{"id":1132,"date":"2026-11-05T06:58:11","date_gmt":"2026-11-05T06:58:11","guid":{"rendered":"https:\/\/perit.ai\/blogs\/?p=1132"},"modified":"2026-11-05T06:58:11","modified_gmt":"2026-11-05T06:58:11","slug":"can-ai-close-the-loop-between-discovery-and-experiment","status":"publish","type":"post","link":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/","title":{"rendered":"Can AI Close the Loop Between Discovery and Experiment?"},"content":{"rendered":"<p data-pm-slice=\"1 1 []\">Scientific discovery has always involved a difficult exchange between what researchers think might happen and what actually happens in the laboratory. An idea may look convincing in a mathematical model, a protein structure may appear promising on a computer screen, or a new material may show remarkable predicted properties. Yet none of these predictions becomes a scientific result until it survives contact with an experiment. This is where one of the most interesting developments in artificial intelligence is beginning to take shape: systems that do not simply suggest what scientists should investigate, but can also help perform the experiments, interpret the outcomes, and decide what to try next.<\/p>\n<p>For decades, computational science and experimental research have operated at different speeds. Modern AI models can evaluate enormous chemical spaces, predict molecular behaviour, and identify patterns across scientific datasets. Physical experiments, however, still depend on laboratory equipment, preparation procedures, measurements, and repeated validation. The gap between a promising prediction and a verified result remains substantial. Self-driving laboratories are emerging as a possible solution, bringing together machine learning, robotics, automated instruments, and experimental feedback to create a more continuous process of scientific investigation.<\/p>\n<p>The important question is no longer whether AI can make a useful scientific prediction. It is whether an AI-directed system can learn from the physical world when that prediction turns out to be wrong.<\/p>\n<h2>From Predicting Discoveries to Testing Them<\/h2>\n<p>The first major shift in AI-assisted science was the ability to identify possibilities that researchers might otherwise overlook. Machine learning models could examine biological sequences, molecular structures, and material properties, narrowing down enormous search spaces before scientists committed resources to physical testing. But prediction and experimental validation are fundamentally different tasks. A model can estimate that a compound should have a particular property without knowing whether the compound can actually be synthesized under practical laboratory conditions.<\/p>\n<p>Consider a hypothetical new material predicted to withstand extremely high temperatures. Its computational profile may look impressive, but the synthesis process could produce unwanted phases, the material might degrade under real conditions, or its measured properties could differ from the predicted values. Each possibility creates another research question that cannot be settled by computation alone.<\/p>\n<p>This is why the development of self-driving laboratories matters. Rather than treating computational predictions and physical experiments as separate stages, these systems connect them through an iterative feedback process. An algorithm proposes an experiment, robotic equipment carries it out, analytical instruments measure the result, and the collected data informs the next decision. The system can then modify experimental conditions, explore an alternative synthesis route, or investigate a different candidate.<\/p>\n<p>The underlying idea is often described as a closed-loop workflow. Instead of following a fixed sequence of experiments designed entirely in advance, the next experiment depends on what the previous one revealed. A failed result is therefore not necessarily wasted effort. It becomes another piece of evidence that can improve the next decision.<\/p>\n<div style=\"width: 100%; box-sizing: border-box; margin: 25px auto; padding: 15px 20px; background: #F0F0E9; border-left: 3px solid #537F7A;\">\n<p style=\"margin: 0; font-family: Georgia,serif; font-size: 16px; line-height: 1.75; color: #344b5d;\">A prediction describes what a model expects to happen. An experiment reveals what actually happens. The real value of autonomous discovery lies in connecting these two, allowing every physical result to influence the next computational decision.<\/p>\n<\/div>\n<div style=\"width: 100%; margin: 28px auto;\" align=\"center\">\n<table style=\"height: 831px;\" border=\"0\" width=\"351\" cellspacing=\"0\" cellpadding=\"13\" bgcolor=\"#F7F5EF\">\n<tbody>\n<tr>\n<td align=\"center\" bgcolor=\"#E8EDF1\"><strong><span style=\"color: #263d56;\">AI PREDICTION<\/span><\/strong><\/p>\n<p><span style=\"color: #45596b;\">Identifies a promising hypothesis<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\"><span style=\"color: #536f80; font-size: x-large;\">\u2193<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\" bgcolor=\"#E4EFED\"><strong><span style=\"color: #315f61;\">EXPERIMENT DESIGN<\/span><\/strong><\/p>\n<p><span style=\"color: #45596b;\">Selects conditions worth testing<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\"><span style=\"color: #536f80; font-size: