Artificial intelligence (AI)-powered applications are increasingly designed and deployed to support international conflict prevention and peacebuilding activities. Some data-intensive activities, such as conflict early warning, have seen steady uptake over several decades. More recently, online communication and social media in conflict dynamics, especially in shaping conflict narratives, spreading misinformation and disinformation and accelerating polarisation, has made it necessary to analyse large amounts of data to detect conflict risks and inform peacebuilding programmes and strategies, which AI now plays an irreplaceable role in.
The considerable diversification of AI technologies, especially large language models (LLMs) and generative AI, has led to a much wider spread of AI into areas traditionally viewed as more human-centred. For example, AI models are now beginning to be used to support peace mediation efforts, including public consultations and broad-based dialogues. Innovations such as AI chatbots have also made AI technologies more accessible to users without specialised skills in data or computer science. This means that AI can be used to shape peacebuilding efforts in spontaneous, unplanned and ubiquitous ways, including internal and public communication, research and knowledge management.
Unsurprisingly, AI is recognised as playing an increasingly important role in efforts to promote peace and security, including by the African Union (AU) and its partners. However, initial enthusiasm about the potential of the ‘AI revolution’ and AI’s contribution to promoting peace, security and stability has given way to a more realistic – and at times sceptical – outlook regarding AI’s prospects for supporting peace and security on the African continent. For example, the AU Peace and Security Council recently emphasised AI risks stemming from algorithmic bias, safety and data protection issues, the protection of vulnerable groups and a lack of compliance with human rights standards.
Many of these concerns can be crystallised into a single question: Is AI trustworthy enough to support conflict prevention and peacebuilding efforts? Asking whether AI usage in peacebuilding could be trustworthy is intuitive: building lasting peace entails building trust among the parties and groups in conflict – and in a digitised world where the relationships among these actors are mediated through digital technologies and AI, those technical agents must also be trustworthy.
For example, it is difficult to imagine experts who use early warning information to decide on preventive action and allocate financial resources, will do so without trusting the information. The same is true when AI is used to obtain, analyse or summarise insights about the needs, interests, opinions or positions of relevant conflict stakeholders. Such information will only play a positive role in a peacebuilding effort if all parties involved have no doubts about its quality or the methods through which it was generated.
On the other hand, naïve and unjustified trust can also have negative effects. For example, chatbots are commonly designed to respond with confidence, increasing the likelihood that we view them as reliable assistants when seeking information. Clearly, trusting chatbots simply because of their appearance and behaviour can be dangerous if inaccurate information or hallucinations are baked into their outputs. Overreliance on them can also erode the agency of human peacebuilders, for instance, in facilitating dialogue processes or shaping options for conflict resolution. It may also make conflict parties and critical stakeholders less inclined to do the hard work of identifying and articulating their grievances and demands, thereby negatively affecting the quality of peace processes.
While this makes clear that AI used to support conflict prevention and peacebuilding should be trustworthy, we are only beginning to understand how peacebuilders can ensure that AI can be trusted. Demands for AI trustworthiness are also likely to vary by peacebuilding activity, context and technology user. Many factors can thus shape the trustworthiness of AI in peacebuilding, including whether AI models produce accurate, transparent, fair, inclusive and safe outputs and whether they align with user behaviour. The recently established Norwegian Centre on Trustworthy AI has begun to study how and to what extent these factors matter. A preliminary look at past and present efforts to design and implement AI for peacebuilding, including on the African continent, already suggests three central and interrelated challenges.
Firstly, enhancing inclusivity: Digital technologies are often used as a vehicle to strengthen the inclusivity of peacemaking and peacebuilding efforts. However, AI-enhanced digital peacebuilding can also reinforce unequal power relations, create new barriers to participation, ownership and partnerships, leading to new exclusions. This has much to do with the global ‘gap’ in digital skills and AI capacities, which disadvantages actors in the Global South, as well as socio-economically marginalised groups, including women. While AI-enhanced peacebuilding tools may be used with the intention to enhance inclusivity, their outputs are unlikely do so if their design and mode of operation do not reflect the preferences and needs of those who participate.
While AI-enhanced peacebuilding tools may be used with the intention to enhance inclusivity, their outputs are unlikely do so if their design and mode of operation do not reflect the preferences and needs of those who participate
Tweet
Additionally, unequal global capacities to further develop AI heightens the risk that digital approaches will not reflect the needs and interests of all relevant stakeholders, especially since many of the leading generative AI models are proprietary, black-boxed and developed by American big tech. Nonetheless, inclusive AI4Peace design is possible, as demonstrated by Build Up’s efforts to develop classifiers for AI-enhanced analysis of social media content to support peacebuilding and community cohesion in places such as Kenya, with the involvement of conflict stakeholders.
Secondly, maintaining human-led efforts: AI-enhanced conflict prevention and peacebuilding may reduce human agency, including that of peacebuilders and local actors involved in conflict. While AI may generate useful outputs, such as early warning information or policy advice, AI-enhanced approaches can undermine the human capacity and interactions essential for sustainable conflict prevention, especially in the Global South. This is because developing and fine-tuning generative AI models and applications requires considerable resources and expertise, which gives companies and organisations based in developed economies an unequal advantage.
At the same time, AI developers often struggle to obtain sufficient training data from conflict contexts, yet this data is necessary to create AI applications that support local stakeholders in their efforts. For instance, AI-enhanced early warning systems for climate-related conflict in places such as Somalia often generate information of limited value for community-based preventive action because they focus on national-level predictions and contain little information about the context-specific causal pathways that drive local tensions and violence.
Thirdly, ensuring reliability and fairness: Peacebuilding actors, conflict parties and conflict stakeholders might find that AI-enhanced peacebuilding produces less reliable outputs if they are not designed and implemented with the active involvement of local stakeholders. In fact, conflict parties and stakeholders may oppose AI-enhanced processes because of potential biases in AI-generated results and the limited transparency of AI systems. Moreover, many AI applications have been trained on data that is culturally associated with their countries of origin, i.e., North America, Europe and China, while training data from conflict contexts is often difficult to come by.
Where AI systems lack alignment with the preferences and needs of conflict parties and stakeholders, and where third parties using them cannot adequately defend these methods against allegations of bias or discrimination, AI4Peace will likely fail the trustworthiness test, and, by extension, undermine trust in the peacebuilding effort as a whole. Fortunately, efforts to advance pluriversal approaches to digital peacebuilding, which reflect the knowledge systems, norms and practices of a diversity of actors, and to localise digital peacebuilding, are being explored, and which can also be extended to AI design and use.
Where AI systems lack alignment with the preferences and needs of conflict parties and stakeholders, and where third parties using them cannot adequately defend these methods against allegations of bias or discrimination, AI4Peace will likely fail the trustworthiness test, and, by extension, undermine trust in the peacebuilding effort as a whole
Tweet
Addressing these challenges – and ensuring that AI4Peace deserves the trust of those affected by violent conflict and working to build peace – requires a global effort. It should involve international and regional organisations, national governments, civil society and the private sector, including actors promoting African peace and security.
Andreas Hirblinger is a senior research fellow at the Norwegian Institute of International Affairs (NUPI).
