AI in global partnerships – how to best use AI systems

The use and development of AI systems often succeed or fail on basics:

• Poorly digitized local knowledge: local practices may not be digitized and are difficult to integrate into data-based processes.
• Administrative capacity: consistent regulation and enforcement.
• Risk of private tech providers exercising threats and pressure on developers and users: AI systems are often under the control of private companies.
• Communication infrastructure: availability and reliability of networks and sustained bandwidth, especially outside capitals and major urban centers.
• Electricity: stable power to run devices and servers.
• User devices: sufficient access to suitable laptops/phones.
• Computing access: affordable GPUs/cloud credits for training and deployment.

Good practices: plan budgets and timelines for electric power, devices, shared computing, and digitization; design for offline or low-bandwidth use from day one; train people to identify ethical risks and comply with applicable regulation.

AI projects inherit all the potential pitfalls of digital projects while also adding specific challenges. Without careful observation and testing, they are costly, fragile, and prone to mismatch with real users and contexts.

• High failure risk: poor user fit or solutions that only work in idealized settings.
• Low resilience: data, policies, or environmental change; performance degrades.
• Costly to plan: hidden work in data preparation, participation, and monitoring.
• Age poorly: technologies evolve; retraining and maintenance are unavoidable.

Here is a list of don’ts when using AI systems in research partnerships:

• Tech for tech’s sake - no clear objective or integration into a broader vision
• “Baby panda” projects - cute, but not scalable or sustainable
• One-way transfer - importing “Northern” solutions to different contexts, reinforcing epistemic domination and colonial legacies
• Context insensitivity - ignoring local context, undervaluing partners’ knowledge and expertise
• Deficit assumptions with respect to the partner - less resources ≠ less knowledge
• Savior mindset - partners are not “beneficiaries” to be saved, but equal co-creators
• Parachute collaborations - avoid short-term drop-ins; invest time in genuine, long-term relationships
• End-only sharing - don’t wait until the end to share results; capacity-building and knowledge exchange should happen throughout