There is widespread public attention to AI’s potential impact on jobs. Everyone is asking questions like, could AI eliminate massive numbers of jobs? How might it change the nature of many jobs? Will it dramatically restructure the labor market? But despite the high level of public interest in these questions, current AI transparency efforts do not cover AI’s labor impacts.
Documentation is one of the only tools that the AI industry & AI research field have widely agreed upon for the assurance of safe and responsible AI systems. But impacts on workers are largely absent from transparency and documentation efforts. In PAI’s latest report, “Workers Participating in Transparency: Addressing the Gap in AI Transparency on Labor,” we explore how to close this gap.
Addressing the Labor Gap in AI Transparency Efforts Panel Discussion
The Importance of Including Workers in AI Transparency Efforts
First, it’s important to understand what we mean by transparency. There are two important concepts of transparency to consider: human-centered transparency and transparency as a process.
The concept of human-centered transparency developed by Liao and Wortman Vaughan starts from the key questions of who and what transparency is for. Transparency is fundamentally for the sake of human understanding, and different stakeholders will have different needs to support their understanding. Critically, Machine Learning (ML) lay users will have different needs than ML experts. Workers are important users and impacted people of ML systems. From a human-centered transparency framework, it’s important to include them and take into account their specific needs.
The concept of transparency as a process, not just an artifact, comes from PAI’s ABOUT ML reference document. Transparency is a process that involves stakeholders in an ongoing critical process of asking, answering and documenting questions about a product and the potential impacts of choices in its design, development and deployment. Workers who are included in this participatory process have key insights to offer as “domain experts” with first-hand experience.
Broadly speaking, worker inclusion can benefit companies. The Ford Foundation, informed by a multisectoral group of corporate leaders, as well as PAI, released a report called Listen to Lead: Raise Retention and Boost Business. The report concludes that engaging workers by listening to them, taking action and being accountable can result in lower turnover, better productivity and more revenue.
Worker involvement in the development and adoption of technology also brings benefits. MIT Sloan School of Management professor Thomas Kochan has a body of research across multiple industries showing that incorporating end users, such as workers, into technological development and deployment results in better products, better implementation, and better jobs. Kochan et al. argue in a recent study that generative AI provides an even bigger opportunity for including worker voice in beneficial ways. As Japanese manufacturers described their philosophy of worker participation in introducing new technology, “it is workers who give wisdom to the machines.”
Including Workers as an Audience and Topic of Transparency
There are three pathways by which we can begin to address the gap in labor transparency – finding ways for workers to be included as a topic of transparency, an audience for transparency and participants in transparency.
One example of including workers as a topic of transparency is the system card for Open AI’s GPT-4. The system card is noteworthy for emphasizing that the impact of GPT-4 on the workforce should be “a crucial consideration” for policymakers and stakeholders. It delves into the potential impact of the model on job automation, job quality and inequality. Open AI also provides transparency about its labor practices for data workers that contribute to the model.
“… engaging workers by listening to them, taking action and being accountable can result in lower turnover, better productivity and more revenue.”
Another example where workers are an audience of and participants in transparency is the model fact sheet for a healthcare tool called Sepsis Watch. Part of Sepsis Watch is a machine learning tool that helps diagnose sepsis, a serious infection that is the leading cause of inpatient death in US hospitals. Healthcare workers helped design the model fact sheet, which resembled a pharmaceutical drug warning label, to convey key information to frontline workers using the tool.
Union Collective Bargaining Offers a Model for Participatory Processes
Collective bargaining is a participatory process by which workers in a union come together as a group and negotiate their working conditions with their employer. Workers help shape the union’s bargaining proposals and ratify the final contract. According to the International Labor Organization, one in three employees in 98 countries are covered under a collective bargaining agreement. As a widespread mechanism, collective bargaining provides an opportunity to address the use of technology in the workplace – see Lisa Kresge’s paper on union bargaining around technology for many examples. For instance, workers have the right to make information requests to the employer during bargaining that could provide greater transparency about technological changes.
Some unions and employers form labor management committees or partnerships that can engage in study or planning around technological change. The Kaiser Labor Management Partnership is one example in the healthcare industry. This union and employer partnership is charged with studying job trends and changing skills required by new technology and promoting job security and workforce training in response to forecasted changes. Such collaboration can also provide a model where workers and employers together can push for greater transparency around labor impacts from model developers.
There is much more we can do to improve labor transparency by including workers as a topic, audience and participants in transparency. Our latest report offers a few examples and explores where further experimentation and research are needed. However, shifting the paradigm and treating transparency as a human-centered process of critical inquiry and documentation can bolster transparency efforts beyond just labor transparency. Taking action to include workers as a topic, audience and participants in transparency ultimately helps us move transparency efforts from a checklist of different artifacts to a process for shared governance that empowers workers.