A new hybrid approach to artificial intelligence could slash energy consumption by up to 100 times while actually improving performance on complex tasks. Researchers at Tufts University have developed a neuro-symbolic AI system that combines traditional neural networks with human-like logical reasoning, offering a radically different path forward as AI's electricity demands threaten to overwhelm power grids .

Why Is AI Consuming So Much Energy Right Now?

The numbers are staggering. According to the International Energy Agency, AI systems and data centers consumed approximately 415 terawatt hours of power in 2024, accounting for more than 10% of total U.S. electricity production . That's enough energy to power a small to mid-size city. The problem is accelerating; demand is projected to double by 2030 as companies race to build larger data centers and deploy more AI applications. Some of these facilities require hundreds of megawatts of electricity, creating a sustainability crisis that researchers say current approaches cannot solve indefinitely.

The inefficiency stems from how today's most popular AI systems work. Large language models like ChatGPT and Gemini rely on statistical pattern matching from massive training datasets. They predict the next word or action in a sequence through brute-force computation, which consumes enormous amounts of energy even for relatively simple tasks. Matthias Scheutz, Karol Family Applied Technology Professor at Tufts, offered a striking comparison: an AI summary at the top of a Google search page can consume up to 100 times more energy than generating the website listings below it .

How Does Neuro-Symbolic AI Work Differently?

Rather than relying solely on data and trial-and-error learning, neuro-symbolic AI combines neural networks with symbolic reasoning, which uses explicit rules and abstract concepts. This mirrors how humans approach problems by breaking them into logical steps and categories. The Tufts team focused specifically on visual-language-action (VLA) models, which are AI systems used in robotics that take in visual data from cameras and language instructions, then translate that information into real-world physical actions like controlling a robot's arms or wheels .

Traditional VLA systems struggle with tasks that require planning and precision. If a robot is asked to stack blocks into a tower, it must analyze the scene, identify each block, and determine placement through repeated attempts. Shadows may confuse the system about a block's shape, or the robot may place pieces incorrectly, causing the structure to collapse. Symbolic reasoning prevents these errors by applying logical rules that limit unnecessary trial and error, allowing the system to reach solutions much faster and more reliably.

"A neuro-symbolic VLA can apply rules that limit the amount of trial and error during learning and get to a solution much faster. Not only does it complete the task much faster, but the time spent on training the system is significantly reduced," explained Matthias Scheutz, Karol Family Applied Technology Professor at Tufts University.

Matthias Scheutz, Karol Family Applied Technology Professor at Tufts University

What Were the Actual Performance Results?

The researchers tested their neuro-symbolic system using the Tower of Hanoi puzzle, a classic problem that requires careful planning and logical sequencing. The results were dramatic. The neuro-symbolic VLA achieved a 95% success rate, compared with just 34% for standard systems . When given a more complex version of the puzzle that the system had never encountered before, the hybrid approach still succeeded 78% of the time, while traditional models failed every single attempt.

Training time collapsed as well. The new system learned the task in only 34 minutes, while conventional models required more than a day and a half. Energy consumption during both training and operation dropped sharply. Training the neuro-symbolic model required only 1% of the energy used by a standard VLA system, and during operation, it consumed just 5% of the energy needed by conventional approaches . These figures translate to the claimed 100x reduction in overall energy use across the full lifecycle of the system.

How to Evaluate AI Systems for Energy Efficiency

• Training Energy Cost: Measure how much electricity is required to teach the system a new task. Neuro-symbolic approaches use symbolic rules to reduce redundant learning cycles, cutting training energy to a fraction of traditional methods.
• Operational Energy Consumption: Track power usage during actual deployment. Systems that rely on logical reasoning rather than statistical pattern matching consume significantly less electricity per task completed.
• Accuracy and Reliability: Compare success rates on structured tasks that require planning and logical sequencing. Hybrid systems that combine learning with reasoning typically outperform pure neural approaches on these benchmarks.
• Generalization to New Problems: Test whether the system can handle variations it has never seen before. Symbolic reasoning enables better transfer to novel situations without retraining.

What Does This Mean for the Future of AI?

The researchers argue that current approaches based on large language models and visual-language-action systems may not be sustainable long-term. While these systems are powerful, they consume disproportionate amounts of energy and can still produce unreliable results, including hallucinations and false information. Neuro-symbolic AI offers a different direction by combining learning with structured reasoning, potentially providing a more efficient and dependable foundation for future AI systems .

The work will be presented at the International Conference of Robotics and Automation in Vienna in May 2026 and will appear in the conference proceedings. The research team published their findings on arXiv in February 2026, with the paper titled "The Price Is Not Right: Neuro-Symbolic Methods Outperform VLAs on Structured Long-Horizon Manipulation Tasks with Significantly Lower Energy Consumption" . As companies continue building massive data centers and governments grapple with AI's impact on power infrastructure, this hybrid approach may represent a critical shift toward sustainable artificial intelligence.