x-large;\">\u2193<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\" bgcolor=\"#E8EDF1\"><strong><span style=\"color: #263d56;\">ROBOTIC EXECUTION<\/span><\/strong><\/p>\n<p><span style=\"color: #45596b;\">Performs the physical experiment<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\"><span style=\"color: #536f80; font-size: x-large;\">\u2193<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\" bgcolor=\"#E4EFED\"><strong><span style=\"color: #315f61;\">MEASUREMENT<\/span><\/strong><\/p>\n<p><span style=\"color: #45596b;\">Collects experimental evidence<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\"><span style=\"color: #536f80; font-size: x-large;\">\u2193<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\" bgcolor=\"#E8EDF1\"><strong><span style=\"color: #263d56;\">MODEL UPDATE<\/span><\/strong><\/p>\n<p><span style=\"color: #45596b;\">Learns from the observed outcome<\/span><\/td>\n<\/tr>\n<tr>\n<td align=\"center\">\n<p style=\"text-align: center;\"><span style=\"color: #536f80; font-size: x-large;\">\u2193<br \/>\n<\/span><strong><span style=\"color: #315f61;\">NEXT EXPERIMENT<\/span><\/strong><\/p>\n<p><span style=\"color: #45596b;\">Refines the next scientific decision<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p align=\"center\"><span style=\"color: #65717d; font-size: small;\">Figure 1. The closed-loop process connecting AI predictions, physical experimentation, and continuous scientific learning.<\/span><\/p>\n<p>This changes the role of AI from a prediction engine into an active participant in the experimental process. It also introduces a more demanding requirement: the system must understand not only what it wants to achieve, but how much confidence it should place in the evidence collected along the way.<\/p>\n<h2>Inside a Self-Driving Laboratory<\/h2>\n<p>A self-driving laboratory is not simply a conventional laboratory equipped with robotic arms. Its defining feature is the integration of automated experimentation with computational decision-making. Hardware performs physical operations, software coordinates the workflow, machine learning evaluates the results, and optimization algorithms determine which experiment should follow.<\/p>\n<div style=\"width: 100%; box-sizing: border-box; margin: 28px auto; padding: 22px 18px; background: #F7F5EF; border: 1px solid #DEDED6; font-family: Arial,sans-serif;\">\n<div style=\"text-align: center; color: #263d56; font-size: 16px; font-weight: 600; margin-bottom: 18px;\">Architecture of a Self-Driving Laboratory<\/div>\n<div style=\"padding: 15px 12px; background: #E8EDF1; border-left: 4px solid #536B82; text-align: center;\">\n<div style=\"font-size: 11px; letter-spacing: 1px; color: #657687;\">LAYER 01<\/div>\n<div style=\"font-size: 15px; font-weight: 600; color: #263d56; margin: 5px 0;\">AI &amp; Decision Engine<\/div>\n<div style=\"font-size: 12px; color: #526273;\">Machine learning \u00b7 Prediction \u00b7 Experiment selection<\/div>\n<\/div>\n<div style=\"text-align: center; color: #536f80; font-size: 22px; padding: 5px;\">\u2193<\/div>\n<div style=\"padding: 15px 12px; background: #E4EFED; border-left: 4px solid #537F7A; text-align: center;\">\n<div style=\"font-size: 11px; letter-spacing: 1px; color: #557774;\">LAYER 02<\/div>\n<div style=\"font-size: 15px; font-weight: 600; color: #315f61; margin: 5px 0;\">Workflow &amp; Control System<\/div>\n<div style=\"font-size: 12px; color: #526b68;\">Protocol generation \u00b7 Scheduling \u00b7 Instrument control<\/div>\n<\/div>\n<div style=\"text-align: center; color: #536f80; font-size: 22px; padding: 5px;\">\u2193<\/div>\n<div style=\"padding: 15px 12px; background: #E8EDF1; border-left: 4px solid #536B82; text-align: center;\">\n<div style=\"font-size: 11px; letter-spacing: 1px; color: #657687;\">LAYER 03<\/div>\n<div style=\"font-size: 15px; font-weight: 600; color: #263d56; margin: 5px 0;\">Robotic Experimentation<\/div>\n<div style=\"font-size: 12px; color: #526273;\">Sample preparation \u00b7 Mixing \u00b7 Heating \u00b7 Synthesis<\/div>\n<\/div>\n<div style=\"text-align: center; color: #536f80; font-size: 22px; padding: 5px;\">\u2193<\/div>\n<div style=\"padding: 15px 12px; background: #E4EFED; border-left: 4px solid #537F7A; text-align: center;\">\n<div style=\"font-size: 11px; letter-spacing: 1px; color: #557774;\">LAYER 04<\/div>\n<div style=\"font-size: 15px; font-weight: 600; color: #315f61; margin: 5px 0;\">Measurement &amp; Analysis<\/div>\n<div style=\"font-size: 12px; color: #526b68;\">Sensors \u00b7 Imaging \u00b7 Spectroscopy \u00b7 Data collection<\/div>\n<\/div>\n<div style=\"text-align: center; color: #537f7a; font-size: 22px; padding: 8px;\"><\/div>\n<div style=\"text-align: center; padding: 11px; border: 1px dashed #78928C; color: #315f61; font-size: 12px;\">Experimental feedback returns to the AI decision engine<\/div>\n<div style=\"margin-top: 15px; padding-top: 12px; border-top: 1px solid #DEDED6; text-align: center; color: #687582; font-size: 11px;\">Human oversight operates across all four layers<\/div>\n<\/div>\n<p style=\"text-align: center; color: #687582; font-size: 12px; margin-top: -17px;\">Figure 3. Functional architecture of a self-driving laboratory, connecting computational decisions with robotic execution and experimental feedback.