Adaptive AI Systems in Engineering

What happens when you aim industrial AI at production scheduling but treat it like every other engineering problem? We built a multi-agent AI system that achieved a 21% increase in profit. Here’s how: 1. Make the goals explicit Production scheduling is a complex process with numerous trade-offs. Highest demand or most efficient run? Overtime or on-time delivery? We spelled out the real goals and KPIs so the agent system knew exactly which knot it had to untangle. 2. Capture expertise through machine teaching Machine teaching breaks the job into bite-size skills. An engineer shows the system why a decision works, not just what happened in the data. Rather than rely purely on data, machine teaching transfers deep human expertise into the system - digitizing decades of experience and knowledge, crucial as expert operators retire. 3. Structuring the Multi-Agent System The multi-agent system was designed to mimic human decision-making: Sensors: Gather real-time data on production status, resources, and external market conditions. Skills: Modular units responsible for specific actions, such as forecasting demand, optimizing scheduling, or adapting to sudden changes. Each skill can evolve on its own, giving the plant the same modular flexibility you expect from any well-engineered system. 4. Establishing a Performance Benchmark Good engineering demands clear benchmarks. We ran a standard optimization-based system as our baseline. This allowed us to objectively measure whether our AI agents delivered measurable improvements. 5. Rigorous Testing & Iteration Engineering thrives on iteration. We created and tested 13 agent system designs, continuously iterating based on performance data. Each iteration leveraged insights from the previous, systematically improving performance until we identified the optimal solution. --- By treating AI as an engineered system (modular, explainable, and configurable) it demonstrates significant potential results: ✅ 21% higher profit margins ✅ Improved adaptability to rapidly changing market conditions ✅ Preservation and amplification of valuable human expertise Full breakdown of the build and tests is below.👇 #ProductionScheduling #IndustrialAI #MachineTeaching #SmartManufacturing

How do materials fail, and how can we design stronger, tougher, and more resilient ones? Published in #PNAS, our physics-aware AI model integrates advanced reasoning, rational thinking, and strategic planning capabilities models with the ability to write and execute code, perform atomistic simulations to solicit new physics data from “first principles”, and conduct visual analysis of graphed results and molecular mechanisms. By employing a multiagent strategy, these capabilities are combined into an intelligent system designed to solve complex scientific analysis and design tasks, as applied here to alloy design and discovery. This is significant because our model overcomes the limitations of traditional data-driven approaches by integrating diverse AI capabilities—reasoning, simulations, and multimodal analysis—into a collaborative system, enabling autonomous, adaptive, and efficient solutions to complex, multiobjective materials design problems that were previously slow, expert-dependent, and domain-specific. Wonderful work by my postdoc Alireza Ghafarollahi! Background: The design of new alloys is a multiscale problem that requires a holistic approach that involves retrieving relevant knowledge, applying advanced computational methods, conducting experimental validations, and analyzing the results, a process that is typically slow and reserved for human experts. Machine learning can help accelerate this process, for instance, through the use of deep surrogate models that connect structural and chemical features to material properties, or vice versa. However, existing data-driven models often target specific material objectives, offering limited flexibility to integrate out-of-domain knowledge and cannot adapt to new, unforeseen challenges. Our model overcomes these limitations by leveraging the distinct capabilities of multiple AI agents that collaborate autonomously within a dynamic environment to solve complex materials design tasks. The proposed physics-aware generative AI platform, AtomAgents, synergizes the intelligence of LLMs and the dynamic collaboration among AI agents with expertise in various domains, incl. knowledge retrieval, multimodal data integration, physics-based simulations, and comprehensive results analysis across modalities. The concerted effort of the multiagent system allows for addressing complex materials design problems, as demonstrated by examples that include autonomously designing metallic alloys with enhanced properties compared to their pure counterparts. We demonstrate accurate prediction of key characteristics across alloys and highlight the crucial role of solid solution alloying to steer the development of alloys. Paper: https://lnkd.in/enusweMf Code: https://lnkd.in/eWv2eKwS MIT Schwarzman College of Computing MIT Civil and Environmental Engineering MIT Department of Mechanical Engineering (MechE) MIT Industrial Liaison Program MIT School of Engineering

🔄 "How do you make AI systems that can reorganize themselves like the human brain does?" The team at Sakana AI just answered this with Transformer² - a breakthrough that lets language models rewire themselves in real-time based on the task at hand, just like our brains activate different regions for different activities. Here's why this is interesting! Traditional fine-tuning is like forcing a model to be good at everything simultaneously. Transformer² instead uses a two-pass approach: first identifying the task type, then dynamically mixing "expert" modules for optimal performance. Think of it as assembling the perfect team of specialists for each specific challenge. The results are ... compelling: - Outperforms LoRA (a popular fine-tuning method) while using <10% of the parameters - Demonstrates consistent gains across model scales (8B to 70B parameters) - Adapts effectively to entirely new tasks it wasn't trained for - Shows surprising versatility in vision-language tasks with 39% performance gains But here's the brilliant technical insight: Instead of modifying entire neural networks, Transformer² only adjusts the singular components of weight matrices – like precisely turning specific knobs rather than rebuilding the whole machine. Could this be the first step toward truly self-organizing AI systems? What industries do you think would benefit most from adaptive AI? 📄 Paper linked in comments #AI #MachineLearning #DeepLearning #AdaptiveAI #NeuromorphicComputing

Self-Adaptive Learning: Smarter LLM/SLM in Real-Time (Models that thinks, learns, and adapts—no retraining required). Traditional AI models are limited by their static nature, requiring massive datasets and periodic retraining to stay relevant. This process is costly, time-consuming, and struggles to keep pace with real-world dynamics. Self-adaptive learning rewrites the rules. These models evolve in real-time, updating their weights as they interact with the world—just like biological neural networks adapting to every new challenge. No bloated training infrastructure. No delays. Just relentless, on-the-fly learning and unstoppable improvement. The result? Models that personalize outputs on the fly, handle edge cases effortlessly, and remain scalable in fast-changing environments. By cutting resource overhead and accelerating adaptation, self-adaptive learning paves the way for more intelligent, efficient, and impactful AI systems. https://lnkd.in/eGRh-7wn