ChatGPT and Beyond – AI as a tool for safety and health professionals
I’m sure many of us have been fascinated by ChatGPT and have played with OpenAI’s free trial for our own entertainment, but can it also be used as a valuable business tool for professional food safety and public health professionals?
To try and answer this question, I asked ChatGPT a range of questions related to some current projects and compared its answers with traditional internet searches. To add the ‘Human Intelligence’ element, I also relied on my 45 years of experience of working in both the enforcement and commercial sectors of the food industry to provide some ‘HI’ to assess the value of the AI.
Before I summarise the results of my review of Chat GPT as a business tool, it may be useful to highlight the real reasons why Open AI launched ChatGPT.
The Real Purpose of ChatGPT
Open AI was originally launched as a non-profit organisation to drive the development of safe AI for the benefit of humanity.
The OpenAI Charter states:
“We commit to use any influence we obtain over AGI’s deployment to ensure it is used for the benefit of all, and to avoid enabling uses of AI or AGI that harm humanity or unduly concentrate power.”
The Charter was written in 2018 when Open AI was still a non-profit organisation, but in 2019 OpenAI transitioned to a ‘for profit’ business and received a $1 Billion investment from Microsoft. At this time, Sam Altman, one of the original founders of OpenAI, stated that the $1 Bn investment may be insufficient, and they may need to raise “more capital than any non-profit has ever raised”.
At the start of 2023, Microsoft invested a further $29 Billion in the now ‘for profit’ OpenAI business, and we should also remember that ChatGPT runs on the Microsoft’s Azure platform, and one of the other major partners is Amazon cloud services. So OpenAI is no longer an altruistic organisation and in addition it has a ravenous appetite for capital investment.
For the first seven years of its existence, much of OpenAI’s work went unnoticed by venture capitalists and mainstream news and social media. The pioneering ‘Open AI Gym’ was a worthy open platform, and its ‘InstructGPT’ was a leading natural language system - but they were not grabbing the attention of mass media or, more importantly, venture capital investors. The launch of ChatGPT as a sibling model to their core InstructGPT system, was a strategic move to dominate the AI conversation with an easily accessible, newsworthy and entertaining free App. This was a very successful move, as ChatGPT is now synonymous with AI in the same way Hoover became the brand name for all vacuum cleaners.
An additional benefit of providing a high-profile free App is that users will help to develop the system for free by providing feedback. As stated on the OpenAI website; “We are excited to introduce ChatGPT to get users’ feedback and learn about its strengths and weaknesses and …. during the research preview, usage of ChatGPT is free.”
So, when evaluating ChatGPT as a work tool, we should be aware that its main objective is as a publicity tool for a company that is hungry for both financial investment and human feedback.
Considerations when using ChatGPT as a business tool
1. Confidentiality
When you enter a question into ChatGPT you are sending potentially confidential information to a server that is outside of your company. Many companies prohibit the use of externally hosted systems such as Dropbox and even Google Translate under their IT security policies, and in April Samsung banned the use of ChatGPT by its employees due to security concerns. So if you’re working for major branded food business, think twice before asking ChatGPT about “how to eradicate listeria from our manufacturing plants”.
2. Accuracy
Remembering that ChatGPT is in effect a publicity stunt, so it’s not surprising that it prioritises natural language over technical accuracy. To give one personal example - a while ago I wrote a (spectacularly unsuccessful) fiction novel, and I asked Chat GPT if it was worth reading. After the standard ChatGPT disclaimer, I was impressed that it described my novel accurately, but then it said the book had ‘mixed reviews’ so I should make up my own mind about reading it. This sounded reasonable, apart from the fact that the book had such limited sales there are only 6 reviews in existence, all on Amazon, and all 5 star and very positive (probably due to most the reviews being written by friends and relatives). So in an effort to create a natural sounding response, ChatGPT told a lie.
This may seem like a minor issue, but if we are using it as a critical business tool, can we trust it? There’s word in colloquial English for a person who ignores the facts in an effort to sound clever, and it involves bovine excrement – so is the tool we should be using to consult on critical business issues?
3. Absence of references
As health and safety professionals we’re often asked for guidance on a technical issue that goes beyond our direct knowledge. We’re all partial to some Googling to obtain additional information before we provide guidance to our businesses – and we’ll always look at the origin of the documents in the search results to decide if they are credible.
With ChatGPT you get highly credible sounding advice, but in effect it is just acting as a sophisticated front end for Microsoft’s Bing search engine, but we cannot see any of the documents used to generate the answer. This means that we cannot evaluate the reliability of the sources that generated the advice, and if challenged on our recommendations we would be unable to provide any references.
