The adoption of AI is giving organizations more ways to arrive at an answer and fewer reasons to pause before acting on it. A system can analyze a large volume of information, apply the criteria it was given, and produce a recommendation that appears well supported. The people approving that recommendation may have every reason to believe it is correct. They still have to decide whether acting on it would be just.
This has been on my mind since yesterday's readings for the 26th in Ordinary Time. In the Gospel, Jesus tells of two sons asked to work in their father’s vineyard. One refuses, then changes his mind and goes. The other agrees, but does nothing. Jesus asks which son did the father’s will, drawing attention to the difference between giving the expected answer and accepting the responsibility behind it. Ezekiel makes a related point stating that what matters is a person’s willingness to turn and do what is right and just.
These readings have a bearing on how we use AI. An organization can publish principles about fairness and human oversight while its systems quietly shape decisions about who receives attention, who is considered credible, or whose application deserves further review. Leaders may sincerely endorse those principles and still have little understanding of how the decisions are made in practice. The gap between what an organization says and what it does becomes harder to see when the work happens inside a system few people can explain.
AI can help people make better decisions. It can also make an existing judgment faster and harder to challenge. If a model relies on incomplete records, a person may be denied an opportunity because the system has an incomplete picture of them. If it is designed to favor outcomes that are easy to measure, it may treat an unusual but reasonable circumstance as an error. The resulting decision might be consistent with the model’s design and supported by the available data. For the person affected, that explanation does little good if there is no meaningful way to correct the record or ask someone to reconsider.
Being right, in this setting, often means that the system performed as intended and the decision followed the approved process. Being just requires us to examine the process itself. We need to understand what the system cannot see and whether the people affected have any real ability to be heard. Those questions become especially important when an AI system influences decisions about employment, access to services, insurance, or care, where an error can have consequences that are difficult to undo.
The reading from Philippians calls for humility and concern for the interests of others. Applied to AI, that means resisting the temptation to treat a technically sound result as the end of the conversation. A person who challenges an automated decision may know something the system was never given. Listening to them should be part of responsible decision making, even when the process appears to have worked exactly as designed.
This is a governance question as much as a technology question. Someone chose the purpose of the system and the conditions under which it would be used. People decide how much authority to give its output, when a human should intervene, and what happens when a decision causes harm. Those choices carry moral responsibility. Saying that the model produced the recommendation does not explain why an organization accepted it, particularly when the organization has reason to know the system’s limits.
What This Means for the Future
As AI moves from recommending actions to carrying them out, the opportunity to exercise judgment may come earlier in the process and be less visible to the people affected. A system might screen a candidate, adjust a customer’s terms, or route a request before any employee looks at the individual case. By the time someone asks whether the outcome was fair, several later decisions may already depend on it. Correcting the original error will then require more than changing a single answer.
Organizations will have to decide how much room they are willing to preserve for reconsideration. If they build processes around the assumption that the system is usually right, exceptions may come to be treated as delays rather than signs that something deserves attention. People could find themselves asked to prove that an automated judgment was wrong without being able to see the information or assumptions behind it. That would leave the organization with an efficient process and the person affected with little practical recourse.
The deeper risk is that responsibility becomes difficult to locate. A leader may rely on a vendor’s system, while the vendor points to the organization’s configuration and the employees involved say they followed the recommended action. Each account may be accurate. None answers who had the duty to notice that people were being treated unjustly and who had the authority to change course. That responsibility needs to be settled before a system is given greater authority, because it becomes much harder to recover after harm has occurred.
The readings offer an uncomfortable but useful test for leaders adopting AI. It is possible to say yes to fairness, accountability, and human dignity while building processes that leave little room for any of them. The work is to examine what those processes actually do, remain willing to change them, and take responsibility for the people who bear their consequences. A correct answer has value, but as more decisions are made at a distance from the people they affect, justice will depend on whether we remain willing to hear those people and act on what we learn.
