Major scientific breakthroughs often begin when we find a new way to observe reality. The telescope revealed that we were not at the center of the universe. The microscope uncovered a world invisible to the human eye, opening new possibilities for biology and medicine. X-rays and magnetic resonance imaging made it possible to see inside the human body without surgery. In all these cases, the world remained the same; what changed was our ability to observe it. Now it is artificial intelligence’s (AI) turn.
Although this technology is commonly associated with conversational assistants, content generation, and task automation, AI systems can also analyze images, recognize patterns, detect anomalies, and uncover signals that are difficult to identify through conventional observation. By expanding our ability to perceive what previously remained hidden—even from specialists with years of experience—we can discover new relationships across vast amounts of information.
Learning to Observe Data with AI
In the age of big data, millions of data points are generated every day through sensors, cameras, monitoring systems, and digital platforms. Yet accumulating data and understanding it are two very different things. For a long time, discoveries depended on asking better questions and designing variables capable of answering them. To some extent, we could only find what we had previously decided to measure. But by learning directly from large volumes of information, AI can identify unexpected patterns and unforeseen relationships. Rather than merely confirming hypotheses, AI is beginning to reveal new possibilities for exploration.
In a recent study published in Nature, we tested this new way of observing at the microscopic scale by analyzing the patterns that emerge when small droplets of plant extracts evaporate. At first glance, many of these structures appeared similar, but by incorporating deep learning models, it became possible to automatically identify, and group patterns of shapes and textures based on their visual characteristics. AI revealed differences that would otherwise have been difficult to analyze systematically.
This finding suggests a shift in the way we generate knowledge. If an image can contain information that was previously invisible to the human eye, the same question is relevant for any organization: how many opportunities, risks, or processes remain hidden within the vast amounts of data a company generates? In an environment where virtually every organization has access to large quantities of information, the advantage will belong to those capable of extracting knowledge that competitors have yet to discover.
When Images Explain Processes
The possibilities of AI go beyond recognizing differences. If this technology can identify hidden patterns, it may also reveal how a phenomenon occurs. The distinction is subtle, but it completely changes the purpose of the analysis. An image ceases to be merely a visual representation and begins to become a source of evidence about the processes that produced it.
In a study published in The International Journal of Advanced Manufacturing Technology, we tested this capability in the analysis of metallic coatings used in manufacturing processes. At first glance, two surfaces may appear virtually identical and meet the same quality standards; however, small variations in their microstructure reflect differences in manufacturing conditions. While traditional methods identify these variations through specialist expertise and complex testing, AI models were able to recover information about the manufacturing process from the result. This contribution demonstrated that a micrograph could contain far richer information than can be interpreted through conventional observation. The material’s surface ceases to be a simple object of inspection and becomes a record of its own history.
This idea extends far beyond the laboratory. Organizations typically use the information they generate to describe processes that have already occurred; AI can reveal relationships that explain why they occurred, and which signals might anticipate future changes. The next competitive advantage, therefore, will not depend solely on having more information, but on developing a new capacity to interpret it. If an algorithm can uncover invisible relationships and use them to understand physical and biological processes, could it also learn how experts observe reality?
The organization that 'learns to see'
As the previous examples illustrate, artificial intelligence is moving beyond its role as a tool for classifying information to become an instrument capable of expanding human perception. The ability to identify patterns takes on a different dimension when human expertise is incorporated into the process. Can AI learn how an expert observes a phenomenon?
That question was the starting point for another study whose objective was not to improve the accuracy of a model, but to harness the accumulated experience of specialists to recognize qualitative patterns that are difficult to express through mathematical rules. Rather than training the algorithm solely with objective labels, the model learned from the human interpretation of images. It was then able to reproduce this approach across approximately thirty thousand records, identifying relationships that classical methods had not revealed.
AI is often presented as a technology intended to replace humans, but these results show that the most significant advances emerge when specialist expertise is combined with the analytical capacity of algorithms. Human knowledge provides context, judgment, and an understanding of the problem, while AI extends that expertise by exploring millions of combinations and maintaining a level of consistency that is nearly impossible to achieve through human observation alone. It does not replace expert perception; it projects it to an unprecedented scale.
How can an organization 'learn to see'?
For executives, expanding an organization’s observational capacity through artificial intelligence does not necessarily begin with choosing a technology. First, they need to recognize the signals that are not yet being fully used:
- Identify where visibility is lost. Detect processes or decisions in which the volume or complexity of information exceeds the organization’s capacity to analyze it systematically.
- Recognize what information is already available. Before generating new data, review the images, transactions, sensor data, operational records, or customer interactions the organization already produces—and how much of that information is being used.
- Explore beyond known questions. Use AI not only to answer predefined questions, but also to search for patterns, anomalies, and relationships that have not previously been considered.
- Interpret findings through expert knowledge. A pattern identified by an algorithm does not constitute an explanation. The experience of those who understand the business is essential to determine what it means, why it may be occurring, and whether it truly matters.
- Turn observations into decisions. The advantage emerges when a previously hidden signal makes it possible to anticipate a risk, discover an opportunity, better understand a process, or make a different decision.
The ability to observe through AI requires organizations to rethink how they understand innovation. Competitive advantage no longer depends solely on generating vast amounts of data or having more algorithms. Organizations that learn to ask better questions and integrate AI findings with the expertise of their teams will derive greater value from this technology. Those that learn to use this new instrument first will not only understand what is happening more clearly; they will also be able to transform it before others do.
References
Acuña, C., Mier y Terán, A., Kokornaczyk, M. O., Baumgartner, S., & Castelán, M. (2022). Deep learning applied to analyze patterns from evaporated droplets of Viscum album extracts. Scientific Reports, 12(1), 15332.
Guglielmetti, G., Castelán, M., Miguel Sánchez, D. K., Acuña, C., Martin, D., Baumgartner, S., & Tournier, A. L. (2026). Human perception-based deep learning classification: an exploratory application to a large chronobiology dataset. in silico Plants, 8(2), diag015.
Acuña, C., Chávez, D., Velázquez, C., Velazco, D., Vargas, G., & Castelán, M. (2025). Deep learning-based identification of electrodeposition time: a case study using synthetic images of copper coatings. The International Journal of Advanced Manufacturing Technology, 138(5), 2109–2119.
Kokornaczyk, M. O., Acuña, C., Mier y Terán, A., Castelán, M., & Baumgartner, S. (2024). Vortex-like vs. turbulent mixing of a Viscum album preparation affects crystalline structures formed in dried droplets. Scientific Reports, 14(1), 12965.
