{"data":{"id":1327,"lang":"en","title":"Tufts Engineers Build Neuro-Symbolic AI That Uses 99% Less Energy and Outperforms Standard Models","summary":"Tufts University researchers cut AI energy use by 99% and boosted accuracy from 34% to 95% using a neuro-symbolic approach combining neural networks with rule-based reasoning.","category":"technology","category_label":"Technology","published_at":"2026-04-07T05:01:55Z","updated_at":"2026-04-07T05:01:55Z","url":"https://techtools.cz/tools/news/article/1327-tufts-engineers-build-neuro-symbolic-ai-that-uses-99-less-energy-and","api_url":"https://techtools.cz/news-api/v1/articles/1327?lang=en","markdown_url":"https://techtools.cz/news-api/v1/articles/1327.md?lang=en","image_url":"https://www.sciencedaily.com/images/1920/server-facility-at-sandia-national-laboratory.webp","word_count":190,"reading_time_minutes":1,"sources":[{"url":"https://www.sciencedaily.com/releases/2026/04/260405003952.htm","domain":"sciencedaily.com"},{"url":"https://arxiv.org/abs/2602.19260","domain":"arxiv.org"}],"translation":{"lang":"cs","id":1385,"title":"Inženýři z Tufts University vyvinuli neurosymbolickou AI, která spotřebuje o 99 % méně energie a překonává standardní modely","url":"https://techtools.cz/tools/news/clanek/1385-inzenyri-z-tufts-university-vyvinuli-neurosymbolickou-ai-ktera-spotrebuje-o-99","api_url":"https://techtools.cz/news-api/v1/articles/1385?lang=cs"},"content":"Researchers at Tufts University's School of Engineering have developed a hybrid AI system that combines neural networks with symbolic reasoning, slashing energy consumption by up to 99% while dramatically outperforming conventional approaches. The team, led by Professor Matthias Scheutz, demonstrated that their neuro-symbolic method reduced training time from over 36 hours to just 34 minutes and used only 1% of the energy during training and 5% during operation compared to standard AI models.\n\nIn testing on the Tower of Hanoi puzzle, the hybrid system achieved a 95% success rate while traditional models managed only 34%. When faced with unfamiliar puzzle variations, the gap widened further — the neuro-symbolic system solved 78% of novel scenarios while conventional AI models failed entirely. The key innovation lies in pairing statistical pattern matching with rule-based abstract reasoning, allowing the system to generalize far beyond its training data.\n\nThe breakthrough arrives at a critical moment. According to the International Energy Agency, AI systems and data centers consumed roughly 415 terawatt-hours of electricity in 2024, and demand is accelerating. The research will be presented at the International Conference on Robotics and Automation in Vienna in May 2026."}}