Chat GPT and Generative AI in the workplace: opportunities and threats
After some very interesting conversations with clients on this issue (and listening to a riveting talk by an expert last week) I have concluded that Chat GPT will prove to be a very valuable tool for workplace productivity and to further business aims – but only if employees can be supported to become educated users.
Chat GPT has been a sudden and runaway success as far as its private use is concerned and it is too familiar to need any detailed explanation of what it is and how it works. But two less prominent features are worth mentioning:
- Firstly, the initial training of the Bot has been done by US-based engineers based on a vast corpus of open access data, largely in English and predominantly with US-based web domains (and it does not distinguish between copyrighted and uncopyrighted material).
- Secondly, it continues to learn by inter-reacting with users, so any inputs from users could well end up as data accessible to other users.
That means that if workers are allowed to use Chat GPT to help with their workplace tasks, employers must be sure that they understand its limitations and how to interpret the results it generates. Dare I say it as an employment lawyer, a policy and some training on its use would be a good place to start.
One obvious limitation is that its answers are based on free data, and as far as I’m aware (I don’t claim to be an expert) it doesn’t have access to specialist data available only to paying subscribers. In my field – apart from the obvious fact that it can make things up – that is a major drawback. We enjoyed reading about the Litigant in Person who presented case law authorities that had been provided by Chat GPT to a Judge, only for the Judge to point out that one of the cases was entirely fictitious and the quotes from the other cases were completely irrelevant (even though the cases themselves were relevant). As lawyers we are quite comfortable using a search engine as a way of finding news and recent policy developments but wouldn’t dream of using it to find the current version of a piece of legislation or to source specialised legal commentary.
That said, its uses are wider than returning natural language-based results in different formats, which are likely, at least currently, to draw predominantly on US-based sources (and only sources in existence pre-2021). I’ve seen it draft a passable employment contract; provide a plan for a course of cognitive behavioural therapy; and write a pitch document for Mills & Reeve. Whilst some of the results in these instances were pretty good, it clearly needs an experienced specialist to review the content carefully and there will need to be proper supervision.
I was impressed by the results when Chat GPT was used as a prompt or to create a checklist against which work can be checked. It will, I am sure, become a standard tool, but we will all need to learn how to get the best out of it.
However, other potential uses can be inferred from what we know about the data it “crawled” as part of its training. For example, researchers based at the University of Norway have reported this includes a “surprising” amount of patented data.
That leads us to an even more important drawback: asking it a specific question to solve a work-based problem could well result in giving away the company’s confidential information, which it will then use for anyone else. Even the fact your company is recorded as having asked a question can be used. For that reason alone, some employers have banned its use for workplace tasks, at least until they understand more about its limitations or at the very least ban the use of the company’s name and any specific references to its business.
One alarming feature of some articles I’ve read is that – guess what – they tell me that they’ve been prepared with the help Chat GPT. This may be no more than the enthusiasm of early adopters keen to demonstrate what it can do, and thankfully, happy to give credit where it’s due. But it does open up the prospect of a Kafkaesque world where AI could subvert our attempts to understand the very risks that using it creates.
Personally, I believe that intelligent use of Chat GPT – and other similar forms of generative AI – will yield real benefits. Progressive employers will I am sure find a way to use it to benefit their businesses, or risk being left behind by others that do.
I expect many of your employees are excited by it and already using it (maybe even in work). Others will be terrified by it. But it has suddenly arrived, and it isn’t going away. So, it’s best to have a plan that’s kept under review.