Ethical Tools for the Expert in the Loop
AI is commoditizing expertise. Without ethical standards, market forces invite conflicts of interest and bias.
Sep 29, 2026 · @Jeff Kluge, EVP Strategy and Corporate Development
The Race to the Lowest Common Denominator
AI is being built to answer everyone, instantly and cheaply. The economics reward scale, speed, and answers that are good enough. But when the objective is to produce more answers at lower cost, something important can get lost: the distinction between an answer that sounds right and a judgment that deserves to be trusted.
There is another race worth running. Not to the bottom, but toward the expert.
In consequential fields, a 90% right answer is not 90% useful. It is a liability waiting to happen. The remaining 10% is where the forgery, the misdiagnosis, and the bad claim can live.
General-purpose AI has three persistent problems when applied to work that demands expertise.
First, it can sound equally confident when it is right and when it is wrong. Fluency is not accuracy, and the person reading the answer may have no reliable way to distinguish between them.
Second, no one necessarily stands behind the answer. There may be no named professional, no credentials, no signature, and no clear responsibility for the consequences. The risk ultimately lands on the person who acts on it.
Third, AI can flatten the distinction between expertise and the appearance of expertise. A novice can produce work that sounds authoritative, while a genuine expert can appear increasingly replaceable. At the same time, systems can draw on experts' published work without adequately recognizing their contributions or distinguishing sound scholarship from error.
The result is not simply the democratization of knowledge. It can be the dilution of accountability.
The question is not whether AI can produce useful answers. It can. The question is what happens when those answers influence decisions involving money, reputations, public trust, or human welfare.
The Expert in the Loop
An expert in the loop is not someone who checks a machine's work as a final formality. It is a named, qualified, accountable person whose judgment the tool exists to support.
The distinction matters. A tool should make an expert faster, more thorough, and better documented. It should help gather evidence, compare information, identify anomalies, and expose questions that warrant further investigation. It should not quietly substitute a generated conclusion for professional judgment.
The person with the relevant expertise must retain the authority to interpret the evidence, reach a conclusion, explain the reasoning, and stand behind the result.
That principle can be expressed through five commitments.
1. The expert decides.
The tool gathers, compares, and flags. The qualified person interprets the evidence and makes the judgment. Where a professional determination is required, the person responsible must review and sign off on it.
2. Evidence is visible.
A conclusion must show its basis. Supporting evidence, sources, and reasoning should be available for examination and challenge. An answer that cannot be examined is difficult to verify, regardless of how confidently it is delivered.
3. Every judgment is attributable.
The record should identify who made the judgment, when it was made, what evidence supported it, and what process informed it. The result should be documented in a form that can be reviewed and audited later.
4. Limits are disclosed.
A responsible system must identify what it cannot establish, where information is missing, and when the evidence is insufficient to support a confident conclusion. Uncertainty should not disappear simply because a system is designed to produce an answer.
5. Expertise is protected.
Experts should be credited for their contributions, and their work should not simply be appropriated to train systems designed to replace them. The economic incentives surrounding AI should support the continued creation of knowledge, rigorous research, and professional judgment rather than undermine them.
These are not merely principles for responsible technology development. They describe the conditions under which AI-assisted work can earn trust.
This is not ethics as a compliance layer bolted on at the end. It is ethics as architecture.
The Economics of Accountability
When answers become free, trust becomes scarce. Scarce things command a price.
A collector may be able to obtain an instant opinion about a painting, but what is that opinion worth when the work is sold, insured, or contested? The collector needs more than a conclusion. They need to know what supports it and who will stand behind it when the stakes become real.
An insurer needs evidence it can underwrite. An institution needs a defensible record. A court needs qualified testimony, not simply an answer generated by a chatbot.
The cost of an unreliable answer does not disappear because the answer is inexpensive. It moves downstream, to the people and institutions that must act on it.
This is where the economics of expertise becomes inseparable from accountability. The value of an expert's judgment lies not only in the conclusion, but in the evidence supporting it, the experience used to interpret it, and the willingness to stand behind it.
AI can change the economics of expert work. It can reduce the time required to gather and compare evidence, make research more consistent, organize documentation, and extend the reach of scarce expertise. Tasks that once consumed hours of professional time may become faster and more systematic.
