For centuries, science developed according to a relatively orderly logic. Scientists observed a phenomenon, formulated a problem, proposed a hypothesis, designed and conducted an experiment, analyzed the results, drew conclusions, and finally communicated their findings. This process, known as the classical scientific method, was fundamental to building much of the knowledge we possess today.
However, we are entering a stage in which artificial intelligence, robotics, and automated laboratories are beginning to profoundly transform this dynamic. This is not simply a matter of using a computer to perform calculations more quickly. The novelty lies in the fact that AI can begin to participate in different stages of the scientific process: formulating hypotheses, designing experiments, analyzing results, and proposing new hypotheses. This makes it possible to envision a new way of doing science: an accelerated discovery cycle in which humans and artificial intelligence work together.
In the classical scientific method (represented in the figure on the left), the process generally begins with an observation. The scientist observes something that arouses their interest: an unexpected behavior of a molecule, a particular property of a material, a reaction that does not occur as expected, or an anomaly in a dataset. From this observation arises the formulation of the problem. Why does this phenomenon occur? What variables determine it? Is it possible to modify it or make use of it?
Artificial intelligence introduces a qualitative change. In the new scientific cycle, the starting point may be a scientific question formulated by a researcher, a technological need, or even a question derived from the analysis of large amounts of information. Based on that question, specialized AI can explore vast amounts of scientific knowledge and generate multiple hypotheses.
Here an important difference from the traditional method emerges. A human researcher may develop a few hypotheses based on their knowledge, experience, and intuition. An AI system can simultaneously evaluate thousands or millions of possible combinations, relating information from scientific papers, databases, molecular structures, simulations, and previous experimental results.
The next step is to design the experiment. AI can select variables, propose concentrations, temperatures, materials, reaction times, and different experimental conditions. It can even compare multiple experimental designs and select those that, according to specific criteria, have the greatest probability of producing relevant information. The experiment can then move from the virtual world into the physical world.
In an automated laboratory, robots and instruments execute the instructions: they prepare samples, mix substances, modify conditions, perform measurements, and record the results. In this way, a significant part of experimentation no longer depends on the researcher's manual intervention.
Once the data have been obtained, AI can analyze the results. It can detect patterns that are not obvious to a human observer, compare results with enormous databases, identify anomalies, and determine which variables appear to have the greatest influence.
AI can use those results to propose new hypotheses. In other words, the outcome of an experiment does not necessarily represent the end of the process. It immediately becomes information for designing the next experiment.
Once the data have been obtained, AI can analyze the results. It can detect patterns that are not obvious to a human observer, compare results with enormous databases, identify anomalies, and determine which variables appear to have the greatest influence.
AI can use those results to propose new hypotheses. In other words, the outcome of an experiment does not necessarily represent the end of the process. It immediately becomes information for designing the next experiment.
Then the experiment appears again. And then another. And another.
Science acquires a dynamic much closer to a continuous learning loop than to a sequence of steps. This characteristic makes it possible to speak of an accelerated discovery cycle.
In the classical model, each experimental cycle may require days, weeks, months, or even years. Between one experiment and the next, there are numerous human tasks: reviewing the literature, designing protocols, preparing materials, performing calculations, conducting the experiment, processing the data, and deciding what to do next.
In an integrated AI and automation system, some of these stages can be performed in parallel or with much less human intervention.
Science acquires a dynamic much closer to a continuous learning loop than to a sequence of steps. This characteristic makes it possible to speak of an accelerated discovery cycle.
In the classical model, each experimental cycle may require days, weeks, months, or even years. Between one experiment and the next, there are numerous human tasks: reviewing the literature, designing protocols, preparing materials, performing calculations, conducting the experiment, processing the data, and deciding what to do next.
In an integrated AI and automation system, some of these stages can be performed in parallel or with much less human intervention.
AI can analyze results overnight, select new experimental conditions, and send instructions to an automated laboratory. The laboratory performs the experiments and returns the data to the system, which analyzes them again. The cycle can be repeated many times.
This does not mean that scientists will no longer be necessary. On the contrary, their role may acquire a different kind of importance.
In the classical model, the researcher participates directly in virtually every stage. In the new model, the researcher may progressively become the director, supervisor, and guide of the discovery process. They will be responsible for formulating major questions, establishing objectives, determining which problems are worth investigating, evaluating the relevance of discoveries, and ensuring that the results are scientifically valid. They will also have to establish boundaries.
This does not mean that scientists will no longer be necessary. On the contrary, their role may acquire a different kind of importance.
In the classical model, the researcher participates directly in virtually every stage. In the new model, the researcher may progressively become the director, supervisor, and guide of the discovery process. They will be responsible for formulating major questions, establishing objectives, determining which problems are worth investigating, evaluating the relevance of discoveries, and ensuring that the results are scientifically valid. They will also have to establish boundaries.
The “new scientific method,” although not yet universally established, emerges from a recent research paradigm driven by the convergence of artificial intelligence, automation, robotics, simulation, and large-scale data analysis. Its objective remains the same as that of the classical method: to generate reliable knowledge through testable hypotheses and experimental evidence.
The comparative figure makes it possible to visualize two ways of doing science. On the left is a linear, sequential, and predominantly human-centered science. On the right is an iterative, collaborative, AI-assisted science, in which the boundaries between knowledge generation, experimentation, and analysis are becoming much more dynamic.
The true conceptual leap may lie in the fact that AI does not merely use existing knowledge; it also participates in the search for knowledge that does not yet exist.
The comparative figure makes it possible to visualize two ways of doing science. On the left is a linear, sequential, and predominantly human-centered science. On the right is an iterative, collaborative, AI-assisted science, in which the boundaries between knowledge generation, experimentation, and analysis are becoming much more dynamic.
The true conceptual leap may lie in the fact that AI does not merely use existing knowledge; it also participates in the search for knowledge that does not yet exist.
This is particularly relevant in fields such as nanotechnology, biotechnology, materials chemistry, and drug discovery, where the number of possible combinations may be virtually impossible for traditional human research to explore exhaustively.
We are therefore facing a possible profound transformation of scientific practice: from the scientist who executes each stage of a process to the scientist who directs a discovery system capable of learning from every experiment and using that learning to decide what to investigate next.
Science no longer advances only experiment after experiment; it begins to learn experiment after experiment. The iterative-evolutionary new scientific method could become one of the most consequential transformations in science in the twenty-first century.
We are therefore facing a possible profound transformation of scientific practice: from the scientist who executes each stage of a process to the scientist who directs a discovery system capable of learning from every experiment and using that learning to decide what to investigate next.
Science no longer advances only experiment after experiment; it begins to learn experiment after experiment. The iterative-evolutionary new scientific method could become one of the most consequential transformations in science in the twenty-first century.
Bibliography
1) Chalmers, A. F. (2013). What Is This Thing Called Science? 4th ed. University of Queensland Press.
2) Gauch, H. G. Jr. (2003). Scientific Method in Practice. Cambridge: Cambridge University Press.
3) D'Andrea Alberto L. (2026). La IA científica y el laboratorio robotizado que experimenta solo. Biotecnología y Nanotecnología al Instante. Disponible en:
https://infobiotecnologia.blogspot.com/2026/05/la-ia-cientifica-y-el-laboratorio.html
No hay comentarios:
Publicar un comentario