The wave of AI continues to impact people’s eyes, especially the great changes brought by ChatGPT, and the arrival of the era of large models today, will make people convinced that the future is the era of artificial intelligence. It is also affecting the manufacturing industry, which is constantly seeking answers – what changes can AI bring to us? But here’s the problem with AI – it seems more reasonable to move on to another topic, what kind of AI industry actually needs. The former is looking for nails with a hammer, while the latter is more in line with the technological development thinking of the industry has been, the problem needs what kind of tools? It’s not a question of what the tools need.
1). Easy to use AI tools
IS420YAICS1B Industrial AI, it needs to be simple to use – after all, in order for industry to solve problems, even the development of tools, it must be combined with the mechanism of the industry itself, programming. Engineers in industry are usually not trained in AI, or even those from the AI field, and need better tools in development projects. This includes how the data is preprocessed, the friendliness of the training method in the configuration, the ease of understanding of the operating interface, and as a tool, it is difficult to popularize widely if very complex operations are required.
Moreover, the training methods and models integrated in this AI tool must be suitable for industrial characteristics. Therefore, this is now the field of industrial automation manufacturers will integrate AI into the original system.
Therefore, for the industrial field, the integration of embedded AI, as well as the data interaction interface of standards and specifications, to achieve the specific AI application task heavy in the industry itself, to achieve rapid configuration, calculation and control task integration, closed-loop iteration, and all these need to be simple and easy to be mastered by engineers.
Therefore, this is the most important topic for industry’s need for AI – as a tool property, it must be friendly to engineers.
2). Industry needs highly interpretable AI
IS420YAICS1B Because AI is more of a “black box” model, it is different from the white box of mechanics – and there is an “interpretability” problem. It is also a difficult topic for commercial AI after it enters industry.
There are many kinds of explainability in machine learning, and now the actual use is called local interpretation, which refers to the decision basis of explaining a single model reasoning. For example, a very complex neural network can infer tabular data and give the importance of the various inputs.
Interpretability, and some lie in the interpretability of data preprocessing, the interpretability of algorithms, the interpretability of results after the event, and multiple levels to achieve.