By Constant Ondo
ChatGPT is buzzing since its launch at the end of 2022. All companies and departments are wondering if they should integrate this tool into their operations, and if so, how.
Industrial companies are no exception. They are wondering, for example, whether they should consider the possibility of plugging generative AI modules into their EDM to facilitate the exploitation of their knowledge.
Constant Ondo and Simon Fuhlhaber, experts in AI applied to knowledge management and founders of PICC Software, provide some answers in this article.
What is Generative AI (GenAI)?
GenAI is a new form of artificial intelligence, designed to respond to instructions (prompts) formulated in natural human language. GenAI learns the structure of the “content” it is fed with, in order to “imitate” human creativity. Its aim is to generate new content, similar in form to that from which it has learned. In its basic model, generative AI has a creative purpose, and we’ll see below what this creativity can be used for.
Forrester unsurprisingly ranked generative AI as the number one of its top 10 emerging technologies for 2023. According to the research specialist, many use cases for generative AI are expected within the next two years, which could generate sufficient benefits for a positive ROI of this technology. Smart manufacturing is one of the application areas Forrester is considering. Let’s take a look at what is possible with the tools available today (summer 2023), for the exploitation of production data and human know-how.
What are the main tools based on Generative AI, and what are they used for?
GenAI is used to create content: text, image, and video generation are its main applications.
ChatGPT, from OpenAI, is a chatbot designed to answer series of questions. Bard is presented by Google as a creative tool designed to stimulate our imagination, make us more efficient, and bring our ideas to life.
DALL.E and Midjourney create images from descriptive texts. DeepBrain and Synthesia can generate videos.
Creative professionals are already seeing how Gen AI might help them automate part of their work. But others see much further ahead: generative AI can be used to create the millions of “models” needed to train the most complex algorithms, for applications such as image recognition, for example.
Understanding the purpose of the algorithm behind each tool is a prerequisite for imagining relevant use cases. And even if the possibilities offered by the various tools available on the market are evolving at an incredible speed, notably with the contribution of LLMs (Large Language Models), the fact remains that Generative AI algorithms have been designed to create content similar to existing content. What kind of industrial applications might this correspond to? Creating a new manufacturing process by replacing reference X325S with reference YY456W? Word already knows how to do this.
Does PICC Software use generative AI?
PICC Software is a knowledge management platform that uses artificial intelligence to solve complex problems that depend on too many parameters and that the human brain can not grasp easily.
PICC Software makes use of neural networks, deep learning, automatic natural language processing (NLP) and LLM, technological building blocks that can be found in generative AI. But it uses these building blocks in a very different way, to carry out tasks that have nothing to do with what you might expect from ChatGPT, for example.
What to expect from generative AI tools in knowledge management?
The main path currently under exploration by industrial companies is the one of integrating generative AI modules to search the company’s knowledge base. It’s hard indeed to imagine an operator on a production line asking ChatGPT to search the Internet for the causes of his machine’s breakdown. If one has to ask for advice outside the company, he might as well connect to the OEM’s help desk.
Let’s take a look at what we could expect from a generative AI module applied to a EMD, for example. We could ask it to extract information: find me all occurrences of the expression “oil leak”. And we could refine the results by specifying in our prompt the machine model, the location of the leak….
We could also try asking GenAI to help us find a solution: “I’ve got an oil leak on cylinder no. 1 of machine XX, what should I do? » In this case, and given the current state of performance of the tools we’ve tested, all Generative AI can do is summarize the various solutions listed in your knowledge base. It might go as far as ranking these solutions from most likely to least likely… but without giving you any guarantee.
To date, it is impossible to solve problems efficiently using generative AI alone. At best, generative AI merely compensates for the problems of access to knowledge presented by most EDMs. But it cannot automate the analysis of complex problems, or propose solutions with a degree of accuracy and security sufficient for use in industrial processes. Error is not accepted here; the cost of a single hour’s downtime is too great.
Today, the best generative AI can do is generate an investigation roadmap. And it will take a long conversation with the chatbot, or a long prompt including all the knowledge of your experts, to extract an actionable solution. How much time will this save your teams? What benefit does that bring for the operator who urgently needs to correct a parameter, or for production managers willing to boost productivity?
How does PICC Software generate rapid ROI on knowledge management?
PICC Software has been designed and trained to perform 3 major functions:
- identify each piece of information related to a problem and/or solution in all knowledge media
- extract and reorganize all knowledge into interconnected problems and solutions
- identify root causes and the best solution for each problem, with 100% accuracy.
PICC Software doesn’t need complex prompts to provide answers.
PICC Software is a problem-solver. It knows which of all the solutions already in use in the company I should use, and tells me why. It can back up its recommendation by integrating data from IoT sensors. He details how the solution has been used X times in the past, on which machine, in which country. He can recommend an expert (KOL) for my problem. He can tell me what are the risks associated with this solution for downstream machines/operations, and what problems linked to mine this solution can solve as well. He can also alert me when a solution doesn’t comply with my country’s regulations.
Generative AI has not been designed to understand and structure knowledge. All it can do, in year 2023 is read knowledge and summarize it.
Of course, this new form of AI will undoubtedly evolve rapidly. But if speed is the key to success, why wait several months or years before solving everyday industrial problems with AI? PICC Software is way ahead of the game and can be up and running in 1 or 2 months.
Want to see for yourself? Contact us to schedule a demo!