The Legal Innovation & Technology Lab's Spot API Beta
@ Suffolk Law School - Spotter Updated: 2021-05-03 (Build 7)
Spot is an AI-powered issue spotter built to promote access to justice. As such, we see sharing Spot as an opportunity to develop and test model frameworks that creators of conscience may use to help mitigate the risks of harm presented by machine learning tools. While not a replacement for future regulation of such tools, our model, based in trust law, aims to provide a concrete enforceable approach creators can adopt to promote a stated mission and mitigate harm. To understand why we suggest the use of trusts, consider the following.
Spot can only recognize a statement of an issue that resembles those it has seen before. If Spot's training data is sourced from one population and used on a different population, there will be differences in how each group talks about issues that will affect Spot's ability to identify them. Additionally, different populations may experience different issues at different rates, skewing their prevalence in training data. Our best defense against such bias is to make sure that we are using data from the populations we aim to serve. One way to do this is to turn the use of Spot into a virtuous cycle, where it learns from the people it is helping so it may better help those in the future with similar issues. Such data sharing, however, is not always appropriate, and when appropriate, the choice to share such data should be based on trust. That is, those providing data to train Spot should be able to trust that their data will be used in a manner they support. Such trust requires an understanding of how Spot will be used. It requires an understanding of what uses are permitted, what uses are prohibited, and how violations of expectations will be addressed.
The open-source movement has proven the potential impact of broadly shared no-cost software including its ability to foster innovation and collaboration. A limitation of traditional software licensing, including open-source, however, is that it creates duties only between parties to the license (e.g., the software author and developers making use of the software). Machine learning tools such as Spot derive much of their value from broad community involvement while impacting these same communities as part of their use. This role suggests such communities should have a seat at the table. We propose a trust structure in part because of its ability to create enforceable responsibilities to this broader community.
To help build a community of shared purpose around Spot, we are collaborating with Duke Law’s Digital Governance Design Studio to develop a “click-trust” to govern the use of Spot’s API and a planned Python library. A trust is a legal device to manage assets on behalf of a person or a community. A trustee has a legal duty to use the trust’s assets—whether digital or physical—to support the interests of the beneficiaries.
Here, rather than licensees, users of the Spot API/library would be asked to become trustees of the Spot community. In agreeing to the terms of the trust, Spot users would need to affirm that their use supports Spot’s mission and beneficiary communities, and that they won’t pursue harmful secondary uses. Trustees will be encouraged, but not required to contribute data back to Spot.
Unlike a license, which creates duties only between those agreeing to it, our trust would create a duty between an API/library user and the Spot community and its access to justice mission. It enables members of the beneficiary community to enforce the trust terms and offers equitable relief against inappropriate uses of Spot. We think this strikes the right balance of empowering people to build creative tools and businesses that use Spot, while protecting against harmful secondary uses of Spot.
Ultimately, we hope we never have to enforce the Spot trust. Our goal is to use the trust to set norms for the nascent Spot community, and to give future contributors to Spot a voice in how the tool is developed and deployed. While this may be overkill for Spot’s current incarnation, we hope this can serve as a foundational building block for a community built around Spot, where access to justice organizations and stakeholders contribute data to collaboratively develop tools that improve access to legal services.
We are seeking comment and feedback from prospective Spot community members. We will hold a series of live feedback sessions in June, and open an asynchronous comment period in July 2021. You can register for the feedback sessions below or sign up to be notified when the comment period opens. This will be followed by additional engagements with potential end users (e.g., legal aid clients).