Warranted reliance on algorithms: make performance and consequences visible
Family Office Private Wealth Journal · Editorial research desk
Family Office Private Wealth Journal · 2026-10-29
CC BY 4.0 · https://creativecommons.org/licenses/by/4.0/
Source: https://journal.research.university/journals/family-office-private-wealth/articles/expansion-10-1371-journal-pone-0278751-solutions

Abstract

AI-assisted editorial analysis. The source reports an experiment with 143 students at a German university. Participants chose between a human expert and an algorithm in one of six framed service scenarios. The algorithm was assigned a seventy percent success probability and the human expert sixty percent, with the same monetary incentive across scenarios. Participants were less likely to select the algorithm in scenarios perceived as more serious. This provides a useful question for family-office technology governance: does the perceived gravity of a task alter willingness to use a tool even when its performance information is held constant?

A controlled choice, not a certification of artificial intelligence

The source reports an experiment with 143 students at a German university. Participants chose between a human expert and an algorithm in one of six framed service scenarios. The algorithm was assigned a seventy percent success probability and the human expert sixty percent, with the same monetary incentive across scenarios. Participants were less likely to select the algorithm in scenarios perceived as more serious. This provides a useful question for family-office technology governance: does the perceived gravity of a task alter willingness to use a tool even when its performance information is held constant?

The experiment does not establish that current investment algorithms outperform qualified professionals. The superiority of the algorithm was stipulated through a lottery rather than established through field validation. The scenarios also differed in context, moral associations and other features, so gravity was not independently varied within one task. The authors acknowledge the small sample and transfer limits. A responsible application therefore examines the reasons for reliance or rejection; it does not declare every objection to automation irrational. Real decisions include accountability, rights, unequal error costs and uncertainty about whether reported performance applies.

Build a comparison before asking people to trust

The technology owner should define the actual task in operational terms: extracting a covenant, flagging an unusual transfer or summarizing a document are different activities from approving a payment or selecting an investment. Identify the intended user, the affected person, the action triggered and the consequence of an error. Preserve this description as the scope of the evaluation. A tool that performs well on a narrow extraction task should not inherit authority over a broader decision simply because its interface appears competent. The study’s fixed success probabilities make task definition especially important when translating its lesson.

The evaluation lead should compare the tool with the relevant human or existing process on the same representative cases. Keep test cases separate from material used to configure the system. Record omitted cases, uncertain labels, population differences and the costs of both false positives and false negatives. A single accuracy percentage is insufficient if the rare failure is an unauthorized transfer or a missed covenant breach. Where an adequate comparison is unavailable, label performance unknown. Do not use the paper’s seventy-versus-sixty example as a substitute for evidence about a vendor, model version or deployment setting.

Investigate resistance without concealing consequences

The operations lead should invite users to explain objections before training begins. Distinguish a mistaken belief about measured performance from a valid concern about confidentiality, recourse, authority, workload or an unrepresented case. Record each concern with an owner and a route to resolution. If the objection identifies a real coverage gap, update the evaluation rather than treating the user as an obstacle to adoption. If it rests on a misunderstanding, explain the evidence and its limits in accessible terms. The objective is justified reliance, not maximum acceptance of whatever system has already been purchased.

The source discusses framing as a possible way to influence adoption. This companion does not recommend hiding important risks to make an algorithm seem less threatening. Communicate consequences and limitations clearly, especially where affected people cannot reverse a decision. Show examples of correct outputs and failures, and explain when human escalation remains necessary. An office can make a description precise without making it deceptively reassuring. A user’s willingness to use a tool after a polished demonstration is not proof that the tool is safe, nor is discomfort proof that the tool lacks value.

Start with bounded authority and observable decisions

The accountable business owner should define what the system may recommend and what requires an authorized human decision. In an initial pilot, use shadow mode where feasible so outputs can be compared without automatically changing client assets or access. Record tool version, input scope, output, human response and the evidence behind any override. Avoid a rule that humans must always agree with the system: that would prevent detection of valid objections. Also avoid treating a human sign-off as meaningful oversight if the reviewer lacks time, information or authority to challenge the output.

Measure performance by case type, error severity, unresolved exceptions, time to correction and the quality of the final decision record. Track both unsupported rejection of useful advice and uncritical acceptance of wrong advice. Adoption rate alone cannot distinguish those failures. Pause the pilot when a material error falls outside the approved risk tolerance, when confidentiality controls fail or when users cannot explain the basis for an irreversible action. Define the evidence required to reopen, such as corrected configuration, independently checked test cases or a narrower authorized scope. These are proposed governance controls, not interventions tested in the paper.

Keep responsibility attached to the system’s use

The committee should review the pilot with the people who bear its consequences, not solely with the vendor or the team that selected the product. Compare the originally approved task with actual use and identify any expansion in authority. Retain the performance comparison, concern register, training examples, override log and stop decisions. Revalidate when the model, workflow, data source or relevant population changes. A favorable comparison from an earlier version should not silently carry forward to a changed product. If reliable comparison cannot be maintained, reduce reliance rather than substituting reputation for current evidence.

This original AI-assisted analysis treats the study as a bounded behavioral experiment. It does not claim that aversion explains every failure to adopt technology or that algorithms should control high-stakes decisions whenever an average success rate looks higher. Source authors retain their research credit and do not endorse this workflow; no new human review is asserted. The practical purpose is to replace arguments about trust with a documented comparison of task-specific performance, consequences, uncertainty and authority, while preserving the right to question automation for reasons the experiment’s simplified choice did not represent.

Source, attribution and limits

[1] Ibrahim Filiz; Jan René Judek; Marco Lorenz; Markus Spiwoks. The extent of algorithm aversion in decision-making situations with varying gravity. PLOS ONE (2023-02-21). https://doi.org/10.1371/journal.pone.0278751. CC BY 4.0.

Student sample of 143; algorithm superiority stipulated by lottery, not field evidence. Context and perceived gravity vary together; their independent causal effects are not isolated. Real-world error costs, rights and uncertainty are broader than the incentive task.

Prepared by the Family Office Private Wealth Journal editorial research desk with AI assistance at the direction of Dr. Freedom Cheteni. Original research authors retain full credit and are not authors or endorsers of this companion. This proposed process has not been evaluated in a new empirical study; no independent peer-review acceptance is claimed. This is educational analysis, not individualized investment, legal or tax advice.

Rights: © 2026 The Research University.
© 2026 The Research University. This original editorial text is licensed Creative Commons Attribution 4.0 International. Attribute the title, editorial research desk and publisher, link to the license and indicate changes. Underlying research, talks and recordings retain their separate attribution and rights.
Edition changes: Original AI-assisted editorial synthesis. Research findings, transcript themes and proposed practices remain distinct. No new participant data or independent peer-review acceptance is claimed.