If you’re interested in exploring how ChatGPT can assist you in your professional endeavors, it’s important to be mindful of potential disclosure of company information during experimentation, warns associate professor Rob Nicholls, UNSW Business School.
With the growing interest in artificial intelligence (AI), many people are now using ChatGPT for various purposes, whether it’s out of curiosity or as part of their unique selling proposition in business.
In a remarkably short span of 60 days, ChatGPT gained an impressive 100 million users after its launch. Notably, Microsoft has invested a substantial $US10 billion in OpenAI, the start-up behind ChatGPT, and has integrated GPT4 into Bing.
One of the key aspects of the user experience with OpenAI’s ChatGPT is the ability of AI tools to generate highly useful text from specific prompts, offering potential time-saving opportunities in various work-related tasks, such as writing emails, through machine learning capabilities.
You can even use the job description provided in an advertisement to get ChatGPT to write the “perfect” cover letter. Or the staff member from People and Culture can generate the “perfect” job advertisement for LinkedIn from the scratchy brief provided by the business unit.
Trying out ways ChatGPT can be used comes with some risks at work
As a starting point, let’s look at the example of the job advert.
If the role is one which is commonly advertised, then little is lost by sharing the form of the job description with a couple of hundred million other uses of ChatGPT.
But if the job description includes information which could be used by a competitor to identify your business, then the risks are significantly higher (especially if recruitment is an important part of insider business strategy in your workplace).
With companies like Samsung having recently been stung by staff members inadvertently giving away material via ChatGPT, it is important to consider the risks carefully before using it at work.
And just like that – the code was gone
This issue is particularly challenging if staff members have started using ChatGPT as part of the code development process.
It’s really appealing to do so. One of the great aspects of ChatGPT is its automation use case in reducing coding time in software development projects for programmers.
This can be done at a design level “How do I sort job application letters using Natural Language Processing in Python?” or at the code level “How do I use Gaussian Naive Bayes in scikit-learn?”.
(One really useful feature is the ability to paste code in as a prompt to ask ChatGPT to improve it. This is as simple as “what is wrong with the following code?”. ChatGPT can even recognise the coding language that you’re using!)
The ChatGPT Problem
The problem is that ChatGPT can then include the material that you have used as a prompt to improve its answers in the future, “training” the algorithms. This is precisely what happened to Samsung developers who used ChatGPT to both improve code and keep meeting notes.
Material that would have been regarded as some of the most sensitive by Samsung was available to developers outside of Samsung, simply because Samsung engineers used ChatGPT to decrease their development time.
Will GPT4 make using ChatGPT at work riskier?
The latest in the GPT series after GPT3, GPT4 has an incredibly accurate voice to text feature. But the risk of the text becoming part of the training set to improve the generative AI is high.
Using ChatGPT to transcribe work meetings is a convenient approach that can even allow for generating meeting minutes before the meeting concludes. However, it’s important to note that the transcription may become part of the GPT4 ecosystem by the end of the meeting.
Recent news headlines have also highlighted Italy’s decision to “ban” ChatGPT due to privacy concerns when it became the first Western country to block AI chatbot
Essentially, the argument made by Italian authorities is that the data collected by ChatGPT was in breach of the European General Data Protection Regulation.
However, and consistent with other European countries, it seems likely that Italy will walk back from this approach by the end of April. The only change required will be to have an age verification (over 18) check on users.
Generative AI uses billions of data points in order to be able to create text on a predictive basis. It improves in response to user feedback. The challenge faced by businesses that employ curious people is that this feedback may include company confidential material.
The solution is simple in theory, but much harder in practice. If material would not normally be disclosed outside of the business, it should not be used as a prompt for ChatGPT or for Bing.
The practical difficulty with this rule is that search engines, including Google with its generative AI called Bard, are an essential business tool. The issue may be to decide whether search engines are there to provide information, or to provide answers.
So, I should avoid using ChatGPT at work?
Not sure what you should (and shouldn’t) share with our friend, ChatGPT? Try this simple test:
Is the output of the ChatGPT session a document that would normally be regarded as confidential by your business? Then it should not be shared on ChatGPT.
If you did write your cover letter or resumes using ChatGPT, the AI system used to filter applicants could also run your cover letter through GPTZero.
This online tool from Open AI can detect whether text was written by a generative AI by examining that text’s “perplexity” (a measurement of the randomness of the text) and “burstiness” (a measurement of the variation in perplexity).
As the text output of ChatGPT continues to improve, it poses a formidable challenge to existing tools. Who can predict with certainty how the landscape of AI technology will evolve in the future?
The rapid pace of development leaves us on the edge of our seats, bracing for the unknown, and eagerly anticipating the transformation that awaits.
