impact-assessment reflection-discussion sustainability
"SustainabilityDesigners and users of AI systems should remain aware that these technologies may have transformative and long-term effects on individuals and society. In order to ensure that the deployment of your AI system remains sustainable and supports the sustainability of the communities it will affect, you and your team should proceed with a continuous sensitivity to the real-world impacts that your system will have.Stakeholder Impact AssessmentYou and your project team should come together to evaluate the social impact and sustainability of your AI project through a Stakeholder Impact Assessment (SIA), whether the AI project is being used to deliver a public service or in a back-office administrative capacity. When we refer to ‘stakeholders’ we are referring primarily to affected individual persons, but the term may also extend to groups and organisations in the sense that individual members of these collectives may also be impacted as such by the design and deployment of AI systems. Due consideration to stakeholders should be given at both of these levels.The purpose of carrying out an SIA is multidimensional. SIAs can serve several purposes, some of which include:(1)Help to build public confidence that the design and deployment of the AI system by the public sector agency has been done responsibly(2)Facilitate and strengthen your accountability framework(3)Bring to light unseen risks that threaten to affect individuals and the public good (4)Underwrite well-informed decision-making and transparent innovation practices(5)Demonstrate forethought and due diligence not only within your organisation but also to the wider publicYour team should convene to evaluate the social impact and sustainability of your AI project through the SIA at three critical points in the project delivery lifecycle:1.Alpha Phase (Problem Formulation): Carry out an initial Stakeholder Impact Assessment (SIA) to determine the ethical permissibility of the project. Refer to the SUM Values as a starting point for the considerations of the possible effects of your project on individual wellbeing and public welfare. In cases where you conclude that your AI project will have significant ethical and social impacts, you should open your initial SIA to the public so that their views can be properly considered. This will bolster the inclusion of a diversity of voices and opinions into the design and development process through the participation of a more representative range of stakeholders. You should also consider consulting with internal organisational stakeholders, whose input will likewise strengthen the openness, inclusivity, and diversity of your project.2.From Alpha to Beta (Pre-Implementation): Once your model has been trained, tested, and validated, you and your team should revisit your initial SIA to confirm that the AI system to be implemented is still in line with the evaluations and conclusions of your original assessment. This check-in should be logged on the pre-implementation section of the SIA with any applicable changes added and discussed. Before the launch of the system, this SIA should be made publicly available. At this point you must also set a timeframe for re- assessment once the system is in operation as well as a public consultation which predates and provides input for that re-assessment. Timeframes for these re-assessments should be decided by your team on a case-by-case basis but should be proportional to the scale of the potential impact of the system on the individuals and communities it will affect.3.Beta Phase (Re-Assessment): After your AI system has gone live, your team should intermittently revisit and re-evaluate your SIA. These check-ins should be logged on the re- assessment section of the SIA with any applicable changes added and discussed. Re- assessment should focus both on evaluating the existing SIA against real world impacts and on considering how to mitigate the unintended consequences that may have ensued in the wake of the deployment of the system. Further public consultation for input at the beta stage should be undertaken before the re-assessment so that stakeholder input can be included in re-assessment deliberations.You should keep in mind that, in its specific focus on social and ethical sustainability, your Stakeholder Impact Assessment constitutes just one part of the governance platform for your AI project and should be a complement to your accountability framework and other auditing and activity-monitoring documentation.Your SIA should be broken down into four sections of questions and responses. In the 1st section, there should be general questions about the possible big-picture social and ethical impacts of the use of the AI system you plan to build. In the 2nd section, your team should collaboratively formulate relevant sector-specific and use case-specific questions about the impact of the AI system on affected stakeholders. The 3rd section should provide answers to the additional questions relevant to pre-implementation evaluation. The 4th section should provide the opportunity for members of your team to reassess the system in light of its real-world impacts, public input, and possible unintended consequences.
## Stakeholder impact assessmentStakeholder Impact Assessment for (Project Name)
1. Alpha Phase (Problem Formulation) General Questions- Completed on this Date:
1. Identifying Affected Stakeholders: Who are the stakeholders that this AI project is most likely to affect? What groups of these stakeholders are most vulnerable? How might the project negatively impact them?
1. Goal-Setting and Objective-Mapping: How are you defining the outcome (the target variable) that the system is optimising for? Is this a fair, reasonable, and widely acceptable definition?
1. Does the target variable (or its measurable proxy) reflect a reasonable and justifiable translation of the project’s objective into the statistical frame? Is this translation justifiable given the general purpose of the project and the potential impacts that the outcomes of its implementation will have on the communities involved?
1. Possible Impacts on the Individual: How might the implementation of your AI system impact the abilities of affected stakeholders to make free, independent, and well-informed decisions about their lives? How might it enhance or diminish their autonomy? How might it affect their capacities to flourish and to fully develop themselves? How might it do harm to their physical or mental integrity? Have risks to individual health and safety been adequately considered and addressed? How might it infringe on their privacy rights, both on the data processing end of designing the system and on the implementation end of deploying it?
1. Possible Impacts on Society and Interpersonal Relationships How might the implementation of your AI system adversely affect each stakeholder’s fair and equal treatment under the law? Are there any aspects of the project that expose vulnerable communities to possible discriminatory harm? How might the use of your system affect the integrity of interpersonal dialogue, meaningful human connection, and social cohesion? Have the values of civic participation, inclusion, and diversity been adequately considered in articulating the purpose and setting the goals of the project? If not, how might these values be incorporated into your project design? Does the project aim to advance the interests and well-being of as many affected individuals as possible? Might any disparate socioeconomic impacts result from its deployment? Have you sufficiently considered the wider impacts of the system on future generations and on the planet as a whole?
1. Alpha Phase (Problem Formulation) Sector-Specific and Use Case-Specific QuestionsCompleted on this Date: In this section you should consider the sector-specific and use case-specific issues surrounding the social and ethical impacts of your AI project on affected stakeholders. Compile a list of the questions and concerns you anticipate. State how your team is attempting to address these questions and concerns.
1. From Alpha to Beta (Pre-Implementation)Completed on this Date:After reviewing the results of your initial SIA, answer the following questions:- Are the trained model’s actual objective, design, and testing results still in line with the evaluations and conclusions contained in your original assessment? If not, how does your assessment now differ?- Have any other areas of concern arisen with regard to possibly harmful social or ethical impacts as you have moved from the alpha to the beta phase?- You must also set a reasonable timeframe for public consultation and beta phase re-assessment: Dates of Public Consultation on Beta-Phase Impacts: Date of Planned Beta Phase Re-Assessment:
1. Beta Phase (Re-Assessment)Completed on this Date:Once you have reviewed the most recent version of your SIA and the results of the public consultation, answer the following questions:- How does the content of the existing SIA compare with the real-world impacts of the AI system as measured by available evidence of performance, monitoring data, and input from implementers and the public?- What steps can be taken to rectify any problems or issues that have emerged?- Have any unintended harmful consequences ensued in the wake of the deployment of the system? If so, how might these negative impacts be mitigated and redressed?- Have the maintenance processes for your AI model adequately taken into account the possibility of distributional shifts in the underlying population? Has the model been properly retuned and retrained to accommodate changes in the environment?Dates of Public Consultation on Beta-Phase Impacts: Date of Next Planned Beta Phase Re-Assessment:"(Leslie, 2019, p.23-24)