Playing around with ChatGPT
I have been intrigue with AI for a bit so I started asking the ChatGPT system safety questions. I found it interesting. Here is an example.
What is the difference between Safety I and Safety II?
Safety I and Safety II are two different approaches to safety management.
Safety I is a traditional approach to safety management, also known as a "defensive" approach, that focuses on identifying and controlling hazards to prevent accidents and incidents. It is based on the assumption that safety can be achieved by identifying and eliminating hazards, and by enforcing rules and regulations to ensure compliance. This approach is often reactive and relies on past data to predict and prevent future accidents.
Safety II, on the other hand, is a more modern and proactive approach to safety management. It is based on the idea that safety is not just the absence of accidents, but also the presence of desired outcomes. Safety II focuses on creating a positive safety culture and promoting a more resilient and adaptive system, where employees at all levels are encouraged to actively identify and mitigate potential hazards. This approach is more proactive, it focuses on using real-time data and feedback from the front-line to continuously improve safety performance.
In summary, Safety I focuses on identifying and controlling hazards, while Safety II focuses on creating a positive safety culture and promoting a more resilient and adaptive system.
I asked it questions about employee engagement and safety. I also asked it to write a machine guarding policy based on OSHA regs. The results weren't bad and I think I can use it as a first draft of the policy for my company.
So what do you think? Do you think this might be a useful tool in the future?
BTW, I was wearing my SkyNet t-shirt while doing this.
Comments (5)
WOW
Yea I played with it a few weeks back. Definitely a crazy tool - not sure if it will replace us completely but definitely could be a useful tool for some of the administrative tasks
Hi Michael,
I've been playing around as well to get a sense of the AI's potential, especially in terms of gathering and summarizing information. It's taken a while to figure out how best to formulate my prompts in order to get back the right information in the right format. This is not a revolutionary insight by any means, but, just like with the average human writer, the quality of the AI's work depends very much on the quality of the 'brief' you give it. The more niche the topic, the more carefully you need to formulate the prompt - otherwise the information you get back often tends to be inaccurate or even false.
Unless you give the AI very clear, narrow parameters, especially when it comes to lesser-known topics, it will serve up whatever it finds in the few available sources as if it is the truth. For example, when starting out I asked the AI for a summary of the first-ever (known) environmental disaster that involved hazardous materials. It spit out a very convincing summary which, upon investigation, turned out to be completely inaccurate. The event it described was not the first environmental disaster ever, and the statistics it provided were nowhere near accurate. It couldn't find a clear answer to my question but, much like many humans, it opted to give me AN answer rather than no answer at all. This is obviously frustrating because it necessitates extensive fact checking. If I wanted to do that much research, I would not have bothered with AI!
So, in my latest experiments I have been telling the AI which kinds of sources to use (i.e., credible sources such as official regulating bodies, academic and research institutions, etc.) and asking it to provide links to multiple sources for the information it provides. That already makes the fact-checking a lot quicker and easier. So, in a nutshell, I'm giving instructions in the same way that I would to a human writer - but unlike with humans, giving the AI better instructions always amounts to better output!