<\/p>\n<p>The process generally begins with a research objective, such as improving a material&#8217;s conductivity, identifying an efficient catalyst, or optimizing a chemical reaction. The AI system considers available experimental data and proposes conditions worth investigating. These conditions may include temperature, concentration, reaction time, pressure, or the proportions of different ingredients.<\/p>\n<p>Robotic equipment then prepares and executes the experiment. Depending on the laboratory, this can involve liquid handling, weighing and mixing powders, controlling heating systems, transferring samples, or operating analytical instruments. Once the experiment is complete, measurements are collected and converted into data that the computational system can interpret.<\/p>\n<p>The most important stage comes after measurement. The AI does not simply record whether an experiment succeeded. It evaluates how the observed outcome compares with the expected result and uses that difference to update its understanding of the problem. An optimization method such as Bayesian optimization can help select the next experiment by balancing two competing goals: exploring uncertain regions of the experimental space and improving conditions that already appear promising.<\/p>\n<p>This balance is important because repeatedly testing only the most promising conditions can cause a system to miss unexpected discoveries. On the other hand, exploring every possibility is inefficient. A well-designed experimental agent must decide when to investigate something unfamiliar and when to refine an existing result.<\/p>\n<div style=\"width: 100%; box-sizing: border-box; margin: 28px auto; padding: 22px 20px; background: #F7F5EF; border: 1px solid #DEDED6;\">\n<div style=\"font-family: Arial,sans-serif; color: #263d56; font-size: 16px; font-weight: 600; margin-bottom: 8px;\">Bayesian Optimization: Selecting the Next Experiment<\/div>\n<div style=\"font-family: Georgia,serif; text-align: center; color: #263d56; font-size: 23px; padding: 22px 5px;\">x<sub>next<\/sub> = arg max <sub style=\"font-size: 12px;\">x<\/sub> \u03b1(x)<\/div>\n<div style=\"border-top: 1px solid #DEDED6; padding-top: 15px;\">\n<div style=\"font-family: Arial,sans-serif; color: #315f61; font-weight: 600; font-size: 13px; margin-bottom: 10px;\">What does this mean?<\/div>\n<div style=\"font-family: Arial,sans-serif; color: #45596b; font-size: 13px; line-height: 1.7;\">\n<p><strong>x<sub>next<\/sub><\/strong> \u2014 the next experimental condition selected.<\/p>\n<p><strong>\u03b1(x)<\/strong> \u2014 the acquisition function, which estimates how valuable an experiment may be.<\/p>\n<p><strong>arg max<\/strong> \u2014 selects the condition with the highest acquisition value.<\/p>\n<\/div>\n<\/div>\n<div style=\"margin-top: 15px; padding: 12px; background: #E4EFED; color: #315f61; font-family: Arial,sans-serif; font-size: 12px; line-height: 1.6;\">The acquisition function balances exploration of uncertain conditions with exploitation of conditions that already show promising results.<\/div>\n<\/div>\n<p style=\"text-align: center; color: #687582; font-size: 12px; margin-top: -17px;\">Figure 4. A simplified Bayesian optimization objective for selecting the next experiment based on an acquisition function.<\/p>\n<p>In this arrangement, the laboratory becomes more than an execution environment. It becomes part of the learning system itself.<\/p>\n<h2>When the Laboratory Starts Learning<\/h2>\n<p>One of the clearest demonstrations of this approach came from the A-Lab, an autonomous laboratory developed for synthesizing inorganic materials. Published in <em>Nature<\/em> in 2023, the research combined computational predictions, machine learning, scientific literature, robotic synthesis, and automated characterization to investigate new materials.<\/p>\n<p>The system was designed to address a persistent problem in materials science: identifying a potentially useful compound computationally does not guarantee that researchers can successfully produce it in the laboratory. A-Lab attempted to bridge this gap by generating synthesis procedures, carrying out experiments with robotic equipment, and analysing the resulting materials using X-ray diffraction.<\/p>\n<p>The reported results were significant. During 17 days of continuous operation, the laboratory performed 353 experiments and synthesized 36 compounds from 57 target materials. More importantly, when a synthesis did not produce the desired result, the system used the experimental analysis to propose improved follow-up procedures rather than simply moving to another candidate. <strong><a style=\"color: #537f7a; text-decoration: none;\" href=\"https:\/\/doi.org\/10.1038\/s41586-023-06734-w\" target=\"_blank\" rel=\"noopener noreferrer\">[1]<\/a><\/strong><\/p>\n<div style=\"width: 100%; box-sizing: border-box; margin: 30px auto; padding: 22px 20px; background: #F7F5EF; border: 1px solid #DEDED6;\">\n<div style=\"font-family: Arial,sans-serif; color: #263d56; font-size: 17px; font-weight: 600; margin-bottom: 6px;\">A-Lab: Experimental outcomes across 57 target materials<\/div>\n<div style=\"font-family: Arial,sans-serif; color: #687582; font-size: 12px; margin-bottom: 22px;\">Results from 17 days of autonomous experimentation<\/div>\n<div style=\"margin-bottom: 18px;\">\n<div style=\"display: flex; justify-content: space-between; font-size: 