The term artificial intelligence (AI) has been around for over 60 years, but it has really only become commonplace outside the world of scientific research in the last 30.
The Oxford English Dictionary defines AI as “the capacity of computers or machines to exhibit or simulate intelligent behavior”.
These machines (or algorithms) are usually trained on data sets to solve specific problems – for instance, customizing social media feeds to the user, detecting cancer cells or to help financial institutions detect fraudulent transactions.
But a new type of AI is being developed – a foundational model.
This is an advanced natural language chatbot tool, which is trained on huge quantities of unlabeled data, and therefore, rather than being developed for a specific purpose, can be easily applied to a range of tasks.
One foundational model that has risen to fame since its launch in November 2022 is ChatGPT.
Developed by Elon Musk’s OpenAI, ChatGPT is built on the foundation model GPT-3, which relies on natural language processing (NLP) and is made up of 175 billion parameters.
While it is not the only foundational AI model out there (Google, Meta, Microsoft are all working on similar technology), ChatGPT stands out because it has opened up the tech to public use (and therefore garnered significant interest), Gartner HR’s senior director Eser Rizaoglu tells UNLEASH.
Since ChatGPT is currently free for public use, individuals have flocked to it and started to interact with the algorithm.
Its responses to all queries are human-like and conversational; through reinforcement learning, ChatGPT uses previous questions and answers to determine its next responses. In many ways, it operates as a more sophisticated, interactive search engine.
The main way it differs from traditional AI, according to Rizaoglu, is that it can “in a self-supervised manner, adapt to solving your problems”.
Ultimately, it uses NLP to “master the extremely skill of social interaction”, adds Tech Mahindra’s global chief people officer Harshvendra Soin.
How can HR teams use ChatGPT?
While many people are using ChatGPT for fun – like getting it to write Shakespearian-style sonnets about pirates or other wacky ideas – others are experimenting with how to use it in their working lives.
One obvious use case for ChatGPT is in the world of content and marketing. It can quickly produce text, which can be used as first drafts for articles, blogs, newsletters or social media posts. This can speed up the content creation process, as usually figuring out where to start is the hardest bit.
It can also be used like a chatbot by businesses to provide clear information about their products to potential customers. But does ChatGPT have a role to play in non-sales or marketing functions, like HR?
Let’s start by asking the chatbot itself. UNLEASH asked ChatGPT, ‘How could HR use ChatGPT?’.
The AI replied:
There’s the theory (and straight from the horse’s mouth). But what is the reality? UNLEASH sat down with a range of experts to discuss the pros and cons of HR teams using this advanced chatbot.
The positives of ChatGPT for HR
The recruitment use case was the primary example mentioned by the experts of how HR is (and could) be leveraging ChatGPT.
St Christopher’s people and inclusion partner Idris Arshad has been personally using ChatGPT in recruitment, including to write job descriptions, create a standard operating produce for recruitment and write generic content about careers on company websites or social media.
— Idris Arshad (@IdrisArshadHR) December 13, 2022
Other recruitment use cases are drafting interview questions for particular roles, creating an email template for candidate communication or writing an employment contract, according to Philipp Petrescu, co-founder and managing director at MVP Match.
Johannes Sundlo, people management consultant at The Talent Company, shares that he has used ChatGPT in a real-life interview scenario.
Sundlo tells UNLEASH that when he was conducting an interview for a leadership role, he shared with ChatGPT the website links to a job advert and a leadership statement and asked the AI to create scientific interview questions, as well as what the interview should look out for to ensure interview preparation was unbiased.
He then inputted the candidates answers to the question to see ChatGPT’s recommendation on whether he should make the hire. This helped Sundlo make his decision on whether or not to hire the leader in question.
As ChatGPT itself states, the technology cannot only be used in HR’s communication with job seekers and candidates, but also with an organizations’ employees. Examples include designing training plans, generating frequently asked questions, or better communication (during onboarding, as well as beyond).
Sundlo shares an example here. “I need training for new managers who felt uncertain about giving feedback in our upcoming performance review” – so he asked ChatGPT what the training should look like (i.e. what essentials must it cover, what should the structure be).
The AI provided a skeleton that Sundlo converted into a presentation.
Gemma Dale, co-founder of The Work Consultancy and lecturer at Liverpool John Moores University, tells UNLEASH she has also used it for job descriptions, to design training materials, as well as write emails or scripts for meetings or presentations.
You might have noticed that these uses cases are largely about speeding up researching, writing and content creation – so HR is essentially leveraging ChatGPT in a similar way to marketing.
“ChatGPT tools will help with…everything that we write today, every bit of content that we create today”, notes Jason Averbook, CEO and co-founder of Leapgen. It can kickstart communications, research and ideation – “stuff that would take hours, and months, to research on the web can now be done within minutes”.
