In a series of commentaries, ALNAP is exploring the use of AI in humanitarian evaluation practice. Whilst other commentaries have focused on the methodological pros and cons of using AI when collecting and analysing data in evaluation, one burning issue (no pun intended) has thus far been absent from the conversation.
Before I go into the impact AI has on the environment, I would like to start somewhere else. Last year, I co-authored ALNAP’s latest evaluation guide, Adapting the OECD criteria for the evaluation of humanitarian action. The development of the guide included extensive consultations with the humanitarian evaluation community. These conversations centered largely on evaluation criteria but naturally also included discussions on key issues for the international humanitarian system writ large. Not surprisingly, accountability to affected people and locally-led humanitarian action were themes that many raised as important. But the environment and climate crisis was also raised by several interlocutors as an issue that humanitarians need to grapple with, in evaluations, and in policy and programming. We therefore proposed to make the environment and climate crisis a priority theme in the guide. During the peer review process some reviewers opposed this, opining that these issues are not relevant for busy humanitarians who should focus on saving lives. As authors, we nevertheless decided to maintain the theme in the guide, as we firmly believe that the humanitarian system cannot ignore what is at stake. It needs to grapple with its own impact on the environment, and better understand the increasing humanitarian needs occurring as the climate crisis continues to escalate.
You might wonder what this has to do with AI. As mentioned in ALNAP’s recent Explain briefing on AI in the humanitarian sector https://alnap.hacdn.io/media/documents/ALNAP_Explain_AI_Briefing_052025_1_X3EUlVd.pdf, the use of AI has very real implications on the environment. Data centres that house AI servers are large consumers of water, so large in fact, that protests have taken place in several countries.1 The centres produce electronic waste, rely on unsustainably mined minerals and elements, and they use massive amounts of electricity, spurring the emission of planet-warming greenhouse gases.2
As humanitarian evaluators should we not take this into consideration when discussing when and how to use AI? As the climate crisis’ effects are increasing across the globe, how can we disregard the negative environmental effects of AI? Can we ignore how AI could potentially exacerbate what is already a dire situation?
In addition to the thoughtful perspectives presented by my evaluation colleagues in other commentaries, I would therefore like to add the aspect of prioritisation when deciding when to use AI in evaluation. Instead of casually using AI, let us prioritise its use to key tasks, where it can add real value. Let us make its use intentional, by choosing AI when it makes sense from a methodological point of view, and when it warrants its environmental cost. For some this might be never, for others more often.