13px; color: #364b5e; margin-bottom: 7px;\">Successfully synthesized<br \/>\n<strong>36 (63%)<\/strong><\/div>\n<div style=\"height: 13px; background: #E3E5E1; border-radius: 2px;\">\n<div style=\"height: 13px; width: 63.16%; background: #537F7A; border-radius: 2px;\"><\/div>\n<\/div>\n<\/div>\n<div style=\"margin-bottom: 18px;\">\n<div style=\"display: flex; justify-content: space-between; font-size: 13px; color: #364b5e; margin-bottom: 7px;\">Inconclusive<br \/>\n<strong>4 (7%)<\/strong><\/div>\n<div style=\"height: 13px; background: #E3E5E1; border-radius: 2px;\">\n<div style=\"height: 13px; width: 7.02%; background: #A6A18B; border-radius: 2px;\"><\/div>\n<\/div>\n<\/div>\n<div>\n<div style=\"display: flex; justify-content: space-between; font-size: 13px; color: #364b5e; margin-bottom: 7px;\">Not obtained<br \/>\n<strong>17 (30%)<\/strong><\/div>\n<div style=\"height: 13px; background: #E3E5E1; border-radius: 2px;\">\n<div style=\"height: 13px; width: 29.82%; background: #9A7770; border-radius: 2px;\"><\/div>\n<\/div>\n<\/div>\n<div style=\"border-top: 1px solid #D9D9D1; margin-top: 23px; padding-top: 15px; display: flex; justify-content: space-between; gap: 10px; flex-wrap: wrap;\">\n<div>\n<div style=\"font-size: 23px; font-weight: 600; color: #263d56;\">353<\/div>\n<div style=\"font-size: 11px; color: #687582;\">Experiments conducted<\/div>\n<\/div>\n<div>\n<div style=\"font-size: 23px; font-weight: 600; color: #263d56;\">17<\/div>\n<div style=\"font-size: 11px; color: #687582;\">Days of operation<\/div>\n<\/div>\n<div>\n<div style=\"font-size: 23px; font-weight: 600; color: #263d56;\">57<\/div>\n<div style=\"font-size: 11px; color: #687582;\">Target materials<\/div>\n<\/div>\n<\/div>\n<\/div>\n<p style=\"text-align: center; color: #687582; font-size: 12px; margin-top: -17px;\">Figure 2. Experimental outcomes of the A-Lab, demonstrating the ability of autonomous systems to synthesize predicted inorganic materials while identifying unsuccessful synthesis pathways.<\/p>\n<div style=\"width: 100%; box-sizing: border-box; margin: 28px auto; padding: 22px 18px; background: #F7F5EF; border: 1px solid #DEDED6; font-family: Arial,sans-serif;\">\n<div style=\"text-align: center; color: #263d56; font-size: 16px; font-weight: 600; margin-bottom: 20px;\">How an Autonomous Laboratory Learns from Failure<\/div>\n<div style=\"display: flex; gap: 10px; align-items: stretch;\">\n<div style=\"flex: 1; padding: 13px 8px; background: #E8EDF1; text-align: center; border-top: 3px solid #536B82;\">\n<div style=\"font-size: 11px; color: #657687;\">STEP 01<\/div>\n<div style=\"font-size: 13px; color: #263d56; font-weight: 600; margin: 7px 0;\">Prediction<\/div>\n<div style=\"font-size: 11px; color: #526273; line-height: 1.5;\">AI selects a promising synthesis route<\/div>\n<\/div>\n<div style=\"align-self: center; color: #78908e; font-size: 19px;\">\u2192<\/div>\n<div style=\"flex: 1; padding: 13px 8px; background: #F0E9E5; text-align: center; border-top: 3px solid #9A7770;\">\n<div style=\"font-size: 11px; color: #876c66;\">STEP 02<\/div>\n<div style=\"font-size: 13px; color: #765d57; font-weight: 600; margin: 7px 0;\">Failure<\/div>\n<div style=\"font-size: 11px; color: #685c59; line-height: 1.5;\">The target material is not obtained<\/div>\n<\/div>\n<\/div>\n<div style=\"text-align: center; color: #537f7a; font-size: 21px; padding: 10px;\">\u2193<\/div>\n<div style=\"padding: 15px; background: #E4EFED; border-left: 4px solid #537F7A; text-align: center;\">\n<div style=\"font-size: 11px; color: #557774;\">STEP 03<\/div>\n<div style=\"font-size: 14px; color: #315f61; font-weight: 600; margin: 6px 0;\">Experimental Analysis<\/div>\n<div style=\"font-size: 12px; color: #526b68; line-height: 1.6;\">Measurements reveal what happened and provide new evidence<\/div>\n<\/div>\n<div style=\"text-align: center; color: #537f7a; font-size: 21px; padding: 8px;\">\u2193<\/div>\n<div style=\"padding: 15px; background: #E8EDF1; border-left: 4px solid #536B82; text-align: center;\">\n<div style=\"font-size: 11px; color: #657687;\">STEP 04<\/div>\n<div style=\"font-size: 14px; color: #263d56; font-weight: 600; margin: 6px 0;\">Model Update<\/div>\n<div style=\"font-size: 12px; color: #526273; line-height: 1.6;\">The system revises its understanding of the experimental conditions<\/div>\n<\/div>\n<div style=\"text-align: center; color: #537f7a; font-size: 21px; padding: 8px;\">\u2193<\/div>\n<div style=\"padding: 15px; background: #E4EFED; border-left: 4px solid #537F7A; text-align: center;\">\n<div style=\"font-size: 11px; color: #557774;\">STEP 05<\/div>\n<div style=\"font-size: 14px; color: #315f61; font-weight: 600; margin: 6px 0;\">Improved Experiment<\/div>\n<div style=\"font-size: 12px; color: #526b68; line-height: 1.6;\">A revised procedure is selected and tested<\/div>\n<\/div>\n<div style=\"text-align: center; color: #537f7a; font-size: 20px; padding-top: 12px;\">Feedback continues<\/div>\n<\/div>\n<p style=\"text-align: center; color: #687582; font-size: 12px; margin-top: -17px;\">Figure 5. Experimental feedback allows an autonomous laboratory to use unsuccessful synthesis attempts as evidence for refining subsequent experiments.<\/p>\n<p>This is what makes the experiment particularly relevant. The achievement was not merely that robots could prepare materials without constant human intervention. It was that experimental outcomes became inputs for subsequent scientific decisions.