Project Management for Responsible AI

Project management is as useful in software development for AI applications as it is in other project domains.

A metaphor for the use of project management to achieve responsible AI. Not a literal view. Supplied by Daiki GmbH.

Artificial Intelligence (AI) is attracting a lot of attention. AI products like other software products are often delivered using project management. This is because the product delivery roadmap is often very complex.

What choices are there for software project management?

Many choices are available for software project management. Agile project management seems particularly useful for AI projects because it emphasises the importance of people involved in the project (on both sides: product development and management; and customers and other external stakeholders such as regulators). Given the rapid change in AI technology, the emphasis in agile project management in dealing with change and using it to support the delivery of value also seems to make agile project management relevant. Agile project values do not restrict you to a particular type of project management since agile values have informed different types of project management.

Agile software development

The Manifesto for Agile Software Development encompasses a number of values summarised in this image.

The values of the Agile Manifesto although produced some time before the current emphasis on responsible AI are still relevant today:

• Individuals and interactions emphasize the importance of interactions between the software engineers, project manager(s), product testers, product owner and senior managers during all parts of the project towards product roll out.
• Working software shows the importance of continual testing and interaction between development team members and the project manager before the AI product roll out to ensure that the emerging software is fully understood, and no unexpected faults arise.
• Customer collaboration reminds any organisation developing an AI application that the development is carried out for a customer, even if that is an internal customer. Clearly in many commercial contexts the customer will be external, business or consumer.
• Responding to change is particularly relevant while AI innovation proceeds rapidly and the most successful start-ups will seek to deploy products quickly. That doesn’t mean that there won’t be unanticipated changes, but that there should be a means of responding to them.

Why is agile project management of interest?

Can agile project management be used to generate responsible AI (AI where a person in the organisation developing it takes responsibility for it)?

Responsible AI has enhanced requirements because of the different ways in which people can interact with AI, and because of the different ways in which value can be obtained from it. Here these are compared with the focus of agile project management, and it is suggested that this is a useful means of managing projects designed to produce responsible AI. Agile project management is particularly useful because it does not refer to a particular type of project management, but has inspired a choice of techniques, including Scrum and Kanban. A project manager of responsible AI can choose one appropriate for their organisation and their project.

Putting people first

The Agile Manifesto refers to (see above): individuals and interactions and customer collaboration (my emphasis). Moving away from a rigid project process all the people involved in delivery of the AI product need to interact throughout, from project manager to the development team, product testers, product owner, senior managers and sales and marketing experts along with present and potential future customers to ensure that the project leads to success.

But responsible AI needs more. Any person involved needs to be able to ensure that the AI product:

• Has a respect for human autonomy.
• Does not cause harm.
• Acts fairly and does not discriminate.
• Produces transparent and explicable results.
• Uses data in a private and secure manner.

All these points refer to people associated with AI, in this case a project delivering an AI product. Different bodies of stakeholders may be involved in each case, but the sophistication of the technology does not avoid the requirement for people to understand what is happening. Other authorshave made the good point that these criteria should not just be achieved with respect to current stakeholders, but that they should be transferrable to any third party that becomes involved.

Generating value

The Agile Manifesto also refers to: working software and responding to change (my emphasis). This means continual testing of the AI product during its development, as well as interaction between the individuals involved in the development and management of that development to learn about the outcome of that testing. Testing doesn’t stop when the product is completed. There will need to be further testing by the software developers of the complete product to check its robustness and to test for unexpected behaviour. There will need to be testing with either potential customers of this product or actual customers of previous products to learn about the behaviour of customers with this new product. The information from these activities should inform sales and marketing activities.

Because this is a discussion of an AI product this feeds directly back towards all the issues that need to be considered under the heading of Putting People First above. But it also informs how the project manager delivering an AI product can deal with the substantial requirements for responsible AI and still be successful.

A successful project must include iterative development, adaptive planning, and flexibility so that change management is conventional rather than exceptional. In this way the project manager can deliver:

• Business value. Return on investment into the project.
• Innovation value. A product that advances insights compared to the competition.
• Social and environmental value. A product that reduces the energy and resource demand for its operation compared to the competition.
• Insight value. A product that delivers technical insights into the organisation for future development.
• Process value. Improvement in project management processes that can be more closely linked to the requirements of future projects.

Conclusion

Many techniques exist for software project management that can inform development of AI projects. Here the emphasis of agile techniques on putting people first and delivering value are suggested to be particularly appropriate for responsible AI because of the flexibility and adaptability they entail. It is for these reasons that current project managers of responsible AI can learn from agile project management.

This is a summary of an article produced for Dai.ki GmbH under the same title. The full version, including references, is available here: https://dai.ki/project-management-for-responsible-ai/. More about the services we provide through Coevolve IT is available here. We thank Mark Coeckelbergh for the initial introduction to Dai.ki GmbH, Jona Boeddinghaus and Mark Coeckelbergh for comments on earlier versions, and Julia Harrison for making the article suitable for presentation.