At Submittable, our mission is to accelerate mission-driven work. Innovation is central to that acceleration. Breakthrough technologies such as artificial intelligence represent enormous potential for humanity as a whole, and especially the social impact sector.
Submittable has the opportunity to create products that empower our customers to deliver exponentially more impact across their programs and beneficiaries. The deployment of our AI tools are a part of making progress on a variety of vital causes across public, nonprofit, and private sectors, from affordable, community-based child care to fighting and building resilience to climate change.
As part of this work, we have a duty to ensure all AI technology we develop is responsible, ethical, and impactful. To that end, Submittable adheres to a set of principles for responsible AI.
Submittable’s Principles for Responsible AI
Submittable first released AI-powered capabilities in the spring of 2021 to support our clients expecting tens or hundreds of thousands—even a million—applications, in response to primarily highly in-demand COVID-relief funding. With Automated Review, the technology takes care of tedious tasks, like extracting data from tax forms or identification documents, or applying automated scoring based on custom-defined business logic.
This opens up space for our customers to focus on program-critical work that requires a human touch—like building relationships with applicants and awardees.
Since then, our science team has grown. We’re hard at work unveiling further products that pursue the mission of AI for good. This new wave will make it possible for each stakeholder in the grant lifecycle to harness the power of AI—not only the grantmakers, but also applicants, reviewers, and others.
Today, on the precipice of the next phase of AI for Submittable, we’re sharing with our community the responsible AI principles that are guiding this work.
We believe AI’s role is to help humans make better decisions and achieve more—but not at the expense of human agency. While our AI tools will help work get done faster, they’ll never cut humans out of the equation. To this end, we work to create AI tools that eliminate busywork, while centering the role of the human brain in key tasks like decision making or judgment. Our AI tools will preserve the right for the human to make the final decision.
All of our AI tools are accountable to a human being that’s constantly kept in the loop. “Human-in-the-loop” (HITL) monitoring means that humans provide direct feedback to the learning model to correct errors and improve accuracy over time. We use HITL to both actively monitor our AI tools and perform ongoing impact evaluations, so that our models improve over time.
We believe all users should have a clear understanding of where and when AI is used—and why. That’s why we make it clear when users interact with AI tools, where we implement AI in our software, and why we are doing so. This transparency is foundational for both trust in the tools and setting expectations for how the tools work.
We work tirelessly to find and eliminate bias in our AI tools. We perform extensive early bias identification and continuous monitoring. This allows for early and constant mitigation leading to more equitable tools in the short and long term. We also take it a step further, by engaging expert stakeholders in model development so that our technology reflects not just our own interpretation of equity, but adheres to industry-leading standards.
Private and secure
At every step of our development lifecycle, we implement world-class measures to ensure all data is private and secure. We will never share data with public models. Users can trust that the data that moves through Submittable’s AI is protected and safe.
Our science process
As we develop our AI tools, the Submittable science team takes an intentionally measured approach, as depicted in the flywheel below.
- Research: Work with subject matter experts to understand the problem we are trying to solve and appropriate ways to solve it.
- Design: Ensure that the solution will enhance user experience while meeting our Responsible AI Principles.
- Build: Create a performant stable service that meets all security requirements.
- Evaluate: Ensure that the solution is performing as expected.
- Improve: Look for and evaluate areas of improvement.
Submittable embraces the non-deterministic nature of AI and algorithmic development. A non-deterministic algorithm is one that, when given the same inputs, may return different outputs. This means that we will often not know the results of our work until it’s complete. We consider this a strength.
In practice, this means we operate within confidence intervals, which indicate the degree to which we can expect outputs to be accurate. To create a confidence interval, we define an upper and lower bound, which is the highest accuracy we can expect and the lowest accuracy we will allow. Then we use HITL methodology, among other best practices, to ensure that the outputs of our AI tools fall within these upper and lower bounds.
By following this best practice method for AI development, we’re able to both ensure accuracy and embrace the dynamic nature that gives AI tools their unique value.
What’s next for AI at Submittable
We’ve already delivered AI that helps our customers automate tasks and improve decision making. Coming next, we’re excited to get AI into the hands of more customers, with enhanced usability for non-technical individuals.
A key part of this journey is our recently announced strategic partnership with Microsoft’s Tech for Social Impact. Joining them in their mission of creating AI for good, we’ll be leveraging Microsoft’s best-in-class suite of tools, including Microsoft Azure Open AI, to create a new wave of AI-powered technology. This new wave will unleash the promise of AI to every member in the grantmaking process.
Here’s a sneak-peek at what’s on the horizon for Submittable users:
- Time-saving answer suggestions for grant seekers: We’re working on an AI-powered service that will auto-fill suggested answers to grant applications, informed by answers the applicant themself has previously provided.
- Faster form creation for grantmakers: Informed by our vast knowledge of best practices among practitioners, our customers will be able to use an AI bot to automatically create grant application forms using natural language prompts.
- Accessible applications across language barriers: We’ll also help extend grant opportunities across diverse populations with an AI bot that will translate forms into multiple languages.
- Accurate data extraction from documentation: We plan to eliminate tedious data entry for applicants and reviewers, while also reducing the potential for human error and fraud with an AI-supported tool that can extract information from official documents, such as W2s.
AI for good requires intention
We’re beyond excited for what the future holds for our customers and philanthropy practitioners and beneficiaries across the globe. But we are proceeding carefully. As we release revolutionary technology to the sector, we’re laser-focused on ensuring that in practice, the technology we develop furthers our purpose to accelerate mission-driven work.