Can we trust AI? Dr. Richard Zhang, a research scientist at Google DeepMind, suggests asking a better question: which direction is that trust heading?
Richard works on some of the most technical parts of building large language models, from prompt optimization to alignment. He's also a committed Christian who helps bring faith into conversations about AI safety and responsible AI inside one of the world's largest technology companies.
We sat down with Richard at Missional AI 2025 in Dallas to talk about trust, what a distinctly Christian approach to AI safety might look like, and what it's like to follow Jesus while building frontier AI.
Richard's work focuses on getting better answers from AI models and aligning them with human values, including values that differ across people and cultures.
He started in hyperparameter optimization, which he describes simply: he's the person tuning the knobs in the background, trying a setting, keeping it if results improve and reverting if they don't. Today he works on prompt optimization, because prompts are "the hyperparameters of the future." Sometimes you have to ask a model the same thing several different ways before it gives the answer you need, and his job is figuring out how to ask the right questions.
He also works on alignment: how to help models reflect human preferences and beliefs, especially where those beliefs are pluralistic.
Instead of asking whether you can trust AI right now, Richard suggests asking where that trust is heading, the same way God looks at us.
Most people treat trust as static: can I trust this answer today? Richard prefers to look at its direction. "Does God trust us? I don't know if God should trust me, honestly," he says. "But thankfully, God is sanctifying me, and that is a direction of trust that is positive."
He sees the same trajectory in AI. A year earlier, people regularly messaged him because Gemini had invented a Bible verse. Humans misquote Scripture too, he admits, but it showed room to improve, and those errors are becoming less common. Today, he says, he tends to trust AI more than humans, which raises ethical questions of its own.
AI agents working behind the scenes can feel unsettling, but Richard sees them like any complex technology we already rely on, and notes that today's models increasingly show their reasoning in plain language.
As agents begin making decisions for us out of sight, complicated pipelines can create fear. But Richard points out that we trust plenty of systems we don't understand, from a microphone to the many processes behind a flight. That trust rests on the people and companies who built them, which is why he wants more open conversation between AI builders and users.
Large language models also bring something new. Older systems ran on numbers no ordinary person could read. LLMs translate their processing into text, so with reasoning models such as OpenAI's o1, Gemini 2.5 Pro and Claude, you can follow how the model thinks through a problem. That makes the system more interpretable, and builds trust.
Richard believes Christians can contribute something distinctive to AI safety: values arranged in a divine order, not a flat list.
Google, other companies and governments have all published AI principles, but they often read as simple lists. Richard sees Christian values differently. There are core values God cares about most, from which secondary and tertiary values flow, and Jesus's message makes clear that missing the core makes everything else pointless.
Even the Ten Commandments, which look like a list, have an order: the first commandments concern God, the rest concern people. Richard sees Christian ethics as more like a tree than a list. Bringing that divine ordering into safety conversations, grounded in how God created the world, is something he believes Christians can step into.
People often describe AI as "magic," but there is real logic under the hood. Richard sees apparent magic as an invitation to understand more. His faith teaches him there will always be mystery in knowing God, yet that mystery never means giving up on understanding. We seek God little by little, and he believes we can approach AI with the same posture of seeking.
A small prayer meeting for AI researchers has grown into a wider community that sustains Richard and helps him bring faith into his work.
Richard credits a more senior Google researcher, Cheryl Chen, who arrived at the company and asked God what He wanted her to do there. Among other initiatives, she started prayer meetings for AI researchers around 2020, where they prayed for their work and their field. That gathering grew into a community called Christians and AI, bringing together believers from nonprofits, startups, industry and academia to discuss the fast-moving field, pray and seek the Lord together.
That fellowship, he says, has sustained him and given him the confidence to bring his faith into work conversations. He now works with Google's responsible AI teams, who are discovering that most people's identities and cultures are rooted in faith. With 80 to 85% of the world's population identifying with a religion, most of the billions of people Google serves are people of faith.
Google doesn't have a formal benchmark for religious understanding, Richard notes, speaking for himself. But he sees Gemini and other models steadily improving, and hopes progress on clearer tasks will carry over to more ambiguous ones, including questions of faith.
Richard also spoke on the main stage at Missional AI 2025 with his keynote, "AI Reasoning & Agentic Flows: Redemptive Use Cases for LLMs." You can find it in the Missional AI Video Library under Dallas 2025 Keynotes.
Where this conversation explores trust and readable reasoning, the keynote looks at how reasoning models and agentic workflows can serve redemptive ends. For more on faith inside big tech, the library also includes "Faith in Big Tech: Lessons from Google & DeepMind Leaders" and Barbara Villasenor's "How a Small Prayer Group Scaled to 11,000 in Big Tech."
Listen to the full episode with Dr. Richard Zhang on the Missional AI Podcast, and subscribe so you don't miss future conversations.
Join us at Missional AI London on July 17, a one-day gathering for leaders across the UK and Europe shaping the future of ministry through technology. Find out more and register at missional.ai.
This episode was made possible in partnership with Gloo, the technology platform connecting the faith ecosystem.