We assess the potentials and opportunities, but in particular the risks
Everyone's talking about ChatGPT. There's no doubt in anyone's mind that ChatGPT is a new milestone in technological development. But how can we use this innovation in HR to good effect, and can it be done entirely without reservations?
Have you often found yourself having conversations like these lately?
"Have you used ChatGPT yet?"
Me: "No, I haven't got round to it yet." Thinking: What's she talking about? I'll have to google that!
"Have you used ChatGPT yet? You really have to try it. Yesterday, I ..."
Me: "I've tried, but the system was overloaded. I couldn't log in at all." Thinking: It's time I found out what everybody's talking about.
Assuming that I'm not the only person in this boat, I found it important to scrutinise the subject in detail. And in particular to appraise it in the context of HR. So I talked to Esther Brand. She is, so to speak, the mother of our chatbot Sophie, making her our chatbot expert.
ChatGPT is a publicly accessible chatbot. Thanks to its extensive training data, it is very well placed to understand and reproduce human language. It can answer questions, write poetry or scripts, create blog posts, summarise texts and even recognise emotions. To do so, it uses the GPT-3 language processing model, which analyses prompts, breaks the text down into small parts and identifies the part of speech and function, such as whether it is a noun, verb or filler. It also recognises the relationships between the words and attempts to deduce the context in order to generate an appropriate answer.
GPT-3 has access to a complex network with around 175 billion parameters and an almost unimaginable amount of terms from freely available internet sources such as Wikipedia or social media platforms. In the Pro version, ChatGPT already works with GPT-4. GPT-4 is said to have been trained with a trillion parameters and therefore offers another enormous improvement over GPT-3.
It's because ChatGPT marks significant progress in the development of chatbots. In contrast to earlier chatbots, which were mostly based on rigid rules or simple patterns, ChatGPT is capable of holding human conversations because of its GPT-3 language processing model. It can deduce the meaning and context of sentences and phrases and generate appropriate answers without having to be specially programmed. In addition, ChatGPT is freely available. Up to now, a lot of people have only been familiar with the simple chatbots from online customer service.
The company behind ChatGPT is the non-profit organisation OpenAI. It was founded in Silicon Valley in 2015. One co-founder was Elon Musk. However, he dropped out again in 2018 for reasons of conflict with Tesla. OpenAI carries out research in the area of artificial intelligence. It is famous for its GPT (Generative Pre-trained Transformer) series language processing models and the DALL-E image generating tool. At the same time, OpenAI is deeply engaged in research and development into the areas of AI ethics and AI security to ensure that AI systems are used securely and responsibly.
OpenAI receives billions in regular financial support from its partner Microsoft. The majority of its experiments has been running on Microsoft's Azure cloud platform since 2016. Microsoft profits from this partnership and is going to incorporate ChatGPT into its Bing search engine and other products.
There are three options, or rather scenarios:
Using ChatGPT via the OpenAI chat platform in the form currently available.
Integrating the functionality of ChatGPT into our own HR tools with the OpenAI API.
Waiting until major HR software manufacturers such as UKG or SAP SuccessFactors integrate ChatGPT's artificial intelligence in the background.
ChatGPT has huge knowledge and can give you guidance, tips and advice. However, you need to be careful in two areas.
First: be careful about the information you feed to ChatGPT when asking a question. The chatbot will re-use it for answers. For example, at the beginning of this year, Amazon warned its employees not to share confidential information with ChatGPT after cases arose in which its answers came "very close" to unpublished internal information. So it is not a good idea to feed ChatGPT with CVs, for example.
Second, you should be careful with regard to legal matters. The answers are only as good as the data behind them and the source is not stated. It makes sense here to have the chatbot's answer validated by a lawyer.
Let's take a look at a practical example. We ask ChatGPT and the HR Cosmos knowledge platform – both free of charge – the same question and compare the answers.
This question was answered in German on HR Cosmos and has been translated for this blog
The devil is in the detail. Although ChatGPT's answer appears correct at first glance, not all its statements are correct. For example, it does not mention the limit at which a tax levy starts, i.e. CHF 148,200 of annual income. What's more, the statement that the tax was introduced in 1999 is incorrect. The tax was first levied back in 1996. The deciding factor is the threshold of CHF 2.5 billion, above which contributions for the year will be due or otherwise. ChatGPT didn't mention this either.
Yes, ChatGPT offers the facility to link in-house systems by API (application programming interface). There is a charge for that. When you pay, the input via API is not used to train ChatGPT. It can be used, for example, to automatically generate texts for letters, agreements or job advertisements. The API can also be used for text analysis, to extract mood, topics or keywords.
You also have to bear in mind that the API currently processes and stores all its user data in the USA. This can soon lead to a conflict with the General Data Protection Regulation (GDPR). To be on the safe side, you should use products with data processing servers in Switzerland or the EU.
Of course, companies are interested in combining the latest technologies with their own products. ChatGPT means a milestone in language processing. However, we don't yet know how existing tool providers will integrate this potential into their solutions.
No, that will never happen entirely. A lot of HR tasks need human intuition, experience and judgement. A chatbot cannot and should not replace these human skills. For example, in developing Sophie, the digital HR assistant, we are trying to use new technologies to offer employees support that's proactive, fast and easy to understand. Decisions and control functions will still be in the hands of human beings.
Here's one example of potential ways to deploy new technologies usefully, taken from the onboarding/offboarding process:
When new employees join, the digital HR assistant or the chatbot can provide information and documents to welcome them and provide orientation. At the same time, HR staff can concentrate on providing the new hires with personal support and answering their individual questions. If someone leaves, the digital HR assistant or the chatbot can help by generating exit forms or performing other administrative tasks. By now, they are also capable of analysing exit surveys. HR staff then have more time for in-depth personal interviews and can focus on using the feedback for optimisation.
Thus, it makes sense to look at individual issues one by one. In the article "Artificial Intelligence in HR" in HR Today, we examined this in more detail.
Chat GPT: Amazon warnt seine Mitarbeiter neues KI-Tool zu nutzen - Business Insider (01.03.2023)
Partnerschaft mit Microsoft (01.03.2023)
Data usage policies - OpenAI API (15.03.2023)
Published: 11. April 2023