Those gains matter. They can make qualified expertise more accessible and allow experts to spend more time on the work that genuinely requires their judgment.
But if the process removes the accountable expert, it may also remove the very thing that makes the result trustworthy.
The opportunity is not simply to make expertise cheaper. It is to make qualified expertise more productive and more widely available while preserving the accountability on which its value depends.
Commodity AI can appear inexpensive because the cost of its mistakes is often borne elsewhere. An accountable expert, by contrast, must consider the consequences of an error, document the reasoning, and be prepared to defend the conclusion. That responsibility is not friction to be eliminated. It is part of the service being provided.
The strongest economic case for AI, therefore, is not that it makes experts obsolete. It is that experts can become more valuable because of AI: able to investigate more thoroughly, respond more quickly, document their work more rigorously, and support decisions with greater confidence.
What This Means for Art
Art authentication makes the stakes tangible.
One incorrect judgment can move millions of dollars and rewrite an artist's legacy. A false attribution can distort a market. A missed forgery can damage a collector, an institution, or an auction house. A poorly supported conclusion can be repeated until it becomes accepted as fact.
Imagine two approaches to the same problem.
In the first, an application examines a photograph and returns a verdict. The answer is fast and inexpensive. But a photograph alone may not provide enough evidence to establish authenticity, and the user may not know what the system could not see, what evidence it considered, or who is accountable if it is wrong.
In the second, a conservation scientist examines the work. AI helps compare available material evidence against relevant reference data, identifies anomalies that warrant further investigation, and assembles research and documentation. The scientist interprets the findings, directs further research, weighs the evidence, and makes the judgment.
The distinction is not that one approach uses AI and the other does not. Both may use it. The distinction is whether the technology produces an unaccountable verdict or supports a defensible expert determination.
In the second approach, the findings, supporting evidence, and authorship of the judgment can be recorded in a provenance record that travels with the artwork throughout its life.
That creates value beyond the initial assessment.
The collector receives a better-supported conclusion. The insurer has evidence it can evaluate and potentially underwrite. Dealers and auction houses can strengthen their due diligence and protect their reputations. The artist's legacy is better served by a process that takes the consequences of error seriously. And the expert's judgment becomes more useful because it is supported by a stronger, more transparent research process.
The economic and ethical outcomes reinforce one another. Better evidence supports better decisions. Clear attribution makes responsibility visible. Documentation allows a conclusion to be examined rather than accepted on authority alone.
None of this eliminates uncertainty. Nor should it pretend that AI can resolve every question of authenticity. Art expertise often involves incomplete records, conflicting evidence, changing scholarship, and judgments that remain open to revision. A responsible process makes those limits visible instead of concealing them behind the appearance of certainty.
The objective is not to manufacture certainty. It is to improve the quality, transparency, and accountability of the judgment.
Designing for the Work That Matters
The principles extend far beyond art. They apply wherever AI influences consequential decisions: medicine, law, financial services, scientific research, insurance, and public institutions.
In each field, the central question is the same. Does AI strengthen the process by which a qualified person reaches and supports a judgment, or does it create the appearance of expertise while obscuring responsibility?
Builders have a choice. They can optimize exclusively for speed, scale, and cost, or they can design systems that preserve evidence, disclose limitations, recognize expertise, and make responsibility traceable.
Buyers have a choice, too. They can ask only how much a system costs and how quickly it produces an answer, or they can ask who stands behind that answer, what supports it, and what happens when it is wrong.
Experts also have a choice. Treating AI only as a threat risks surrendering the opportunity to shape how it is used. But adopting AI without demanding standards worthy of professional judgment risks something more consequential: allowing the value of expertise to be separated from the accountability that gives it meaning.
The long-term economic value of AI will depend not only on what it can automate, but on whether people can trust the work it helps produce. That trust cannot be generated by fluency alone. It must be earned through evidence, transparency, responsibility, and the continued contribution of qualified people.
AI will be used for work that matters. That is settled. The open question is whether it will be used by the people qualified to answer for it.
The race to the lowest common denominator has plenty of runners. We are changing the race. If you are an expert whose judgment matters, or someone who depends on one, run it with us.