Click Image to Enlarge

AI is already part of our everyday lives, but in the future, it is expected to fundamentally change our society. Eighty percent of companies say they are planning or have already started AI-related projects. Overall, global GDP is expected to grow by up to 7% over the next ten years thanks to AI.

AI has also been used in maintenance and in the coming year, the potential of new generic AI models will be increasingly brought to maintenance. In the future, this will also strongly change the operating models and practices of maintenance.
What is AI?

AI is a broad and multi-dimensional concept. It is also used loosely in very different contexts. In simple terms, AI refers to the ability of a machine to use skills traditionally associated with human intelligence, such as reasoning, learning, designing or creating. Even the definition of human intelligence is not unambiguous but, in some ways, a philosophical conundrum. AI is therefore not a single technology or solution, but a set of different technologies, applications, methods and research directions, depending on the point of view.

At a high level, AI is often divided into weak and strong AI.

Weak AI

Weak AI is also referred to as applied, narrow and traditional AI. Weak AI refers to systems that are designed to solve a specific task. Examples of weak AI solutions include Apple's Siri assistant and Netflix’s recommendation engine. Unlike traditional programming, where all rules must be defined individually, weak AI models are trained for a specific task using training data. The trained models can then find regularities, anomalies, relationships, etc. in the data. Most current AI solutions are based on weak AI.

Click Image to Enlarge

- Despite the undoubted benefits of AI, there are many aspects of generative AI in particular that those using it need to consider, says the author of the article, Hannu Niittymaa, Solutions Engineer, IBM Sustainability Software.

Weak AI technologies include machine learning, deep learning, neural networks, speech recognition and machine vision. In particular, in deep learning, complex model structures can lead to a loss of generalisability of the solution.

Strong artificial intelligence

Strong AI, also referred to as general, creative or human-level, aims to mimic human cognitive ability in the most general way and is capable of performing tasks that go beyond human intelligence.

Such an AI would be able to learn and understand complex concepts, apply knowledge to different situations, and demonstrate creativity and abstract thinking in a wide range of tasks. This requires a deeper understanding of the meanings of data sets. However, solutions based on strong AI are not yet in practice.

Click Image to Enlarge

Generative AI

Currently, AI is often referred to as generative AI, but this is not the same as strong AI. Generative AI refers to a subset of machine learning that focuses on creating new content, such as images, music or text. It is often based on deep learning and neural networks that are trained to create complex models and generate new content.

While generative AI can be highly advanced and capable of producing impressive content, it has not yet reached the level of strong AI, where a machine is capable of independent thought and understanding in the same way as a human.

Generative AI is suitable for illustrating solutions and creating something new. For example, it can be used to generate entirely new text, speech, images, sound or code. These can be varied based on previous implementations. Humans do this quite commonly. Creating something new is demanding for both AI and humans.

Generative AI has been made possible by new advanced AI foundation models developed by corporate research and development teams using massive computational and data resources.

Click Image to Enlarge

From a security point of view, information uploaded to the ChatGPT service is public information for everyone.

One example of an advanced AI foundation model is the Large Language Model. These are the basis for AI solutions for text analysis and research. These models can also be used to summarise and translate previously written texts.

For example, ChatGPT, developed by Open AI, is based on a large language model. A Chatbox-like interface allows you to ask questions. These questions are answered based on the information that is used by the model - in practice almost all public information on the Internet – and the answer is given in the way the model sees fit.

In addition to the broad language models, there are a large number of other advanced AI-based models for a variety of uses.

Advanced AI models have been developed by Amazon, Google, IBM, Microsoft, NVIDIA and several smaller players.

What can AI be used for?

Solutions based on weak AI are already in use in maintenance today. For example, equipment condition monitoring and solutions that predict future equipment failures are based on weak AI technologies such as machine learning. The next part of this series of articles will explore these solutions in more detail.

Typical applications of generative AI include automation of various customer service situations, automation of IT support services, automation of repetitive routines such as job application reviews, etc. The interactive use of experience by varying questions is the basis for this growing set of applications.

Despite the huge attention that generative AI has received, the use of generative AI in general is still limited and has not yet been widely exploited in maintenance. However, various maintenance system vendors are heavily developing their systems in such a way that generative AI can be used in those use cases where it makes sense. The final part of this series of articles will explore these in more detail.

Click Image to Enlarge

Concerns related to the use of AI Source: 2023 IBM IBV Generative AI global survey

It is of course an intriguing idea that AI could easily and autonomously handle all of the above. However, there is a huge amount of research, algorithms, models and technology behind the various AI models. Deep human expertise is still needed to make use of all this.

The challenges of harnessing AI

Despite the undeniable benefits of AI, there are many issues, especially with generative AI, that those using it need to consider. Some of these issues are very practical, while others may involve larger ethical issues.

An AI model is trained against a set of data, and the quality and representativeness of this set of data will determine to a large extent, the usefulness of the solution for a specific application. There are many ways to improve the quality of the data. For example, signal analysis and feature extraction can be used to build indicators that describe performance better than mere measurements. Data analysis would provide a more meaningful basis for cognitive analysis, i.e. AI proper. So how much data should and should not be manipulated?

In using AI, it is also important that the problem to be solved is of the right size. AI can be used to try to find solutions to large-scale problems, but this should be done with great caution. The result may look good, but is it reliable?
From a security point of view, it is important to note that the data exported to public intelligence services, such as ChatGPT, is public data for everyone from a security point of view.

AI models can be trained on data sets whose quality and origin are not known. This can lead to offensive, biased or outright wrong answers. Not knowing how the AI arrived at its answer also calls into question the reliability and usefulness of the answer. Verification is needed to identify biases and errors.

There may also be copyright restrictions associated with AI solutions and the data sets used, which need to be taken into account.

For these reasons, there must be governance mechanisms in place to ensure that the risks associated with AI are properly assessed. Risks may be related to the reputation of the company and regulatory or operational issues. Ultimately, it is always up to the user to evaluate the quality of the answers provided by AI.

Data for decision-making

According to a wide range of studies, companies consider the use of AI to be one of their top priorities for the future. AI provides additional data for decision-making. It can deliver significant business benefits and the payback period for investing in AI can be very short. It would therefore be good for companies to explore the potential of AI while taking into account the challenges and risks associated with AI for decision-making. Expertise and AI are mutually supportive.

In the following parts of this article series, we will discuss in more detail how AI can be used specifically in the development of maintenance.

Text: Hannu Niittymaa, Solutions Engineer, IBM Sustainability Software
PHOTOS: IBM, SHUTTERSTOCK, freepik