I want to build AI systems I can explain to my family

"The integrity of the upright guides them, but the crookedness of the treacherous destroys them." (Proverbs 11:3, ESV)

One of the tests I want to apply to my own AI work is simple: could I explain this system honestly to the people closest to me? Not with a technical diagram first, and not with a list of impressive features. In plain language. What does it do? What information does it use? What can go wrong? Who's responsible when it does? What should it never do?

If I can't answer those questions, I should be careful about asking anyone else to trust the system.

This isn't hypothetical for me right now

I'm in the middle of rebuilding my own personal site into an interactive resume, and I'm planning tailored versions of it for specific job applications. That means I'm building something whose entire job is to represent me honestly to a stranger who's deciding whether to trust me with something important. I've decided the site needs to be open about my actual weaknesses, on purpose, as a filter for finding a workplace that will actually support and build me up rather than one I had to oversell myself into.

That decision changes how I think about every AI system I build after it. If I'm not willing to let a system overstate my strengths to land a job, I have no business letting a system overstate its own confidence to land a customer's trust.

Complexity is not the same as wisdom

I enjoy the technical side of building systems, retrieval, databases, workflows, the practical details that make software useful. But technical sophistication can become a hiding place. A system can have an impressive architecture and still produce careless results. It can use a powerful model and still expose information it should have protected. Responsible AI needs both technical competence and moral clarity, and I don't get to substitute one for the other.

The questions a trustworthy system should answer

Those questions are understandable to Elizabeth, to a coworker, to a client, or to a hiring manager. They're also genuinely useful engineering requirements, which is exactly the point. A Christian standard for honesty and a good technical standard for accountability turn out to be the same standard.

The family test is really a neighbour test

I don't mean every family member needs to understand vector search or database schemas. I mean the system should be explainable at the level of responsibility. I'd rather be able to say, "This tool organizes repetitive information. It doesn't make the final decision. It doesn't receive private information it doesn't need. Important outputs get reviewed by a person. If it doesn't know something, it's supposed to say so," than say, "It usually gets it right."

Grounding Scripture: "Therefore, having put away falsehood, let each one of you speak the truth with his neighbor, for we are members one of another." (Ephesians 4:25, ESV)

Building for the person who isn't in the room

I've been building my own suite of personal tools for years, task managers, habit trackers, a Scripture memorization app, and I'm working toward selling some of them. Every one of those decisions gets easier when I ask who isn't sitting in front of the screen. The employee affected by a workflow. The client whose financial picture is inside an Ascent AI report. The family member whose time gets redirected because a tool made something feel more urgent than it was. Love of neighbour means those people belong in the design conversation even though they're never the one clicking the button.

An honest admission

I lead people because I want them to experience me as honest and trustworthy, someone who leads by example and has their best interest in mind. If I'm being fully straight, I still don't fully know how to trust and let go of control when I delegate, whether that's to a person or to a piece of software. That's worth naming here, because building AI I can explain to my family starts with the same discipline that letting a person on my team own something starts with: being honest about what I actually know, what I don't, and who's really accountable for the outcome.

The kind of intelligence worth building

The goal isn't a system that sounds like it knows everything. The goal is a system that's useful without becoming deceptive, efficient without becoming careless, and powerful without becoming unaccountable. I want to build things that can be explained in a way that respects the person listening, that protect people who aren't in the room, and that know when to stop. That's the kind of intelligence I can actually stand behind, and it's the kind of intelligence Faithful Intelligence is named after.

Frequently asked questions

What makes an AI system trustworthy?

A trustworthy AI system has a clear purpose, uses appropriate information, protects privacy, communicates uncertainty honestly, includes real human accountability, and has defined behavior for when it's wrong or doesn't know.

Why should AI be explainable?

AI should be explainable so people can understand its purpose, its data boundaries, its limitations, and who's accountable for it. Explainability lets users challenge errors and lets affected people understand how decisions actually get made.

What does Christian faith contribute to AI design?

Christian faith emphasizes human dignity, truth, responsibility, love of neighbour, and humility about the limits of anything we build.

Start here: the free F.A.I.T.H. AI Stewardship Audit is twenty questions about your week, not about technology. About five minutes, no cost, and your answers stay on your device.

I'm Dave, a developer, IT and facilities lead, and father of three, rebuilding my own portfolio into an honest interactive resume.