stipulations of the master agreement between UL Research Institutes and Northwestern University.  

There is no limit to the number of proposals that may be submitted from a single institution or organization. However, each proposal must identify a single principal investigator; an individual may serve as the PI on only one proposal.  

Proposals may also be submitted by Northwestern eligible PIs, as determined by the Northwestern Office for Research. An individual Northwestern PI may submit only one proposal.   

Proposal Requirements and Deadline

A complete proposal package requires the following materials:  

• Statement of Work  
• Budget  
• Budget Justification  
• Biographical Sketch(es)  
• Letter of Intent

See the Proposal Requirements section for explicit detailed guidance on developing each element of the proposal package.  

Proposals are due by Monday, September 26 at 11:59 pm GMT (UTC+0). 

Review Criteria and Selection Process

Proposal submissions will be reviewed by the CASMI team for eligibility and completeness. 

The CASMI Executive Committee will then request review of all eligible proposals by an advisory review panel to recommend proposals for funding to the Executive Committee. Proposal review and recommendation will be based primarily on the following key factors: 

• Intellectual merit of the proposal 
• Importance or urgency of the problem that the proposal would address 
• Broader impacts of the work, including the approach to diversity and engagement of affected communities 
• Connection to the overall CASMI research vision and mission, and demonstration that the work furthers the path toward operationalization of safe machine intelligence 
• Likelihood of the work to lead to successful outcomes and useful deliverables.  

The CASMI Executive Committee will ultimately determine the portfolio of projects to award in order to best further the CASMI mission, to build and broaden the research network, and to progress crucial work with distribution across the spectrum of the research vision.  

Anticipated Award Notification Timeline

Additional Opportunities and Expectations for Principal Investigators

In leading a CASMI-funded research project, Principal Investigators will have full autonomy to execute their research initiative, and the opportunity to engage in the research network it provides. PIs will provide quarterly reports of progress to the Executive Committee and periodic updates to the Governance Advisory Committee. A key objective of CASMI is disseminating the outcomes of this crucial research work and seeing those outcomes incorporated in practical applications. To that end, as deliverables are produced from CASMI projects, PIs are encouraged to work with the CASMI team to develop communications approaches to share the work with broad and diverse constituencies beyond research peers and academic publications.   

PIs are expected to plan to attend CASMI semiannual workshops, anticipated to be hosted at Northwestern’s campus in Evanston, Illinois, and may have opportunities to help guide the focus and development of those workshops. Proposal budgets should include travel costs to support attendance as noted in the proposal Budget guidance.  

CASMI is structured as a research network not only to further individual critical research efforts but also to magnify their outcomes as a collective. To that end, each participating institution will nominate an individual to serve on the Governance Advisory Committee. The Governance Advisory Committee provides critical guidance and advice on CASMI’s progress, direction and opportunities, and provides advice on collaborations and future programmatic direction of the Center. In addition, CASMI and its GAC intend to explore opportunities to provide additional support to early-career researchers, graduate students and research fellows who are part of the research network.  

Opportunities to Learn More

CASMI hosted virtual information sessions to discuss the CASMI research mission and proposal process. Researchers who are considering developing a proposal can watch an edited recording of the info sessions below.

Questions?

We encourage anyone interested in proposing a project to consider watching the virtual information session to learn more. If you have a specific question please reach out to the team at casmi@northwestern.edu.

Back to Top