Silva Ferretti, a colleague in international evaluation, has written an inspiring post on AI in evaluation that she has kindly allowed me to reproduce here. Sit back and enjoy the read!
>> I have been playing with Artificial Intelligence for some time now. I am amazed by it and actually surprised by the lack of debate regarding its role in development and humanitarian program management. Have I missed any discussions on this topic? If anyone has any information or pointers, I would greatly appreciate it. It is a game changer. We seriously should look into this NOW.
I learnt that:
It can write well-crafted logical frameworks and program concepts, as well as sectoral strategies, that are on par or even better than some real ones. It is able to anticipate risks and limitations, and propose detailed activities.
It is inclusive and politically aware, in a positive way. It has been trained to value inclusion and diversity, and is skilled at articulating ideas of participation and accountability, while also understanding that these ideas can generate conflict.
It is progressive and embraces a variety of methods and approaches. It can easily determine when rigorous/objective research is needed and when more constructivist methods should be used. It understands the advantages and areas of application for complexity-aware and feminist approaches.
It is creative and can use various communication styles. It suggested that conventional monitoring and evaluation methods may not be suitable for some programs and helped me generate anecdotes, commercials and even a rap song.
It excels at concepts, not facts. It does not provide references or links, and may sometimes confuse the names of standards or approaches. However, it understands the core concepts and can provide valuable insights. It is not a search engine, but a different paradigm.
What do I take from it?
1) the AI looks so good because a lot of developmental and humanitarian work is based on set approaches and jargon. We play by the book, when writing projects, when monitoring and evaluating change. This has advantages of course (we should not always reinvent the wheel!). But this is also where an AI works best. It is like these professionals good at making any project look cool, using the right words: nice, streamlined, even when reality is messy. And, sadly, what surfaces about many projects and programmes are just these sanitized proposals/reportings: confirmation of preset causal chains, with pre-set indicators… whilst local partners and change makers would tell more interesting and varied stories. It is the sanitized stories which eventually travels up the reporting chain, and into the AI of the future. This generates confirmation bias. And strengthens models accepted and established because we keep using them with the same lenses and logic. But reality is not like the blueprint.
2) the AI is more progressive than several professionals/institutions, in recognizing the whole field of complexity and complexity-driven approaches. Have a chat with it, asking what approaches are best in diverse contexts. It is adamant that participatory and empowerment processes require ad-hoc approaches. The lesson? That available evidence already indicates that there is not only one appropriate way to manage and evaluate (the bureaucratic/rigourous one). The fact that a machine understands the importance of the non quantifiable, of emergence, of feminist approaches – and some human managers don’t get it… – well, it makes me think a lot.
3) The AI can be really „creative“ when prompted. Try it out, and discover the many ways we could use to share the same concepts: poems, songs, riddles, conversations, anecdotes, stories. It is liberating, and a great way to free our own creativity and reach out to new audiences – when talking about change. It can add a whole new „communication dimension“ to monitoring, evaluation, and programming.
4) It is already happening. Artificial intelligence is not going to materialize in the far away future. You can do pretty decent work with it now. For routine tasks, including proposal writing, it is at least as good as a middle level officer needing supervision. How are we going to react? How should we use this tool? What will we teach to the next generation of professionals?
I am not afraid that AI can substitute humans. I am amazed, mesmerized by it. I find it stimulating. It provides, on a whim, the approach „by the book“ and free time to then go ahead, explore its details, the alternatives, the possibilities.
I fear that it can substitute humans as they are now used and considered by the development / humanitarian systems: bureaucrats applying tried and tested approaches, and having to confirm them using a standardized set of tools (which really do not allow to „go out of the box“). If this is the understanding, whole layers of professionals can be made redundant. Soon you can run your perfect programme, monitored by other AIs.
Which, to some extent, will (and probably also should!) happen.
What many people are already advocating re: monitoring, evaluation, learning is becoming more urgent. Do we have the tools to address the complexity ahead?
What is in that black box in between your question and AI response? How is the knowledge feeding it being built?
Our role will be, more and more, to check how ideas are formed, to explore mindsets behind the blueprints being formed, to check the boundaries of the issues we confront, to explore the links that are not yet there. Issues of interrelations, worldviews, boundaries, inclusion: it is the deep level use of systemic thinking that we should embrace, if we want to be enhanced, not trapped by the AI.
Am I optimistic? Not much, because the trend in the sector is for control, compliance, procedures. Even this list sometimes worries me when I see over and over the same debates about things that AI can already do better. But it consoled me that even the AI feels that humans still matter. In its words:
Humans have several unique abilities when it comes to understanding complexity that AI currently cannot replicate, here are some examples:
Empathy: Humans have the ability to understand and relate to the emotions, thoughts, and experiences of others. This allows them to understand the perspectives of different groups and to make decisions that take into account the needs and feelings of all stakeholders.
Intuition: Humans have the ability to make decisions based on their intuition or gut feelings, which can be particularly useful in complex and rapidly changing situations where there may be limited data or time to analyze information.
Creativity: Humans have the ability to think creatively and come up with innovative solutions to complex problems. They can think outside the box and come up with new and unexpected ideas.
Complex decision making: Humans can take into account multiple variables, such as cultural and political context, community dynamics, and the specific needs of different groups, and make decisions accordingly. They can also consider the ethical implications of their decisions.
Contextual understanding: Humans can draw on their experiences and knowledge to understand the specific context in which a humanitarian or development project is being implemented, and how it relates to the larger social, economic, and political environment.
It’s important to note that AI can assist in providing information and ideas, but the human ability to empathize, use intuition, think creatively, make complex decisions and contextual understanding are unique.In summary, my main function is to provide information and ideas, but the real work is done by the humans, who have the ability to understand the complexity of the situation, make decisions, and adapt to the context. Humans also have the ability to empathize and provide compassionate support, which is essential
All things that are recognized by complexity-driven, adaptive, participatory management practices. But are the development and humanitarian sector really run on this understanding?
Again… there are many urgent debates to have. And I would love to have pointers, resources, insights about them. Thanks in anticipation.
[If you wonder: I asked the AI if it agreed with the ideas shared in this mail „As a language model, I am not capable of having opinions or feelings. However, I can confirm that the statement you’ve provided is consistent with the capabilities that I have been trained on“. And, as I was there, I asked the AI to do some edits for readability – the initial bullet points. So, this mail is not co-written… but it is co-edited, discussed, validated. 🙂 ]