ChatGPT has an incredible ability to summarize information, generate new content and answer questions. Despite this phenomenal advancement in AI, industry leaders face a long journey to digitally transform the manufacturing sector.
Manufacturers should start small by automating repetitive operational tasks, like material data management and reporting, according to Jagadish Bandla, enterprise performance CTO at global consulting firm Deloitte. This incremental deployment lets enterprises reallocate teams toward enabling successful generative AI implementations. The goal is to have generative AI capabilities reduce design and development timelines of parts and components and slash raw material use through materials discovery, Bandla said.
While manufacturers have the opportunity to creatively use ChatGPT’s advanced capabilities to their advantage, there are inevitably risks and challenges. ChatGPT and other generative AI technologies can create inaccurate or hallucinated information, i.e., when models create confident-sounding outputs not grounded in any of the original training data. In the manufacturing sector, this flaw could cause physical damage and injure people. Experts believe that the right approach could help mitigate these concerns.
How ChatGPT can assist manufacturers
One early use case is helping manufacturers perform text-based tasks.
“Much of the product manufacturing process is handled by workflows and documentation that are heavily text-based, from machine maintenance logs and service entries to status reports and alerts,” said Bret Greenstein, cloud and digital data and AI partner at professional services firm PwC.
Additionally, manufacturers use document-heavy workflows for parts orders, materials movement, workers’ shift assignments and logs. All of these elements of the business can be analyzed and generated with the help of generative AI.
ChatGPT’s capabilities go beyond rote tasks. Generative AI is applicable to the entire manufacturing value chain: market research, product concept, design, engineering and supply chain management. It could also improve customer engagement through better product configuration and recommendation engines.
Raghuram Mocherla, vice president at global engineering consulting firm Capgemini Engineering, believes top manufacturing use cases for ChatGPT include the following:
- Synthesize unstructured market needs, regulatory environment changes, supply change constraints and other inputs that go into product design.
- Record and apply problem-solving approaches used by engineers and designers during the product design and manufacturing phases.
- Build natural language engines to amplify B2B configure, price and quote functions.
- Automate product testing and validation by generating test scenarios and analyzing the results.
- Assist with purchase decisions based on cost, supply chain constraints and manufacturing capacity.
- Link product performance and field issues to flaws in product design and storage.
Risks and challenges to using ChatGPT in manufacturing
Manufacturers face many risks and challenges when letting large language models (LLMs) loose in factories.
A top concern is hallucinations, according to Beena Ammanath, executive director of the Deloitte AI Institute. Inaccurate and imagined information has no place in a manufacturing company or other business.
Another concern is that generative AI models are limited by their training data. Integrating additional data sets requires costly retraining and fine-tuning. Ammanath said these limitations could endanger core applications, such as data management and predictive maintenance. Consequences could be as severe as machine downtime and failure.
Consider also the legal risks associated with ChatGPT in manufacturing. Companies must ensure that engineers don’t inadvertently expose confidential information to these services. Samsung banned the use of ChatGPT after discovering data leakage in 2023. Manufacturers could face potential liabilities for using models built on data input without permission. Similar legal risks could arise from assigning liability when AI models directly or indirectly cause physical harm to equipment or personnel. Therefore, safety, information security and privacy risks associated with processing and analyzing personnel data must be explored when adopting ChatGPT and similar applications in manufacturing.
Best practices for adopting ChatGPT in manufacturing
Making ChatGPT’s capabilities work for the manufacturing sector requires a technical, organizational and cultural approach, according to Kamlesh Mhashilkar, global head of AI practice in the business transformation group at IT services and consulting company Tata Consultancy Services. He recommended careful use case and data planning, security design, ongoing monitoring, transparency mechanisms, user training and risk management planning.
It’s also crucial to focus on improving data as a starting point since generative AI tools rely on data training. “Success will still boil down to streamlining processes, data and the systems of record to provide reliable, diverse and statistically significant training data sets,” Mocherla said. Also, the manufacturing and business-specific language used as prompts for ChatGPT must be standardized.
Technical experts in the manufacturing space must collaborate with AI experts. “To make generative AI tools run well, organizations need people who understand the technicalities of these tools and work closely with those who understand the business processes that are being augmented with AI,” Greenstein said.
Much of this work is about getting access to the right data, or examples of past work, to train the LLM and prompting generative AI to produce accurate results. Get input from the business side and manufacturing workers in this process to assess the quality and accuracy of the data or responses.
Manufacturing will inevitably adapt to new trends
As we’ve seen with recent GPT iterations, generative AI models will get bigger. They will include more parameters, as well as multimodal and multilingual capabilities. They’re also headed for more integration with other technologies, like computer vision, IoT and robotics.
Greenstein predicted that new innovations will enable the technology to move from just responding to prompts to acting like an agent that uses prompts and responses to achieve business goals. These developments will expand the types of work that generative AI can do and scale the degree of productivity and impact it can have. If these predictions play out, ChatGPT capabilities will utterly transform the dusty old UIs currently found in manufacturing equipment, Mocherla said. More natural human-machine interfaces will capture unstructured information to be incorporated into the process without significant effort.