Key Trends and Observations in GenAI and MERL for 2025

The State of GenAI and MERL: Reflections on 2025

As we dive into 2026, it’s vital to reflect on the trends and developments surrounding Generative AI (GenAI) and its intersection with monitoring, evaluation, research, and learning (MERL). Back in November 2024, I summarized notable trends from that year, and now I’m excited to share an updated perspective based on significant shifts and ongoing challenges.

The World is Still on Fire

2025 marked another year of turmoil, particularly in the realms of democracy, justice, and environmental sustainability. Big Tech’s influence proved detrimental in these areas, as partnerships with the government shifted focus from safeguarding AI to pursuing dominance. The establishment of the newly formed "Department of Government Efficiency" (DOGE), under tech mogul Elon Musk, resulted in the dismantling of critical organizations like USAID, disrupting vital programs and services globally. This led to instances like the Kenyan Government monetizing citizen health data, illustrating how U.S. policies often favor Big Tech at the cost of public welfare.

AI: Rapid Evolution and Adoption

Artificial intelligence is undoubtedly advancing, with organizations increasingly realizing its potential. For instance, several European entities have successfully implemented custom AI solutions that leverage internal data. Events like the Global Digital Health Forum showcased AI’s role in diagnostics and health education, reflecting a shift from theoretical application to practical usage. Despite this progress, challenges such as data hallucinations—where AI generates inaccurate information—persist.

At the MERL Tech Initiative (MTI), we categorize AI applications into three main buckets: backend functions, frontline worker support, and community-focused projects. As we’ve seen more real-world examples emerge, the utility of GenAI in various sectors is becoming clearer.

The Need for Internal Capacity and Data Hygiene

Organizations that successfully harness AI for MERL capabilities typically have internal data science expertise and prioritize data cleanliness. A recent report highlighted a range of approaches among foundations and NGOs towards AI adoption, developing four archetypes: the Curious, the Doers, the Dreamers, and the Skeptics. While many remain at the "Curious" stage, some "Doers" are beginning to systematize their efforts and share their progress, reflecting a growing understanding of AI’s potential.

Understanding Failures Driven by Perverse Incentives

A recent analysis by Priyanka Lakpattu Vasudevan revealed that organizational failures in AI adoption often stem from the push to adopt technology without proper governance and alignment with mission-driven goals. The hype surrounding AI has led to a speculative economy, concentrating wealth among tech elites while masking deeper societal fragilities. With AI companies significantly influencing stock market gains, a disproportionate economic reliance on a single sector is emerging.

The Ethical Debate: Should We Use AI?

Even amid technological advancements, ethical concerns about AI’s application persist. Questions surrounding its impact on ongoing issues—such as climate change—highlight the debate on whether AI should be utilized to tackle these challenges. Current projects like Humanity AI demonstrate an attempt to address these ethical quandaries while encouraging responsible AI development.

The Development of AI Policies and Guidance

The number of organizations developing AI policies is on the rise, although drafting effective policies is challenging due to the ever-evolving nature of AI. Strategies for embedding AI policy within broader governance frameworks are essential. Experts like Alberto Ortega Hinojosa emphasize that as AI influences all organizational facets, a holistic approach to governance is necessary.

The Need for Sector-Level Guidance in Evaluation

Insights from the recent American Evaluation Association Conference reveal a critical need for specific principles regarding AI use in evaluation contexts. Attendees expressed concerns about the potential promotion of harmful practices in AI implementation discussions. Addressing ethical considerations alongside AI’s advantages is crucial moving forward.

Exciting Advances in AI Evaluation Frameworks

In 2025, we saw commendable strides in evaluating AI’s effectiveness. The Agency Fund launched the AI Evaluation Playbook, which details a structured framework for assessing community-facing AI tools. Tools like ID Insight and Weval’s platform provide critical resources for conducting experiments and evaluations.

Building on Existing Knowledge

A successful application of AI in MERL requires returning to foundational ICT4D principles, recognizing the limitations imposed by infrastructure and accessibility issues in frontline worker initiatives. MTI aims to bridge this gap by focusing on inclusive and equitable designs in AI development.

Alternatives to Big AI Are Emerging

Interest in Small Language Models (SLMs) that function efficiently on personal devices is on the rise, offering advantages such as sustainability and improved privacy. However, the lack of user-friendly interfaces poses challenges to widespread adoption. Notable advancements like Microsoft’s Fara-7B signal progress, but the sector requires streamlined interaction methods for non-technical users.

Investment in Non-English Language Models Is Increasing

Signature investments in diverse language models, such as those initiated by The Gates Foundation, underscore the growing trend towards inclusivity in AI development. These initiatives aim to harness local contexts and languages for effective technological solutions.

Conclusion: A Promising but Uncertain Future

The developments in AI and the MERL landscape throughout 2025 provide both opportunities and significant challenges. Looking ahead, 2026 presents a chance for deeper reflection on responsible AI adoption and evaluation practices. As the global AI narrative continues to evolve, it will undoubtedly reshape our collective understanding and utility of technology in ways we are yet to fully grasp.

Stay Updated with MERL Tech Initiative

For more insights and guidance on navigating the complexities of AI in the MERL field, subscribe to the MERL Tech Initiative. Join us as we continue exploring the tools and practices that will shape our future!