ICTworks

As machine learning applications become more widely adopted, so does optimism for the ability for artificial intelligence to reshape the development sector, from automating labor-intensive tasks to offering new insights from complex datasets. However, with this growing opportunity comes the need to carefully assess when and if advanced algorithms are a good fit for any given project. Development practitioners need to understand if and how these new technologies can be used in each unique problem and context.

The USAID Practical Guide to Managing Machine Learning Projects in International Development will help highlight whether artificial intelligence or machine learning is an appropriate and relevant tool for the problem and context and ensure its effective and fair use.

This practical guide has been designed for development practitioners who may not be trained technologists but are involved with or responsible for implementing projects that might have a technical machine learning/ artificial intelligence component. It is informed by USAID’s previously released Making AI Work for International Development. In some instances, this guide incorporates images and language from that original USAID report.

How to Manage Artificial Intelligence Projects

This guide is less about development of a technical artificial intelligence model and more about management of a project that includes machine learning algorithms.

Following the phases of a project lifecycle, this practical USAID guide provides guidance and examples of implementations of artificial intelligence and machine learning with the goal of strengthening your understanding of how these technologies can be appropriately applied, integrated, and managed for impact.

• Evaluate Feasibility: Determine the underlying problem to be solved and whether advanced algorithms are the appropriate tool for you to use.
• Model, Design, and Build: Assess the proposed solutions, identify any risks associated with its implementation, and ensure your project aligns with best practice.
• Implementation: Monitor the performance of your application and whether the project is having the intended outcomes.
• Post-Implementation: Evaluate the performance of your project as a whole and take note of learnings that can be used in future projects.

Along with the project lifecycle modules, four key thematic areas are woven throughout, providing a framework for enhancing positive, responsible impact and sustainability:

1. Responsible, Equitable and Inclusive Design

This theme is the cornerstone of ensuring positive impact and is important to consider throughout the project lifecycle. It explores three principles that build on the responsible data movement:

• Responsible: How can you take appropriate measures to increase the likelihood of your project having the desired outcomes and proactively identify and mitigate any harmful outcomes to individuals or groups?
• Equitable: How can you ensure your project isn’t disproportionately benefiting or harming individuals or groups more than others?
• Inclusive: How can you include end-users and relevant stakeholders in conversation, observation, or co-design to ensure their perspectives are represented, their needs are properly addressed, and that this technology is designed with their interests in mind?

2. Strategic Partnerships and Human Capital

Technology is given meaning through the people who build and use it. This theme focuses on identifying in-house skills and strategic partnerships, from local knowledge experts to technical consultants, that you may need for your implementation.

3. Adaptive Management

This theme focuses on mechanisms for allowing continuous learning and iterative development given the unique needs and challenges of implementing artificial intelligence solutions. By building in mechanisms for adaptation, you will be better placed to achieve your project goals.

4. Enabling Environment

Lastly, this theme highlights enablers for sustainably implementing machine learning algorithms. These enablers may be externalities such as local infrastructure, legislation, or availability of data. This thematic area focuses on helping you identify whether the right elements are in place for your project to succeed.

Artificial intelligence is a relatively new field, and in many cases, we will be learning while actively shaping and developing the field as we go. In some cases, you may find that working to strengthen the enabling environment is a prerequisite or concurrent objective of your project that will facilitate responsible use of your model in the long-term.