<\/p>\n<p>A newer direction is visible in research published in 2026. The RoboChem-Flex platform, for example, explores a modular and more accessible approach to autonomous chemical experimentation. Its combination of customizable hardware, Python-based control, and Bayesian optimization supports both autonomous and human-guided workflows.<br \/>\n<strong><a style=\"color: #537f7a; text-decoration: none;\" href=\"https:\/\/doi.org\/10.1038\/s44160-026-01053-0\" target=\"_blank\" rel=\"noopener noreferrer\">[2]<\/a><\/strong> This matters because the future of experimental AI will depend not only on how capable individual laboratories become, but also on whether such systems can be adapted without requiring every research group to build an expensive, specialized facility.<\/p>\n<p>Another important development involves AI systems that translate scientific instructions into executable laboratory procedures. The AutoLabs research, published in <em>Scientific Reports<\/em> in 2026, demonstrates a multi-agent architecture designed to generate and self-correct chemical protocols for robotic equipment. <strong><a style=\"color: #537f7a; text-decoration: none;\" href=\"https:\/\/doi.org\/10.1038\/s41598-026-45593-z\" target=\"_blank\" rel=\"noopener noreferrer\">[3]<\/a><\/strong> It highlights a different but equally important challenge: connecting high-level scientific reasoning with the precise operations that physical instruments must perform.<\/p>\n<p>Together, these developments suggest that autonomous experimentation is moving beyond isolated demonstrations toward more flexible research infrastructure.<\/p>\n<h2>The Problem With Letting AI Run Experiments<\/h2>\n<p>Despite the progress, a laboratory that can execute experiments autonomously is not necessarily a laboratory that can produce reliable scientific knowledge. Physical experiments introduce uncertainties that are difficult to capture completely in a computational model. Instruments can drift, samples can become contaminated, reactions can behave inconsistently, and measurements can contain noise.<\/p>\n<p>An AI system may also interpret an unexpected measurement as a meaningful scientific result when it actually reflects a procedural error. If the system uses that incorrect interpretation to design its next experiment, the mistake can propagate through the entire workflow.<\/p>\n<p>Reproducibility is another serious concern. A scientific result should not depend entirely on the particular robotic system, instrument configuration, or environmental conditions under which it was obtained. Researchers need access to experimental parameters, sample histories, calibration information, measurement procedures, and the decisions made by the AI throughout the process. Without this information, repeating an experiment or independently evaluating its conclusions becomes difficult.<\/p>\n<div style=\"width: 100%; box-sizing: border-box; margin: 25px auto; padding: 15px 20px; background: #F0F0E9; border-left: 3px solid #536B82;\">\n<p style=\"margin: 0; font-family: Georgia,serif; font-size: 16px; line-height: 1.75; color: #344b5d;\">Automation can accelerate an experiment, but it cannot independently establish why a discovery matters. Scientific judgment remains essential for distinguishing a promising measurement from reliable evidence and an interesting result from a meaningful discovery.<\/p>\n<\/div>\n<div style=\"width: 100%; box-sizing: border-box; margin: 28px auto; padding: 22px 18px; background: #F7F5EF; border: 1px solid #DEDED6; font-family: Arial,sans-serif;\">\n<div style=\"font-size: 16px; font-weight: 600; color: #263d56; text-align: center; margin-bottom: 20px;\">Two Approaches to Scientific Experimentation<\/div>\n<div style=\"display: flex; gap: 12px; align-items: stretch;\">\n<div style=\"flex: 1; min-width: 0;\">\n<div style=\"padding: 12px 7px; background: #E8EDF1; text-align: center; border-top: 3px solid #536B82; margin-bottom: 12px;\">\n<div style=\"font-size: 14px; font-weight: 600; color: #263d56;\">Human-led<\/div>\n<div style=\"font-size: 10px; color: #657687; margin-top: 4px;\">Traditional workflow<\/div>\n<\/div>\n<div style=\"font-size: 12px; color: #45596b; line-height: 1.7;\">\n<div style=\"padding: 9px 5px; border-bottom: 1px solid #DEDED6;\">Researcher selects experiments<\/div>\n<div style=\"padding: 9px 5px; border-bottom: 1px solid #DEDED6;\">Manual sample preparation<\/div>\n<div style=\"padding: 9px 5px; border-bottom: 1px solid #DEDED6;\">Human-operated measurements<\/div>\n<div style=\"padding: 9px 5px; border-bottom: 1px solid #DEDED6;\">Researcher interprets outcomes<\/div>\n<div style=\"padding: 9px 5px;\">Next experiment planned manually<\/div>\n<\/div>\n<\/div>\n<div style=\"width: 1px; background: #D8D8D0;\"><\/div>\n<div style=\"flex: 1; min-width: 0;\">\n<div style=\"padding: 12px 7px; background: #E4EFED; text-align: center; border-top: 3px solid #537F7A; margin-bottom: 12px;\">\n<div style=\"font-size: 14px; font-weight: 600; color: #315f61;\">AI-assisted<\/div>\n<div style=\"font-size: 10px; color: #557774; margin-top: 4px;\">Autonomous workflow<\/div>\n<\/div>\n<div style=\"font-size: 12px; color: #45596b; line-height: 1.7;\">\n<div style=\"padding: 9px 5px; border-bottom: 1px solid #DEDED6;\">Algorithm proposes experiments<\/div>\n<div style=\"padding: 9px 5px; border-bottom: 1px solid #DEDED6;\">Robotic sample preparation<\/div>\n<div style=\"padding: 9px 5px; border-bottom: 1px solid #DEDED6;\">Automated data collection<\/div>\n<div style=\"padding: 9px 5px; border-bottom: 1px solid #DEDED6;\">Model evaluates outcomes<\/div>\n<div style=\"padding: 9px 5px;\">Next experiment selected using feedback<\/div>\n<\/div>\n<\/div>\n<\/div>\n<div style=\"margin-top: 17px; padding: 13px; background: #E9E9E2; text-align: center;\">\n<div style=\"font-size: 12px; font-weight: 600; color: #263d56;\">Human scientific oversight remains essential<\/div>\n<div style=\"font-size: 11px; color: #596773; line-height: 1.6; margin-top: 5px;\">Both approaches require experimental validation, safety controls, reproducibility and scientific interpretation.<\/div>\n<\/div>\n<\/div>\n<p style=\"text-align: center; color: #687582; font-size: 12px; margin-top: -17px;\">Figure 6. A conceptual comparison of traditional human-led and AI-assisted experimental workflows. Actual laboratory configurations vary.<\/p>\n<p>There is also a distinction between optimizing a measurable outcome and understanding the underlying science. An algorithm may discover a combination of conditions that produces an unusually high experimental yield without necessarily explaining why those conditions work. Such a result can be useful, but scientific understanding requires more than finding a successful configuration.<\/p>\n<p>A 2026 review of self-driving laboratories identifies scalability, generalizability, and complete experimental data provenance as important requirements for the next generation of these systems. <strong><a style=\"color: #537f7a; text-decoration: none;\" href=\"https:\/\/doi.org\/10.1038\/s41570-026-00847-2\" target=\"_blank\" rel=\"noopener noreferrer\">[4]<\/a><\/strong> In practical terms, an autonomous workflow should be capable of transferring knowledge across different laboratories, working with different experimental setups, and preserving enough information for its results to be independently examined.<\/p>\n<p>Human supervision therefore remains essential. Researchers must define the scientific objective, establish safety boundaries, evaluate unusual results, and decide whether the evidence is sufficient to support a conclusion. Autonomy should reduce repetitive work and accelerate investigation, not remove scientific responsibility.<\/p>\n<h2>Toward a Continuous Scientific Discovery System<\/h2>\n<p>The longer-term possibility is a research environment in which computational models, autonomous laboratories, and human scientists work as interconnected parts of the same discovery process. A model could identify a promising hypothesis, an experimental agent could design and execute a test, and the resulting measurements could update the original model. The revised model could then suggest another experiment, creating a cycle that continues until the evidence becomes sufficiently strong or the research direction needs reconsideration.<\/p>\n<p>Such a system could be particularly valuable in fields where experimental conditions involve many interacting variables. Materials science, chemistry, biotechnology, and drug discovery all contain enormous spaces of possible combinations that cannot realistically be investigated one experiment at a time by human researchers.<\/p>\n<p>However, the goal should not be to create laboratories that operate without scientists. It should be to create laboratories that allow scientists to investigate more possibilities while spending less time on repetitive procedures and manual optimization. Researchers would still contribute domain knowledge, question unexpected findings, interpret scientific significance, and determine which discoveries deserve further investigation.<\/p>\n<p>There is a subtle but important difference between a machine that performs an experiment and a machine that contributes to scientific learning. The first automates laboratory work. The second uses experimental evidence to change what it believes, what it tests, and what it investigates next.<\/p>\n<p>That distinction may ultimately determine how transformative self-driving laboratories become.<\/p>\n<h2>The Experiment Is Not the End<\/h2>\n<p>AI has already demonstrated that it can identify patterns, generate hypotheses, and propose structures that would be difficult to explore through conventional research alone. But scientific discovery does not end when a promising possibility appears on a computer screen. It becomes meaningful when the physical world provides evidence that supports, challenges, or completely changes what researchers expected.<\/p>\n<p>Self-driving laboratories bring these stages closer together. By connecting computational reasoning with robotic experimentation and continuous feedback, they offer a way to make scientific investigation more iterative, adaptive, and efficient. The technology is still developing, and questions around reliability, reproducibility, cost, and human oversight remain unresolved. Nevertheless, the direction is becoming clearer.<\/p>\n<p>Perhaps the most important contribution of AI to science will not be its ability to predict what researchers should discover. It may be its ability to help researchers test those predictions, learn from the results, and return to the next question with better evidence than before.