research<\/a>, published in the high-impact Journal of Manufacturing Systems in April, introduced a framework that combines Agentic AI with multi-agent systems,
\n where multiple specialized AI \u201cagents\u201d work together to complete tasks. The framework
\n was tested by using real data to demonstrate how it could function in industrial practice.<\/p>\n\u201cAgentic AI is not new; it\u2019s become reinvigorated with the possibilities that GenAI
\n models, such as ChatGPT and Gemini provide to understand and work with human language,\u201d
\n Wuest says.<\/p>\n
Despite its potential, applying AI in manufacturing comes with challenges. Companies
\n are cautious about sharing data, especially when cloud systems are involved. And mistakes
\n can lead to faulty parts, high costs or safety risks for workers.<\/p>\n
\u201cThere\u2019s not much margin for error,\u201d Wuest says. \u201cManufacturing is a precise discipline
\n and in the worst case if something goes wrong someone could die.\u201d\u00a0<\/p>\n
To reduce risks, Wuest and Farahani aim to utilize Agentic AI for helping with operations
\n by bringing together multiple AI agents to interact with each other and achieve a
\n goal. The Agentic AI framework can also accommodate the need for cybersecurity by
\n not exposing vulnerabilities or data. But Agentic AI will not completely takeover
\n \u2013 a human expert will remain in control.<\/p>\n
\u201cAgentic AI utilizes the strengths of General AI, but it doesn\u2019t have the full decision
\n power to execute without supervisory control of a human,\u201d Wuest says.<\/p>\n
Farahani added that human oversight is critical in a manufacturing environment since
\n experts provide judgement, accountability and final approval; things that Agentic
\n AI cannot fully provide.<\/p>\n
\u201cThis makes Agentic AI appropriate as a decision-support tool that recommends actions,
\n while the human approves and manages the operation,\u201d Farahani says.<\/p>\n
This essentially breaks complex tasks into smaller steps. One AI agent could collect
\n and clean data from a machine. Then, another could analyze that data, present the
\n results and suggest an approach. At each step, the human expert checks the results
\n before moving forward.<\/p>\n
\u201cIt\u2019s the best of both worlds with the capability of dedicated models that are computationally
\n cheaper and efficient to run because they are focused on one task,\u201d Wuest says. \u201cIt
\n removes non-value adding tasks from human subject matter experts, but it never takes
\n control.\u201d<\/p>\n
Manually collecting and cleaning a data set is time consuming but allowing an AI agent
\n to handle those steps can free up human experts to focus on higher-level decisions
\n instead of repetitive tasks.<\/p>\n
\u201cInstead of having one monolithic system, you can replace agents or expand capabilities
\n depending on the need,\u201d Wuest says. \u201cHaving the human stay in control and the combination
\n of LLMs and SLMs as hosted agents is unique in our approach.\u201d<\/p>\n
Wuest and Farahani are currently taking theory to practice by attempting to implement
\n Agentic AI on a physical system with live data.<\/p>\n
\u201cWe want to integrate perception, analysis and reasoning components into a framework,\u201d
\n Farahani says. \u201cThe next step is to demonstrate the framework across multiple assembly
\n scenarios in the USC Future Factories testbed, compare it against traditional baselines,
\n and evaluate transferability to other manufacturing systems and processes.\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"If you ask a traditional AI system, \u201cWhen will this equipment fail?\u201d the answer might be a prediction…\n","protected":false},"author":2,"featured_media":41750,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[179,7493,24,25406,1180,2074,1817,25409,25408,25410,22405,25407,1181,25404,25405,18844],"class_list":["post-41749","post","type-post","status-publish","format-standard","has-post-thumbnail","category-agentic-ai","tag-agentic-ai","tag-agentic-artificial-intelligence","tag-ai","tag-college-of-engineering-and-computing","tag-colleges","tag-human","tag-manufacturing","tag-mechanical-engineering","tag-mojtaba-farahani","tag-molinaroli-college-of-engineering-and-computing","tag-sc","tag-thor-wuest","tag-universities","tag-university-of-south-carolina","tag-uofsc","tag-usc"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/41749","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=41749"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/41749\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/41750"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=41749"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=41749"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=41749"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}