The release of ChatGPT and generative AI has democratized access to artificial intelligence (AI). “Maybe everyone’s not using it, but they’re talking about it,” said Cesar Bravo, AI solutions director, Honeywell. “AI is not only a solution that emulates the way we think; it emulates the entire human system,” he noted during his presentation at the 2024 Global Honeywell Users Group in Dallas.
Bravo categorized AI as either narrow AI, meant for specific purposes, such as solving a problem limited in scope, or general AI, which is more of an aspirational class of technologies. “It’s how we create models that understand,” he explained. “ChatGPT aims to be general AI. Large language models are meant to be of broad scope, but they are still under development.”
AI encompasses specific technologies, including machine learning, computer vision, natural language processing and generative AI. Machine learning fits data to a model, and that model can learn from data. Computer vision entails analysis of the visible spectrum. Natural language processing is how we communicate, speech, being translated to text. And generative AI learns from data, but it also generates new content based on context.
Starting with neural nets
AI has been around since 1956 when the first neural network, Perceptron, was introduced. Over the following 50 years, we’d seen the development of convolutional neural networks, recurrent neural networks, fuzzy logic and long short-term memory networks, but the real power of the technology couldn’t be harnessed because we didn’t have access to the necessary data, Bravo noted.
“Around 2008, two important things happened,” he said. “The iPhone was released. It can receive and access data from anywhere and generate data. We can share data. The second event, around 2010, was the evolution of cloud technology. The computing power doesn’t need to be in the device. That computing power can be in the cloud.”
The amount of data we’re generating has unleashed the power of AI. And machine-learning applications are on the rise. DeepMind was founded in 2008 and acquired by Google in 2013. OpenAI was founded in 2015. In 2017, the principles of generative AI were defined. “You need to pay attention to the context,” explained Bravo. “The system is learning. It was an inflection point when AI could communicate with a user in natural language.” Now we have multimodal ways to interact with AI, including text, images or even speaking.
Field, operations and the enterprise
The impact of generative AI can be seen in value creation from optimized processes, the augmentation of workers’ capabilities, work process automation and increased productivity. At Honeywell specifically, AI can bring value, Bravo said. “If you look into our portfolio of solutions for control, you can use machine learning and digital twins,” he explained. “For detection, you can do prediction. And there’s faster time to expertise. With the workforce retiring, generative AI can provide guidance to new workers and help to train them.”
Honeywell sees AI as an enabler. “We apply a concept called human-autonomy teaming,” explained Bravo. “These solutions can interact with the human and collaborate to team up. We see AI as a technology. At Honeywell, we are using AI to give guidance. We don’t sell AI. We sell solutions to overcome your business challenges.”
Honeywell has categorized its solutions in three levels/pillars for enterprise, operations and the field. The enterprise AI solution is Honeywell Forge Production Intelligence, which is built in partnership with Microsoft. “We are using generative AI to interact with the user,” explained Bravo. Experion Operations Assistant is the operations AI solution. “We want to empower the operator to get the information, and we want to guide the operator with electronic work instructions, watch list, predicted events, advice and a conversational user interface,” he said. The field AI solution is Experion Process Knowledge System Field Device Manager. It enables AI-powered field maintenance and AI-assisted field operations. “It embeds all information to do the job, so the field operator can access information with a handheld device.”
