Our evolving AI journey: Laterite's exploration into generative AI for research

In 2021, long before generative AI became ubiquitous, Laterite made a strategic decision to invest early in its potential. As the field shifted from GPT-2 to GPT-3, we saw enormous promise and became one of the first adopters of OpenAI's API, driven by the belief that generative AI could significantly accelerate the demanding tasks involved in data, research, and analytics.

There wasn't a clear roadmap at first. Early on, we tackled foundational challenges inherent in applying AI to research - such as dealing with model hallucinations and reliably extracting information from diverse documents. We built experimental tools aimed at detecting bias in survey questions and assisting with SurveyCTO workflows, often learning through close collaboration with experts like Christopher Roberts at Higherbar.ai / Dobility. Reflecting back, those initial tools appear modest compared to the powerful capabilities of today's AI models, but they represented critical early steps in our AI journey.

Initially, we envisioned developing a subscription service for researchers, crafting our first web presence. However, we quickly realized this ambitious goal required additional resources beyond our scope as a research firm. We strategically pivoted, narrowing our focus to areas where we could achieve more immediate impact: enhancing internal efficiencies and developing intelligent knowledge hub solutions for external clients. This refined approach enabled meaningful collaborations, including our partnership with the Tony Blair Institute (TBI), exploring sector-wide knowledge solutions in education and other critical areas.

Today, these focused efforts have culminated in an internal AI assistant seamlessly integrated within Slack – called LateriteAI – which supports our teams by helping to:

• Automate survey code generation.
• Streamline quality checks for qualitative transcripts.
• Identify themes in qualitative transcripts (in software like Nvivo or MAXQDA)
• Accelerate reviews of contracts and proposal documents.
• Enhance support for statistical programming (such as Stata).
• Provide rapid access to internal knowledge on policies, projects, and team expertise.
• Improve efficiency on operational tasks like timesheets and finances.
• Support complex research processes, including qualitative data analysis and literature reviews.

Our exploration continues. Currently, we're testing AI agents designed for automated quantitative analysis - transforming datasets directly into analytical code and initial findings. While promising, this work remains cutting-edge and firmly within our development and validation pipeline.

Throughout this journey, we remain committed to responsible innovation. Building effective AI tools for research involves continuous testing, refinement, and essential human oversight. This careful balance helps us work towards solutions are not only efficient but trustworthy and ethically sound.

Sharing our journey & seeking collaboration

We invite you to follow our progress and learn more on our newly relaunched website: laterite.ai. The site is a hub to share insights from our AI journey and connect with potential partners.

While direct access to some of our tools might become available in the future, our current focus remains collaborative. Through our AI-as-a-service model, we actively help organizations accelerate their research processes. If you're interested in leveraging AI for specific research tasks, please reach out—we'd be delighted to discuss how our tools and expertise can support your work today.

We continue to be amazed by the rapid advancements and opportunities within generative AI. This journey is one of ongoing learning, adaptation, and a deepening understanding of both the extraordinary possibilities and significant responsibilities that come with using AI to advance social impact research.