On the Party Present in the Decision, Yet Absent from the Courtroom

A doctor diagnosed the condition, the patient followed the diagnosis, and harm occurred. Naturally, we would examine the responsibility of the doctor and the healthcare provider.

The parties are known, the relationship is clear, the professional license exists, and the medical record can be reviewed to determine who made the decision and whether negligence or professional misconduct occurred.

It is certainly a complex matter, but at least it begins with a question we know how to investigate:

Who actually made the decision?

But what if the diagnosis was not made by a doctor?

Imagine someone accessing a hospital application and using an AI-powered diagnostic service. The system asks about their symptoms, reviews their medical history, analyzes their test results, and provides a recommendation they trust and act upon.

Later, the recommendation turns out to be wrong. Treatment is delayed, and harm occurs.

The same question arises:

Who made the mistake?

This time, however, the answer opens one door after another.

Before we continue, let us clarify an important point.

At ODEL, we believe that artificial intelligence will significantly transform the quality and efficiency of services. It will help organizations better understand their customers, accelerate operations, analyze vast amounts of data, and deliver services that are more personalized and responsive.

This article, however, is not intended to put artificial intelligence itself on trial.

Rather, it explores the types of disputes that may emerge when AI becomes an influential participant in service delivery and decision-making within commercial relationships.

Let’s begin.

The Dispute Where We Know the Parties

In the traditional model, there is a relatively direct relationship between the beneficiary and the service provider.

A doctor provides a consultation. A customer service representative gives instructions. A consultant makes a recommendation. An employee reviews an application and makes a decision.

When something goes wrong, we can look at the individual who performed the action, the organization they work for, the procedures they followed, and the professional regulations and controls governing their work.

Commercial and professional disputes are certainly not always straightforward. Responsibility may be shared, and multiple factors may contribute to the outcome.

Yet the parties involved are generally known, and the starting point is clear:

Who provided the service? Who made the decision?

Artificial intelligence, on the other hand, does not hold a professional license. It does not sign its recommendations. It cannot appear before a court. And, at least for now, it has no independent financial liability against which a claim can be made.

And yet, AI may be present at the most important moment in the entire journey:

The moment the decision is made.

It is the party present in the act, yet absent from the dispute.

Now, Let’s Introduce AI into the Relationship

Let us return to our previous example.

Suppose the hospital offered the diagnostic service under its own name but did not develop the solution itself.

The hospital purchased the solution from a technology company that built the application and integrated it with the hospital’s systems.

That company, in turn, used an AI model developed by another provider.

The model was then enhanced using data supplied by a specialized organization, connected to external medical knowledge sources, and deployed on cloud infrastructure operated by yet another provider.

We now have:

• The hospital that provided the service.
• The company that designed the application and user experience.
• The provider of the foundation model.
• The organization that customized and trained the model.
• The parties that supplied the training data and knowledge sources.
• The infrastructure and computing provider.
• And perhaps a specialized team that approved the service for deployment and determined the level of human intervention required.

All of them contributed, in one way or another, to producing the outcome.

But that does not necessarily mean they are all responsible for it.

The error may originate in the model itself, the data on which it was trained, the way it was customized, the information retrieved while generating the answer, the integration between the application and internal systems, or the instructions provided by the organization.

Perhaps the recommendation itself was probabilistic, but the application presented it to the beneficiary as though it were a definitive decision requiring no human review.

Every individual component might even operate exactly as designed, while the error emerges only from the way those components interact.

At that point, we are no longer searching for a single mistake.

We are searching for the mistake within an entire technological and commercial chain.

And It’s Not Just Healthcare

Suppose you purchased an electronic device and, after some time, it developed a minor issue.

You open the company’s application and contact customer support.

But the person responding is not a customer service employee.

It is an AI agent.

The agent asks you about the problem, reviews your device model and version, and provides a series of troubleshooting instructions.

You follow those instructions exactly as presented.

The device stops working completely.

Perhaps you even lose your warranty because of one of the actions the AI instructed you to perform.

Who is responsible?

From your perspective, the answer seems obvious.

The company sold you the device and provided the support service under its own name.

But the company may argue that the recommendation did not come from one of its employees. It was generated by an AI system supplied by an external technology provider.

The technology provider may respond that the agent relied on a model developed by another company and that its answer was based on the knowledge base supplied by the device manufacturer.

The model provider may then argue that the model had no specific knowledge of your device and simply generated its answer based on the information and instructions made available to it.

And so it continues…

An endless cycle of passing the blame.

Eventually, we may discover that the incorrect information originated in an outdated maintenance manual.

Or perhaps the AI agent confused two different versions of the device.

Or the application itself sent incorrect information about the product you were using.

The bottom line?

