Much of the focus on AI technology in the workplace has centred on whether it is replacing humans or not—but what happens when employees are asked to work alongside AI-based systems?

How does this type of technology change work? And how do AI systems (such as chatbots) change the experiences, and potentially even the emotions, of workers?

Our research studied the implementation of an AI chatbot in a large organisation, including interviews with 46 employees, to understand what it means for workers when their colleagues are increasingly taking the form of ‘bots’.

The importance of human interaction to AI learning

The AI chatbot we studied was initially implemented to provide IT helpdesk support to employees and was subsequently extended to include other capabilities such as translation services, holiday bookings, and sentiment analysis. Traditional business IT systems, such as accounting and warehousing systems, are relatively static; they come with a particular set of features and limited potential for users’ feedback to influence change.

In contrast, human interactions are crucial to the learning process of AI-based systems which can improve and accommodate user feedback either in short cycles of supervised learning or on a constant basis through unsupervised learning. Hence, in AI systems, the more they are used, the more usage data can be collected. Then, the more learning the AI system gets, and hence the more it can improve and the more it can better serve users and organisations.

To facilitate this human-AI interaction, some enterprise AI-based systems are embodied in ‘digital human’ avatars or an interface of animated or still images, which mimic attributes of personality. The AI chatbot we studied, for example, provides conversational interface based on natural language, seeking to communicate with users through human-like interactions. They can also be personalised and can be designed to also interact in ways that look to create feelings of togetherness and being with another human. Like other AI systems, AI chatbot employs machine learning technology so they improve with use.

The AI chatbot elicited feelings of connection and amusement

We found that while workers experienced a range of positive and negative emotions towards the AI chatbot in their workplace, they also experienced two novel categories of emotions that are rarely experienced towards traditional business systems. We identified them as connection and amusement emotions.

Connection emotions stemmed from the form and functional design of the AI chatbot. Users interpreted the functional characteristics of machine learning as human-like learning and felt responsible for educating the AI chatbot. Some of our interviewees referred to the AI chatbot as “he” and expressed that they have duties to teach it so it becomes better in the future and eases their workload. For example, one interviewee said:

“when I’m with a chatbot, I know this is a machine and the more I write, the more he gets used to it”

Another interviewee expressed tolerance; while the AI chatbot does not have all the answers it will in the future if they keep using it:

“Well, I think it’s fine. I mean, it’s, I think, just a matter of time before it gets smarter and better. I mean the more training it has, it would definitely get better.”

Workers also found the AI chatbot entertaining, providing an element of fun and gaming.  Users were trying to “challenge the bot”, “play with the bot” and “crash the bot” as users expressed it.

Users also personalised the interface of the bot and expressed a wish to personalise its name.  They saw it as a way to have fun at work as well. One interviewee eloquently gathered these views:

“I would prefer to make it personal. For example, I would give it my name not the name for everyone else. So, mine would be for example Eva or someone’s will be Matthew or something like that. [. . .]”  Exactly, as much customisation is there then it’s better. For example, one day I would like to see an orc talking to me, next one the cyberwoman from the Battlestar Galactica [TV series and science fiction on Cybercivilisation] or something like that [. . .] It would help me more for sure. It is something that is just fun and everyone from time to time has the time that he wants to spend on something not directly connected to the world. And for example, this is something nice.”

These connective and amusement emotions were also mixed with other contentment and frustration emotions. However, the study found that the experience of these multiple emotions offset or enhance each other. For example, workers’ excitement, hope and playfulness, in addition to the connection emotion of empathy, led to tolerance and the continued use of the chatbot in a forgiving way in spite of the frustration experienced when the chatbot provided wrong results. It made employees hopeful that it “will definitely get better” as many expressed it. Most interviewees were hopeful that “he is learning from the interaction” and looking forward to the future when “it will be a true colleague in the office” as one employee expressed it. It was only when users experienced solely strong negative emotions that they stopped using the chatbot.

The feelings of connection and amusement created a rapport with the AI chatbot that fuelled its continuous use. This continuous use, in turn, created the necessary data that enabled the AI chatbot to evolve in its organisational context to better suit users’ needs resulting in significant improvement in its function. This continuous improvement encouraged the organisation to expand the AI chatbot domain to come closer to be a virtual enterprise assistant: booking holidays, checking availability of other services, translating documents among others.

Organisations considering implementing AI chatbots for the internal use of their employees need to consider that it is a continuous journey of use, learning and improvement and that listening to employees’ feedback and lived experience propels this journey. Hence, adopting participant approaches to AI chatbot’s continuous improvement is essential for the success of this technology in organisations.