You can use ChatGPT to summarise a provided list of technical publications, but then you are relying on its programming to decide what parts of the reports are relevant to your business, and it is very unlikely that ChatGPT would be a better of judge of what is important to your business than you are.
4. Sustainability
Your company will have goals for environmental sustainability, probably with a date for achieving net zero impact. ChatGPT’s servers consume significant amounts of power as well the large volume of water needed for cooling. You would therefore need to factor this into your sustainability policy and carbon accounting if you use it as a core business tool.
We should also consider if using what a free publicity stunt is a sustainable business solution – as it’s likely that it will not remain free in the longer term, and it may even be withdrawn once it has served its primary purpose.
5. Potential Copyright Infringement
ChatGPT and its’ image generating sister programme DALL.E, have been taught using multiple data sources from the internet, and they also utilise data from Bing searches. Without seeing the original source references, it would be impossible to spot any possible copyright infringements. The EU has recognised this issue, and the new EU AI Act specifically requires the AI businesses to declare any use of copyrighted material. This is not a serious consideration when using ChatGPT for personal use, but it may be an issue if you are suing it for official business purposes.
6. Some Examples
I was recently asked a question by a restaurant business:
“Should food handlers wear a mask when wearing open food, and should serving staff wear a mask to prevent issues from viruses such as COVID-19?”
Answer from ChatGPT about food handlers:
Yes, it is a good practice for food handlers to wear masks while handling open food. This can help prevent the spread of germs or viruses that may be present in the food handler's respiratory system. Wearing a mask can also prevent hair or saliva from getting into the food and can help reduce the risk of contamination. It is important for food handlers to follow good hygiene practices to ensure the safety and quality of the food.
Answer from ChatGPT about wearing a mask help to protect from COVID-19:
Yes, wearing a mask is one of the proven ways to help reduce the spread of COVID-19. The mask acts as a barrier that prevents the respiratory droplets from getting into the air when you talk, cough, or sneeze. Wearing a mask also helps prevent you from contracting the virus if someone around you is infected. However, it's important to note that wearing a mask alone isn't enough to prevent the spread of COVID-19; it should be combined with other preventive measures.
How to convince food handlers to wear a mask while they handle open food?
Here are some potential steps you could take to convince food handlers to wear masks while handling open food: 1. Educate them on the importance of wearing a mask. Provide clear and concise information on how the virus spreads, and how wearing a mask can help prevent transmission. Explain that this is not only for their own safety, but for the safety of their colleagues and the customers they serve. 2. Make it a company policy. If masks are mandatory in your state/country, make sure everyone is wearing one.
What evidence is there that wearing a mask helps to protect public health?
There is strong evidence that wearing a mask can help to reduce the spread of respiratory viruses, including COVID-19. Studies have shown that the virus can spread through respiratory droplets released when an infected person talks, coughs, or sneezes. A mask can help to prevent these droplets from spreading to other people. In fact, the Centers for Disease Control and Prevention (CDC) recommends that people wear masks or face coverings in public settings, especially when social distancing measures are difficult to implement.
Using the same wording as a Google Search resulted in a wide range of search results, much of it of limited value so it needed some filtering, but I thought the following chart was a useful resource for any safety and health professional:
Effectiveness of Mask Wearing to Control Community Spread of SARS-CoV-2
The inclusion of references allowed me to verify the reliability of the data, and it was also helpful in making an informed recommendation to the restaurant business.
The weakness of the Google search compared to the succinct answer from Chat GPT was that the search results included some dubious sources of information and some irrelevant results that happen to include the word string.
Conclusion
The potential data security risks, the possibility or erroneous answers and the absence of references leads me to believe that ChatGPT is not a suitable tool for safety and health professionals. This is not surprising considering the true purpose of ChatGPT as a publicity tool. However, the impressive ability of the base programme, InstructGPT, to filter a vast amount of data and produce a coherent answer to a complex question could be of great value to professionals.
What could be an ideal tool would be a search engine driven by AI such as InstructGPT that could filter out the irrelevant search results and provide a summary that included references for the sources.
There are two current developments that may be exactly what we need. Google is combining their Bard AI into its’ search engine, and this should be available shortly. And Microsoft has just incorporated OpenAI’s InstructGPT into a special version of their Bing search engine. The screenshot below shows the results of a search for the test question – as it includes the essential references.
Bing Search