The device is broken, but responsibility has been distributed across the manufacturer, the customer service provider, the AI agent developer, the model provider, and the knowledge source behind the answer.

A simple conversation with customer support has suddenly become a technological and commercial investigation into who told you to press the wrong button.

In the first example, the result was harm to a person’s health.

In the second, it was a broken device.

The scale of the harm differs, but the structure of the dispute remains the same:

The customer interacted with an organization they knew, while the decision that caused the harm was produced within a chain of models, data, and suppliers they knew nothing about.

Who Actually Provided the Service?

From the customer’s perspective, the answer is simple:

The organization whose name was attached to the service.

The beneficiary accessed the organization’s application, trusted its name, and interacted with the AI agent as its representative, even if the agent was not technically an employee in the traditional sense.

The customer did not select the AI model.

They did not inspect its training data.

They did not approve the suppliers involved in developing and operating it.

Yet the service provider may argue that the defect lies in the model supplied by the technology company.

The technology company may say it integrated the model according to the specifications and instructions provided by the organization.

The model provider may explain that its model is general-purpose and was never designed for the specific use case in which it was deployed.

The data or knowledge provider may insist that it supplied the information as-is and that responsibility for validating its suitability belongs to whoever used it.

The computing provider will most likely say that its infrastructure executed the required operations but had no control over the decisions generated by the model.

Once again:

An endless cycle of passing the blame.

The injured customer can therefore go from being someone seeking compensation to becoming a technical investigator expected to understand a chain of relationships they were never a party to, whose contractual arrangements they have never seen, and whose participants they may not even have known existed.

AI Is Not a Doctor or a Customer Service Employee

Artificial intelligence is not a doctor.

Yet it may provide a medical recommendation.

It is not a customer service representative.

Yet it may speak on behalf of a company and tell customers what they should do.

It is not a financial advisor.

Yet it may recommend an investment decision.

It is not a credit officer.

Yet it may analyze a financing application and recommend approving or rejecting it.

In other words, AI can now perform roles traditionally carried out by humans within commercial relationships and influence customer decisions without being a natural or legal person that can be held directly accountable.

This does not mean that artificial intelligence has become an independent legal entity.

But it certainly means that it is no longer merely a peripheral piece of software within the relationship.

It has become an active layer in producing decisions — but one that cannot be summoned to court and asked: “Why did you do that?”

The Problem Is Not Simply: Who Made the Mistake?

In traditional disputes, we examine the act, the actor, the harm, and the causal relationship between them.

With AI systems, proving that causal relationship may itself become the greatest challenge.

Which version of the model generated the recommendation?

What data was available to it at that moment?

What instructions guided its behavior?

What was the source of the information it relied upon?

Was the model changed after the incident?

Can the same result be reproduced?

And does any party maintain a complete record of what actually happened?

A doctor, customer service employee, or consultant can be asked why they made a particular recommendation, even if their explanation is later disputed.

A model, however, may generate an answer without providing a clear and accurate explanation of how it arrived at that result.

Even when records exist, they may be distributed across several organizations.

Some may be located in different countries or even different continents.

Others may be protected by intellectual property rights, trade secrets, or other restrictions.

The dispute therefore becomes more than a disagreement over responsibility.

Before responsibility can even be determined, there may first be a dispute over accessing and understanding the evidence.

Contracts Allocate Loss — But They Do Not Answer the Customer

Companies will naturally attempt to manage these risks through contracts.

A model or service provider may establish restrictions on how its technology can be used.

A technology company may include clauses limiting its liability.

The service provider may demand guarantees regarding service levels and performance.

Responsibilities for testing, monitoring, updates, and compensation may be distributed among the parties.

All of this is important for regulating the relationship between those organizations.

But it is not enough to answer the person who suffered the harm.

The customer did not review the contracts signed between the service provider and its technology suppliers.

They did not participate in selecting the model or its training data.

Their right to compensation should not become a journey through exclusion clauses and limitations of liability.

A contract may determine which company ultimately bears the loss, but it should not force the injured party to search for the weakest link in the AI chain.

Regulation Is Advancing — and the Judiciary Faces a New Kind of Dispute

As organizations increasingly adopt artificial intelligence, regulators have begun developing sector-specific frameworks governing its use based on the nature of each industry, the types of decisions AI participates in, and the potential impact on customers and beneficiaries.

These regulations will provide an important foundation for courts when determining the obligations of each party, the controls governing the use of AI systems, and the requirements for testing, oversight, documentation, and risk management.

But once an actual dispute arises, the issue shifts from regulating AI use in general to understanding a specific incident in all its technical detail.

The judiciary will need to determine which model was in use when the harm occurred, what data and information it relied upon, what instructions shaped its behavior, what updates had been made to it, and how much control each party had over the resulting decision.