<\/p>\n<div style=\"width: 100%; margin: 35px auto 20px; font-family: Arial,sans-serif;\">\n<h2 style=\"font-family: Georgia,serif; font-size: 22px; font-weight: 600; color: #263d56; margin-bottom: 20px;\">References<\/h2>\n<ol style=\"padding-left: 22px; margin: 0; color: #45596b; font-size: 13px; line-height: 1.8;\">\n<li style=\"padding-left: 5px; margin-bottom: 15px;\">Szymanski, N. J., Rendy, B., Fei, Y., et al. (2023).<a style=\"color: #537f7a; text-decoration: none;\" href=\"https:\/\/doi.org\/10.1038\/s41586-023-06734-w\" target=\"_blank\" rel=\"noopener noreferrer\"><br \/>\nAn autonomous laboratory for the accelerated synthesis of inorganic materials.<\/a><br \/>\n<em>Nature, 624<\/em>, 86\u201391.<\/li>\n<li style=\"padding-left: 5px; margin-bottom: 15px;\">Pilon, S., Savino, E., Bayley, O. M., et al. (2026).<a style=\"color: #537f7a; text-decoration: none;\" href=\"https:\/\/doi.org\/10.1038\/s44160-026-01053-0\" target=\"_blank\" rel=\"noopener noreferrer\"><br \/>\nA flexible and affordable self-driving laboratory for automated reaction optimization.<\/a><br \/>\n<em>Nature Synthesis, 5<\/em>, 1552\u20131564.<\/li>\n<li style=\"padding-left: 5px; margin-bottom: 15px;\">Panapitiya, G., Saldanha, E., Job, H., &amp; Hess, O. (2026).<a style=\"color: #537f7a; text-decoration: none;\" href=\"https:\/\/doi.org\/10.1038\/s41598-026-45593-z\" target=\"_blank\" rel=\"noopener noreferrer\"><br \/>\nAutoLabs: Cognitive multi-agent systems with self-correction for autonomous chemical experimentation.<\/a><br \/>\n<em>Scientific Reports, 16<\/em>, 19554.<\/li>\n<li style=\"padding-left: 5px; margin-bottom: 15px;\">Canty, R. B., &amp; Abolhasani, M. (2026).<a style=\"color: #537f7a; text-decoration: none;\" href=\"https:\/\/doi.org\/10.1038\/s41570-026-00847-2\" target=\"_blank\" rel=\"noopener noreferrer\"><br \/>\nThe past, present and future of self-driving laboratories.<\/a><br \/>\n<em>Nature Reviews Chemistry, 10<\/em>, 523\u2013537.<\/li>\n<\/ol>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>Scientific discovery has always involved a difficult exchange between what researchers think might happen and what actually happens in the laboratory. An idea may look convincing in a mathematical model,\u2026<\/p>\n","protected":false},"author":1,"featured_media":1134,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[22,19],"tags":[],"class_list":["post-1132","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-artificial-intelligence","category-research"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.6 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Can AI Close the Loop Between Discovery and Experiment? - Perit Blog<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Can AI Close the Loop Between Discovery and Experiment? - Perit Blog\" \/>\n<meta property=\"og:description\" content=\"Scientific discovery has always involved a difficult exchange between what researchers think might happen and what actually happens in the laboratory. An idea may look convincing in a mathematical model,\u2026\" \/>\n<meta property=\"og:url\" content=\"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/\" \/>\n<meta property=\"og:site_name\" content=\"Perit Blog\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-05T06:58:11+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/11\/AI-Discovery-to-Experiment-Loop.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1672\" \/>\n\t<meta property=\"og:image:height\" content=\"941\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Kshitij@Perit\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Kshitij@Perit\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"12 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/\"},\"author\":{\"name\":\"Kshitij@Perit\",\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/#\\\/schema\\\/person\\\/954939b1549e8f3089b46850881d9775\"},\"headline\":\"Can AI Close the Loop Between Discovery and Experiment?\",\"datePublished\":\"2026-11-05T06:58:11+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/\"},\"wordCount\":2411,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/AI-Discovery-to-Experiment-Loop.webp\",\"articleSection\":[\"Artificial Intelligence\",\"Research\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/\",\"url\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/\",\"name\":\"Can AI Close the Loop Between Discovery and Experiment? - Perit Blog\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/AI-Discovery-to-Experiment-Loop.webp\",\"datePublished\":\"2026-11-05T06:58:11+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/#\\\/schema\\\/person\\\/954939b1549e8f3089b46850881d9775\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/#primaryimage\",\"url\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/AI-Discovery-to-Experiment-Loop.webp\",\"contentUrl\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/AI-Discovery-to-Experiment-Loop.webp\",\"width\":1672,\"height\":941},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/can-ai-close-the-loop-between-discovery-and-experiment\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Can AI Close the Loop Between Discovery and Experiment?