<aside>

WORKING DRAFT FOR PUBLIC FEEDBACK For more context on this draft, please see here. Please submit feedback here.

</aside>

Previous: Platforms for Scalable Surveys

Table of Contents

Next: Who These Guidelines Are For

What This Resource is For

This resource is aimed at people working directly on the design and development of AI-driven technological solutions - especially within the commercial, corporate, or for-profit sector. It provides guidance on how to build collaborative working relationships with organizations representing, and individuals of, socially marginalized communities. “Socially marginalized” is meant to be defined within specific localized contexts, as different cultures and parts of the world may have different dimensions of difference shaping its social hierarchies and dynamics. So, while the Global Task Members responsible for the initial draft are predominantly based in the United States and United Kingdom, this resource is drawn from both their expertise having worked with communities outside of the “West”, as well as their own lived experiences. As such, it is meant to serve as a guide for working with stakeholders in these countries and in very different cultural contexts, such as stakeholders based in Global Majority nations.

There are many different ways to solicit feedback from employees, clients, customers, users, or members of the general public.

• However, they do not all yield the same kind of insights, relationships or benefits. These guidelines get into some nuances around building trust-based, mutually beneficial relationships with different types of stakeholders. How stakeholders are engaged - including how they feel throughout the engagement - is just as important as whether they are engaged. Unfortunately, it is possible to build technology that may be socially beneficial, but built in a way that is exploitative and harmful. It is also possible to establish an ethical working relationship with stakeholders, yet fail to engage them in a way that leads to the development of a useful AI system.

Given the complexities of both developing AI systems and engaging stakeholders and the broader public, the resource cannot address every possible pain point and scenario. The intention is to provide a broad-but-specific framework that can support practitioners in better understanding their own unique circumstances under which they are trying to integrate greater stakeholder input and navigate the common challenges that arise so that the stakeholder engagement process they design and implement remains authentic and as equity-oriented as possible. There are substantial limitations to how much benefit can be generated and how much harm can be mitigated, but the framework should support individuals’ efforts to bring their work into greater alignment with the needs of the communities they wish to serve, while reducing the likelihood of harms and risks those communities may face.

Finally, it cannot be emphasized enough that **stakeholder engagement is not a fix for broader social and economic issues related to AI and other digital technology**, especially their impact on socially marginalized communities. Working with communities can help developers better understand the harms, risks, and failures (as well as possible opportunities and positive impacts), but it also cannot eliminate or lessen harms that are sometimes inherent to the technology or the problems that the technology is trying to address. For example, there is growing evidence that operating large-scale AI systems comes as a major environmental cost (e.g., the water needed to cool computing systems); the environmental damage cannot be undone by conducting stakeholder engagement activities, but it may be possible to better understand the expansiveness of the environmental impact and jointly make difficult decisions about how to oversee and govern the use of AI systems to mitigate the environmental damage.

What We Mean By “Stakeholder Engagement”

• Stakeholder refers to individuals and/or groups of people who are affected or may be affected by (but are usually not directly involved with) the design and development of the AI-driven product, service, or system. Depending on the circumstances, these may be individuals who are likely to be end users of a given AI system or product (affected end-users), those who are unlikely to be users (excluded individuals or communities), and those who are even unaware they are interacting with AI systems (impacted non-users).
  • For these guidelines, we are specifically concerned with stakeholders who exist outside of the organization or company (external). We also focus on those who work with, or consider themselves a member of, a socially marginalized group, as they require particular focus having been, and continue to be, excluded from benefits and opportunities made possible through innovation. In some cases, they may experience the worsening of systemic harms due to the development and deployment of AI.
  • While a commonly used term, it should be noted that the word “stakeholder” - in contexts affected by colonialism, especially concerning Indigenous groups and marginalized communities - underscores a ‘language of power’ that perpetuates rather than disrupts colonial logics, inequalities and injustices, as it origins go back to when colonial settlers “staked” a claim on indigenous land. Some alternatives to the term “stakeholder” that could be used, depending on the context, include: partners, rightsholders, participants, citizens, communities, collaborators, and interested or relevant parties.
• Engagement refers broadly to activities undertaken by the developer, development team, or developing organization/company in order to solicit direct or indirect input from stakeholders for the benefit (e.g., risk management, product improvement) of their company/organization or AI-driven product, service, or system.
• Company-led refers to stakeholder engagement that is initiated, designed, implemented, and financially sponsored by an entity (individual, team, or organization) that primarily operates in the for-profit, private sector (one of their primary objectives is to be financially solvent and profit-generating).
• The version of “stakeholder engagement” being prescribed in this resource is one where decision-making is increasingly shared with the people whose lives will be ultimately impacted and transformed by the development and deployment of the technology and one where how something is developed is as important as what is developed.

The Current Situation

Pace of Development & Deployment

There is a rapidly growing amount of investment going to AI development driven by competition between companies and countries. Technology companies seek to build more advanced systems that reach global audiences. Countries are in a geopolitical race in a race towards AI hegemony by advancing AI innovation in their countries through large investments and different AI governance policies. This has led to the hasty deployment of a broad range of AI products. This pace increases the risk of deployment without sufficient safety mechanisms in place before, during, and after a system is released. Stakeholder engagement is a key part of ensuring these systems are relevant and, at the very least, avoid harm. However, the pace of development risks leaving meaningful stakeholder engagement out of the process.

• While the focus of this resource is towards commercial/private sector AI developers, the guidance may be applicable to the many other developers and deployers working across the public sectors (e.g., civil society organizations, government agencies) who are also seeking to use AI tools for social benefit. Adaptations and additional use cases may be necessary to enhance the relevancy of the guidance and guardrails for the different dynamics and circumstances that exist for these different sectors.