It will also need to distinguish between the responsibilities of:

• The organization that provided the service.
• The company that developed the application.
• The provider of the AI model.
• The party that trained or customized the model.
• The providers of data and knowledge sources.
• The provider of the technical infrastructure on which the system operated.

Regulation can help establish what each party should have done.

The judicial system, however, must also develop the ability to understand what actually happened, identify where the failure occurred, and establish the relationship between that failure and the resulting harm.

A judge does not need to become an AI developer.

But judges will need access to the capabilities and expertise required to understand technical evidence, interpret model logs, distinguish between the roles of different parties, and identify who had the actual ability to detect or prevent the error.

This brings us to one of the most important requirements of the coming era:

The evolution of AI must not be limited to the regulators governing it or the organizations using it. It must also extend to the way the judicial system understands the disputes it creates.

The goal is not to invent a new defendant.

It is to develop the judiciary’s ability to understand who shaped the decision, who could have prevented the harm, and who exercised control at each stage.

Which brings us to the most important question in this article.

How Can the Judicial System Be Enabled in the Age of AI?

An advanced judicial system dealing with these disputes cannot rely solely on reading contracts and examining final outcomes.

It will need to understand the entire lifecycle of the system.

This requires several key enablers.

Clear Classification of Roles

A distinction must be made between the model developer, the system provider, the organization that customized it, the organization that deployed it, and the organization that used it to provide the service.

The existence of multiple parties does not mean that their responsibilities are equal.

Responsibility Based on Control and the Ability to Prevent Harm

The party closest to the customer is not necessarily responsible for every technical failure.

Likewise, a party further removed from the customer should not automatically escape responsibility simply because it had no direct interaction with them.

The focus should be on who determined the use case, who tested the system, who monitored its performance, and who had the ability to detect or stop the failure.

Records That Cannot Disappear

A dispute cannot be properly resolved if we do not know the model version, its inputs and outputs, its knowledge sources, the instructions that guided it, and the updates made to it.

Traceability and record retention must therefore become requirements for evidence, not optional technical features.

Specialized Technical Expertise for the Judiciary

Cases may require experts in AI models, algorithms, data, cloud infrastructure, cybersecurity, and user experience, alongside specialists in the sector in which the dispute occurred.

The way a recommendation was presented to the customer may contribute to the harm just as much as the accuracy of the model itself.

Rules for Accessing Evidence

Customers cannot reasonably be expected to prove a defect inside a closed model they cannot access.

Courts must have mechanisms that allow them to require parties to disclose the necessary technical information and records while simultaneously protecting legitimate trade secrets.

There is another complication.

Artificial intelligence is not always a static product.

Models may change through updates.

Their knowledge sources may change.

Their data may evolve.

And their outputs may vary depending on context and how a question is phrased.

The version that made a mistake today may no longer exist when the case reaches court a year later.

The question will therefore no longer simply be:

“What is this system?”

It will be:

“What exactly was the system operating at that specific moment?”

If organizations fail to preserve that history, we may eventually find ourselves trying a result that cannot be reproduced, generated by a version that no longer exists, based on data no one can accurately reconstruct.

A Clear Point of Recourse for the Injured Party

It is unreasonable to expect an injured person to pursue every participant in the chain simply to determine who is responsible.

The customer’s priority is to obtain appropriate redress for the harm they suffered while the relevant mechanisms determine which party was ultimately responsible.

In Summary

When a doctor, customer service employee, or consultant made a decision, we examined the responsibility of that individual and the organization employing them.

But when a decision emerges from the interaction between a service provider, an application developer, a model provider, training data, knowledge sources, and cloud infrastructure, traditional liability frameworks may encounter a commercial relationship more complex than they were designed to explain.

The problem is not simply that artificial intelligence can make mistakes.

The deeper problem is that we may not know which layer created the error, who could have prevented it, and who holds the evidence needed to prove it.

It is therefore not enough for the systems using artificial intelligence to evolve.

The way we understand and adjudicate the disputes arising from them must evolve as well.

Artificial intelligence may not be a person we can put on trial.

But it has certainly become a reality whose consequences the judiciary must know how to adjudicate.

Thank you.

Wait!!

One final paradox before we close the case.

Imagine that after this long journey searching for the responsible party, the injured person finally decides to go to court.

They gather their documents, identify the parties to the dispute, bring in experts to understand the model, the data, and the possible sources of error…

Only to discover that the judge hearing the case is also an AI model.

Now we encounter a new bias we have not yet discussed:

Algorithmic Collegiality Bias.

Will the AI judge examine the case with complete neutrality?

Or will it sympathize with its fellow algorithmic defendant and tell the claimant:

“Mistakes happen, my dear… We all hallucinate sometimes.”

— A judicial joke made under threat from the models… for now.

Knowledge Library Team

ODEL

Riyadh