\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/#website\",\"url\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/\",\"name\":\"Perit Blog\",\"description\":\"Frontier training data, collected and graded by people who do the work.\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/#\\\/schema\\\/person\\\/954939b1549e8f3089b46850881d9775\",\"name\":\"Kshitij@Perit\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/cropped-Perit-Logo-150x150.png\",\"url\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/cropped-Perit-Logo-150x150.png\",\"contentUrl\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/cropped-Perit-Logo-150x150.png\",\"caption\":\"Kshitij@Perit\"},\"sameAs\":[\"https:\\\/\\\/perit.ai\\\/blogs\"],\"url\":\"https:\\\/\\\/perit.ai\\\/blogs\\\/author\\\/perit_ops\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Can AI Close the Loop Between Discovery and Experiment? - Perit Blog","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/","og_locale":"en_US","og_type":"article","og_title":"Can AI Close the Loop Between Discovery and Experiment? - Perit Blog","og_description":"Scientific discovery has always involved a difficult exchange between what researchers think might happen and what actually happens in the laboratory. An idea may look convincing in a mathematical model,\u2026","og_url":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/","og_site_name":"Perit Blog","article_published_time":"2026-11-05T06:58:11+00:00","og_image":[{"width":1672,"height":941,"url":"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/11\/AI-Discovery-to-Experiment-Loop.webp","type":"image\/webp"}],"author":"Kshitij@Perit","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Kshitij@Perit","Est. reading time":"12 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/#article","isPartOf":{"@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/"},"author":{"name":"Kshitij@Perit","@id":"https:\/\/perit.ai\/blogs\/#\/schema\/person\/954939b1549e8f3089b46850881d9775"},"headline":"Can AI Close the Loop Between Discovery and Experiment?","datePublished":"2026-11-05T06:58:11+00:00","mainEntityOfPage":{"@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/"},"wordCount":2411,"commentCount":0,"image":{"@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/#primaryimage"},"thumbnailUrl":"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/11\/AI-Discovery-to-Experiment-Loop.webp","articleSection":["Artificial Intelligence","Research"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/","url":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/","name":"Can AI Close the Loop Between Discovery and Experiment? - Perit Blog","isPartOf":{"@id":"https:\/\/perit.ai\/blogs\/#website"},"primaryImageOfPage":{"@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/#primaryimage"},"image":{"@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/#primaryimage"},"thumbnailUrl":"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/11\/AI-Discovery-to-Experiment-Loop.webp","datePublished":"2026-11-05T06:58:11+00:00","author":{"@id":"https:\/\/perit.ai\/blogs\/#\/schema\/person\/954939b1549e8f3089b46850881d9775"},"breadcrumb":{"@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/#primaryimage","url":"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/11\/AI-Discovery-to-Experiment-Loop.webp","contentUrl":"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/11\/AI-Discovery-to-Experiment-Loop.webp","width":1672,"height":941},{"@type":"BreadcrumbList","@id":"https:\/\/perit.ai\/blogs\/can-ai-close-the-loop-between-discovery-and-experiment\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/perit.ai\/blogs\/"},{"@type":"ListItem","position":2,"name":"Can AI Close the Loop Between Discovery and Experiment?"}]},{"@type":"WebSite","@id":"https:\/\/perit.ai\/blogs\/#website","url":"https:\/\/perit.ai\/blogs\/","name":"Perit Blog","description":"Frontier training data, collected and graded by people who do the work.","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/perit.ai\/blogs\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Person","@id":"https:\/\/perit.ai\/blogs\/#\/schema\/person\/954939b1549e8f3089b46850881d9775","name":"Kshitij@Perit","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/09\/cropped-Perit-Logo-150x150.png","url":"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/09\/cropped-Perit-Logo-150x150.png","contentUrl":"https:\/\/perit.ai\/blogs\/wp-content\/uploads\/2026\/09\/cropped-Perit-Logo-150x150.png","caption":"Kshitij@Perit"},"sameAs":["https:\/\/perit.ai\/blogs"],"url":"https:\/\/perit.ai\/blogs\/author\/perit_ops\/"}]}},"_links":{"self":[{"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/posts\/1132","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/comments?post=1132"}],"version-history":[{"count":1,"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/posts\/1132\/revisions"}],"predecessor-version":[{"id":1133,"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/posts\/1132\/revisions\/1133"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/media\/1134"}],"wp:attachment":[{"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/media?parent=1132"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/categories?post=1132"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/perit.ai\/blogs\/wp-json\/wp\/v2\/tags?post=1132"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}