Whose Image?

Why AI Alignment Starts with the Human Heart

The alignment problem is the challenge of ensuring that AI systems respond reliably in line with human intentions. It’s about ensuring that an AI does what we actually want, not just what we ask it for.

The technical issues around alignment are clear and, in many ways, well understood. We can name them plainly: safety, interpretability, robustness, and guardrails. These are real problems, and they demand serious engineering effort, but they are not mysterious. They are solvable precisely because they live in code, systems, and processes we already know how to build and refine.

This work, however, has to be preceded by the harder question. Which values do they align to? Who defines good, or true, or worthwhile? AI systems reflect the incentives, assumptions and power structures of their creators. Do you even know who those people are? The more complex AI systems become, and the more dependent we become on them, the more serious this issue becomes.

The hardest alignment problem is not on the machine side - it’s on the human side. It’s us understanding, owning, and communicating our intentions with clarity, and in a way that drives the right action. AI doesn’t create moral fragmentation - it exposes what is already in the human heart and community. What we call misalignment is really often referring to core failures common among humans: conflicting moral frameworks, economic systems that reward profit over care, transactional views of humans.

This is why our focus on human flourishing matters so much.

When systems prioritize efficiency over relationship, when we ask them for prediction over presence, we see this fragmentation grow. When we structure our work to value control over commitment, it’s no wonder that we end up wondering just what it even means to be human.

AI will align to what we actually value, not what we claim to value. Where our treasure is, there our hearts - and our algorithms - will be also.

So what do we do about that? We come back to the imago Dei. If we don’t know whose image is in us, we cannot possibly hope to reflect it in the systems we build.

If we view ourselves as mere biological processors - data in, data out - then we shouldn’t be surprised when our AI treats us as nothing more than data points to be optimized. When we lose sight of the inherent dignity and mystery of the human person, we outsource our discernment to the loudest or fastest algorithm available. It’s like aligning a high-speed technology to a low-resolution understanding of ourselves.

At Gloo, we acknowledge that this has implications for everything from our team culture to our product design choices to what our criteria are for the benchmarks. Solving the actual alignment requires a return to the virtues that make life worth living. It looks like:

• Valuing Presence over Productivity: Choosing to use AI to clear the path for deeper human connection, rather than just doing more “stuff” faster.

• Prioritizing Wisdom over Information: Recognizing that just because a machine can predict the next word doesn’t mean it understands the weight of the message, and that only happens in community.

• Building for the Vulnerable: Aligning systems not just to the needs of the shareholders, but to serving the “least of these” in our global community.

If we want AI to be kind, we must cultivate kindness in our boardrooms and our living rooms.

If we want it to be just, we have to repent of our own injustices.

If we want it to serve others, we have to start by putting ourselves in their shoes.

If we want it to drive human connection, we have to choose human connection ourselves.

The machines will follow where we lead.