Ultimately this is positive for HR as it automates their repetitive, time-consuming work, thereby “giving them more bandwidth to focus on the human-facing parts of the job”, adds Petrescu.
Dale agrees that ChatGPT is a time saving tool – although it doesn’t generate a final product – HR would often have to add detail and change the tone – but overall it takes “considerably less time than completing the entire task” manually.
Sundlo is a big fan; “It’s like talking to an incredibly knowledgeable HR friend [or] having a HR assistant on steroids, providing a level of efficiency and accuracy that would be difficult for a human to match.”
One question, however, is whether HR needs to use an advanced chatbot to produce this type of content.
Gartner HR’s Rizaoglu notes: “A lot of applicant tracking systems or employee communications tools already have present templates, which [HR teams] can work off and [can] standardize work with your marketing team or legal and compliance folks”.
The current downsides of ChatGPT
A major challenge of ChatGPT and foundational models in general is that the information they provide may not always be correct.
The major issue is the data they are trained isn’t always reliable – the AI can’t necessarily tell the difference between reliable sources and opinions being shared on social media. This can cause biases to creep in.
Petrescu explains that because ChatGPT is currently “block boxed”, it doesn’t explain why it wrote what it wrote (or where it got its information from). This is a major flaw, and could negate the time savings for HR teams, as they would need to double check a lot of the information.
Arshad adds: “I find it to be an easy way out that is not a way out as it doesn’t provide the solutions you need.” He warns it can make you lazy “in thinking you are giving a good quick answer”.
Averbook agrees that HR teams mustn’t depend too much on ChatGPT; instead, they should experiment, but remember to use their brains.
Another issue is that ChatGPT and GPT3 currently only leverages data from before 2021. “That means, right now, it can’t give accurate and updated output [for instance in areas like] laws that might affect an employment contract or the latest tech stacks”, notes Petrescu.
Ultimately, Averbook warns that HR needs to think about hands, heads and heart. “ChatGPT is helping so much with the hands work. It is starting to help with the heads work, but then people take over, and find the heads work and add the hearts work. That’s the power of where we’re at today with ChatGPT”.
There is a need to be careful between distinguishing what ChatGPT is able to do right now and what the potential future applications are.
ChatGPT told UNLEASH it could answer questions about benefits and company policies, or it could analyze employee sentiment and help reduce attrition.
While it can answer questions about pay gaps, benefits or employee happiness, it can only leverage publicly available data.
Taking DHL as an example because it was ranked by Fortune as the best place to work in 2021, UNLEASH asked ChatGPT if DHL workers were happy. It replied:
We then tried to ask it about UNLEASH, which has less publicly available data about it as a workplace than bigger brands, and the response was:
For employees or HR teams to use ChatGPT to get quick answers to these types of questions, organizations would have had to input significant amounts of its own data, much of which is highly sensitive and therefore would breach privacy laws.
“Alarm bells are going off thinking about it”, shares Gartner’s Rizaoglu. It’s a huge risk and a red flag.
Dale adds that HR teams need to “think about the potential misuse or inappropriate use in their own particular context and educate their employees. In particular, they may wish to issue guidance about GDPR and why employees should never upload personal data”.
A message for HR vendors
Due to this major drawback around the dataset, the only solution is to bring ChatGPT or a foundational model in-house.
According to Rizaoglu, “for 99% of companies that is near impossible” – because of the cost, the complexity and the computer resource (and skills) needed to bring it into your organization.
Christa Reed, head of job market research at Jobsearcher, shares her organization’s personal experience: “The software is a bit difficult to set up and use, particularly for those without a technical background. We had difficulty customizing and integrating into our existing HR system.” She adds it was pricey to bring in-house (and to maintain).
As a result, Gartner’s Rizaoglu believes that ChatGPT and foundational models in general will become useful in the field of HR indirectly “as plugins to other technology”. Vendors can build parameters around the model in order to make it safe.
Therefore, he calls on HR teams to talk to their current vendors about their plans for foundational models. Many them already have chatbots and NLP models, but ChatGPT is even more sophisticated, so HR needs to find out their vendors’ plans for foundational models (and how they are correcting the issues around the data and biases).
All of this is not to say that HR teams should not start experimenting (and having fun) with ChatGPT and foundation models in their current forms.
Dale concludes: “HR professionals need to understand ChatGPT and embrace it. These sorts of tools aren’t going anywhere – they are only going to be better and more powerful, and we cannot afford to get left behind. If we don’t get involved, the agenda will be set without us”.
Tech Mahindra’s Soin states: “The future of ChatGPT appears bright for HR, even though we are still a few years away from its full implementation”.
The International Festival of HR is back! Discover amazing speakers at UNLEASH America on 26-27 April 2023.
Sign up to the UNLEASH Newsletter
Get the Editor’s picks of the week delivered straight to your inbox!