What role does the Human play in industrial AI solutions?
- November 26, 2025
- 3 mins
For an AI system in industry to be meaningful, it must deliver added value, whether qualitative or quantitative. To achieve this, the AI system will leverage data to provide insights that help humans in decision-making or, in some cases, recommend optimized decisions. A crucial question for the project’s success then arises: where should we draw the line between what the Machine does and what it leaves to the Human? This question, more complex than it seems, can find an answer through both technical and methodological approaches.
The Power of Computation vs. Human Sensitivity
In many industrial AI applications, such as delivery route optimization, substantial gains on paper are often observed between machine recommendations and those produced by human experts (gains of 15-30% are common in route optimization). Facing these potential gains, the temptation is to assign total decision-making responsibility to the machine, given its evident superiority over experts (a superiority that is not accidental; the high computational complexity means it is no surprise that a computer can test many more scenarios than an expert).
However, if we attempt to completely replace humans with AI, we quickly realize there are many aspects that the machine cannot handle, such as:
- Customer Relationship: For example, in cases of high demand, deciding which customers to disappoint and which to prioritize, and respecting customer habits to maintain a good relationship.
- Human Management: For example, allowing a driver going through a difficult period or a junior driver to have lighter routes or more flexible schedules.
- Local Knowledge: For example, when certain delivery areas have significant security concerns that increase delivery time.
It is the strength of humans: they manage to maintain a comprehensive view of issues beyond just short-term productivity to ensure the long-term sustainability of industrial activity.
Thus, the dilemma quickly arises: should we choose between the significant performance gains that AI can bring, and the complex management of multiple priorities that humans handle daily in the industry?
What if the answer was: both?
The Human-Machine Synergy: The Challenge of Balanced and Optimized Task Sharing
The difficulty of the issue lies in the fact that we are trying to make two radically different ways of thinking interact. For example, in delivery routes, while a human might base routes on neighborhoods or postal codes to assist them, a machine can transcend these limitations to achieve all possible performance gains. As a result, the machine may produce routes that seem unreadable to the expert, as they do not initially appear “logical.” Indeed, the machine has not followed the expert’s “logic,” which is essentially a manual route construction method aimed at producing a correct result.
The first step is therefore a step of trust from the expert to the Machine. However, AI must in turn allow humans to regain control over what is happening. This is the main challenge.
Indeed, without AI assistance, humans have total control over their decisions. At least on the surface, as in reality, their computational capacity only allows them to explore a few possibilities compared to the millions that the Machine can handle. This total control thus translates into practice as a tedious, manual manipulation of data to produce a decision. If AI takes over this ‘calculation’ part, it is natural to think that humans can then achieve a higher, more strategic level of control over the decision. However, this requires providing them with the means to exercise such control!
Augmented Decision-Making: An Iterative and Collaborative Process Between Humans and Machines
To put it simply, one can consider that humans are capable of making decisions that are imperfect but complete (in the sense that they encompass all important aspects), whereas machines are adept at producing perfect but incomplete decisions. The goal is to establish an iterative system of suggestion and adjustment, where the Machine generates an optimized decision through computation, which the human expert reviews and adjusts by providing additional information (constraints, data changes, etc.) back to the Machine. The Machine then revises its suggestion accordingly, and this process continues until final validation by the human. Ultimately, the human has the final say, as they are the only ones who can truly “know” if everything is right. Throughout the process, the Machine utilizes its remarkable computational capabilities to offer optimized decisions at each stage.
This is the kind of system we implement at Kardinal, which is why our algorithms operate continuously: it allows them to make the best possible recommendation in real-time as the human expert provides ongoing adjustments.
Of course, this presents significant technical and theoretical challenges, as AI must solve a problem iteratively while being subordinate to human input, which is very different from the theoretical problems studied in a lab setting. Nevertheless, it is one of the keys to successful decision-making AI projects.
The Author
Cédric Hervet is a PhD in Applied Mathematics and Co-Founder of Kardinal. For over a decade, he has been researching and designing Artificial Intelligence systems for industrial applications in telecommunications, digital marketing, and transportation.
His dual expertise in statistics/Machine Learning and algorithmics/Operations Research allows him to integrate these two major sets of techniques to design the intelligent systems of tomorrow.
