3D AI logo with city lights in the background

Can AI Do Safety?

There are currently a lot of debates on the use of AI in safety systems. ISO TR 5469 is I believe now almost published. However, my topic today is different and on the topic of whether AI could replace safety experts.

If you want to try this yourself ChatGPT is available at https://chat.openai.com/. You need to create an account but otherwise, it is free. There is a web and app version available to use.

The question I submitted was “write me an essay on the difference between a component, an element, and a sub-system according to IEC 61508”. I had tried it previously with a question related to SFF(safe failure fraction) and the response had errors but this time it was not bad.

Below is the output from ChatGPT

----------------------------------------------------

Title: Understanding the Distinctions between Components, Elements, and Sub-Systems in IEC 61508

Introduction

The International Electrotechnical Commission (IEC) 61508 is a globally recognized standard for functional safety, providing a framework for the development and assessment of safety-critical systems in various industries. This standard introduces several key terms and concepts, including components, elements, and sub-systems, which play critical roles in ensuring the safety and reliability of systems. To better understand their distinctions, this essay explores the definitions and significance of these terms in the context of IEC 61508.

  1. Components

In IEC 61508, a component is defined as the lowest level of hardware or software that can be individually replaced. Components are essential building blocks of a safety-related system, and they can be a single device, a piece of software, or a combination of both. Components can include sensors, actuators, controllers, communication modules, and any other discrete elements that contribute to the overall functionality of the system.

Components serve as the fundamental units that perform specific tasks within a safety-related system. They are typically designed to meet certain safety requirements, and their reliability and performance directly impact the system's overall safety integrity level (SIL). Components must be selected, integrated, and configured to ensure that the safety objectives of the system are achieved. IEC 61508 emphasizes the importance of evaluating and certifying components to ensure they meet the necessary safety standards.

  1. Elements

The term "element" in IEC 61508 is used to describe the functional part of a system that is not interchangeable. Unlike components, elements cannot be individually replaced without significantly affecting the overall system's safety and functionality. Elements encompass a collection of components and software that are designed to work together to achieve specific safety-related functions.

Elements play a crucial role in defining the architecture and functionality of the safety-related system. They provide the structure that integrates components to perform safety-critical tasks. Elements are often associated with specific safety functions, such as emergency shutdown, fault detection, or diagnostic monitoring. Their design and implementation are critical to ensuring that the system achieves the desired safety integrity level.

  1. Sub-Systems

Sub-systems in IEC 61508 are a level above elements and encompass multiple elements that work together to achieve a higher-level safety function within a safety-related system. Sub-systems represent a higher level of integration, typically responsible for a broader set of safety-critical tasks. Unlike elements, sub-systems are often designed to be partially interchangeable or replaceable, allowing for flexibility and maintainability within the system.

The concept of sub-systems facilitates the organization and management of complex safety-related systems. These sub-systems can include redundancy, diversity, and other design strategies to enhance system reliability and fault tolerance. Their correct design, integration, and validation are crucial to achieving the specified safety integrity level for the entire system.

Conclusion

In IEC 61508, the distinctions between components, elements, and sub-systems are critical for ensuring the safety and reliability of safety-critical systems. Components are the foundational building blocks, elements represent functional, non-interchangeable parts, and sub-systems integrate multiple elements to achieve higher-level safety functions. Understanding and implementing these concepts according to the standard's guidelines is essential for developing and assessing systems with the required safety integrity levels. By adhering to the definitions and significance of these terms, practitioners can contribute to safer, more reliable systems across various industries.

-----------------------------------------------------

End of output from ChatGPT

The output is not bad. I would have done it differently. If I had more time, I would have done my version and had a poll on which my readers found better but perhaps I wouldn’t have liked the results.

Just to show off I then asked it to shorten the above to one paragraph and translate it into Welsh. If you want to see the output, then try it yourself. Since I don’t speak Welsh I couldn’t really judge other than it looked like Welsh and the shortened English version was reasonable.

As a matter of interest, I asked the Will Robots Take My Job website about the future of functional safety experts. The closest available job description I could see was “Occupational Health and Safety Specialist”. Not quite functional safety but close enough. The result was somewhat reassuring with only a 7% probability that AI would take the roles belonging to the human experts but given the ChatGPT output above I wish I was that confident. The power of AI is growing continuously and while safety guys have come a long way since our cavemen days, I don’t think we are evolving at near the rate of AI.

Figure 1 - output from will robots take my job for a safety professional

For the full set of blogs in this series – see here

Parents
  • I'm not surprised that the output was good. GPT suffers from hallucination, but less so when given source material to train on, or a broad topic. In general, it's reliable and impressively good at summarizing existing information. As such, I don't think this is a good test to answer your question. What this proves seems to be something like, "Can AI summarize the main points of an existing document?" which we know is answered, "Generally, yes."

    This is a good start toward answering your question of whether AI can do safety. But actually doing functional safety is hardly about merely summarizing a document; in general, most cognitive jobs have value because they apply knowledge to novel situations. I don't have the background to test ChatGPT on applied functional safety, but I would think that a necessary next step in this investigation would be to prompt it to apply the IEC 61508 standard to a novel situation and produce its own analysis. You'd also want to define your metrics for "success" prior to its prompt generation, because of the fuzzy nature of what might constitute a "good" answer. Also, I would expect that this is where it may start hallucinating more. My prediction is that the answer is "no", or more precisely, "not yet" — I agree with you that AI is evolving more quickly than us, and it's only a matter of time before the answer is "yes." But at least for now, I think further exploration will cause it to fail quite spectacularly.

    For reference, my conclusions are based on a similar analysis applying ChatGPT to applications in engineering. It did well initially, and I found it surprisingly good; however, as I narrowed the scope to specific parts or particular problems with a single solution where a "successful" answer was more narrowly defined, it failed to be reliable.

Comment
  • I'm not surprised that the output was good. GPT suffers from hallucination, but less so when given source material to train on, or a broad topic. In general, it's reliable and impressively good at summarizing existing information. As such, I don't think this is a good test to answer your question. What this proves seems to be something like, "Can AI summarize the main points of an existing document?" which we know is answered, "Generally, yes."

    This is a good start toward answering your question of whether AI can do safety. But actually doing functional safety is hardly about merely summarizing a document; in general, most cognitive jobs have value because they apply knowledge to novel situations. I don't have the background to test ChatGPT on applied functional safety, but I would think that a necessary next step in this investigation would be to prompt it to apply the IEC 61508 standard to a novel situation and produce its own analysis. You'd also want to define your metrics for "success" prior to its prompt generation, because of the fuzzy nature of what might constitute a "good" answer. Also, I would expect that this is where it may start hallucinating more. My prediction is that the answer is "no", or more precisely, "not yet" — I agree with you that AI is evolving more quickly than us, and it's only a matter of time before the answer is "yes." But at least for now, I think further exploration will cause it to fail quite spectacularly.

    For reference, my conclusions are based on a similar analysis applying ChatGPT to applications in engineering. It did well initially, and I found it surprisingly good; however, as I narrowed the scope to specific parts or particular problems with a single solution where a "successful" answer was more narrowly defined, it failed to be reliable.

Children
No Data