What Do We Actually Mean By Human-Centred AI?

Human-centred AI has become one of those phrases that appears almost everywhere. It features in strategy documents, conference presentations and product announcements, usually with the assumption that we all understand what it means. Organisations talk about designing human-centred AI, building human-centred AI and putting people at the centre of AI, but considerably less time seems to be spent defining what that actually looks like.

When I spoke to Natalie Calvert from HuddleCX about the human side of AI, she recalled a panel discussion where three senior leaders were asked a seemingly simple question: what does human-centred AI actually mean?

One Phrase, Three Different Ideas

One speaker saw human-centred AI as making AI sound and feel more human, while another focused on helping employees perform better by giving them better tools. The third viewed it as allowing AI to take care of routine work so that people could concentrate on more complex, higher-value tasks.

None of those definitions is necessarily wrong, and that’s partly what makes the conversation so interesting. Each represents a perfectly reasonable ambition for AI, but each could also lead an organisation in a very different direction.

• If human-centred AI means creating technology that feels more human, the focus might be on conversational design, natural voices and making automated interactions feel less robotic.
• If it means improving the employee experience, investment might go towards AI assistants, knowledge tools and systems that help people do their jobs more effectively.
• If the aim is to free people from routine work, automation and job redesign could become the priority.

Three organisations could therefore describe themselves as pursuing human-centred AI while building very different things.

The Problem Isn’t That Anyone Is Wrong

The interesting thing is that all three definitions make sense, but they could lead to very different decisions about how AI is used. A leadership team could agree that it wants AI to be human-centred, while one person is thinking about improving the customer experience, another is focused on making employees’ jobs easier and someone else sees automation as the priority.

They all think they’ve agreed on the direction, but they may actually be trying to achieve very different things. That’s why defining the phrase matters, particularly when it starts influencing which technology organisations invest in, which problems they try to solve and how they eventually decide whether their AI strategy is working.

Human-Centred For Who?

Natalie suggested thinking about human-centred AI through three groups: customers, employees and the organisation itself. The difficulty is that something which works brilliantly for one of those groups might make life harder for another, so putting “humans” at the centre isn’t quite as simple as it sounds.

Customer service is a good example; AI might make it quicker and easier for customers to resolve simple problems, which sounds like a better experience, but that could leave employees dealing with a much greater proportion of the difficult or emotional conversations. The same AI could therefore improve the experience for the customer while making the employee’s job considerably harder.

It can work the other way too; AI might make life easier for employees without customers noticing much difference, while automation could save an organisation money without necessarily improving the experience for either group. Rather than asking which one should come first, perhaps human-centred AI needs to consider all three.

When Human Needs Conflict

That doesn’t mean every AI project has to work equally well for everyone, because there will always be trade-offs. It does mean organisations need to understand what those trade-offs are rather than assuming that calling something human-centred automatically makes it so.

• If AI takes over more simple customer interactions, what does that do to the jobs of the people handling everything that remains?
• If a new tool saves employees hours of work, does that translate into anything better for the customer?
• If an organisation saves money through automation, where does that benefit actually show up?

Those feel like much more useful questions than simply asking whether an AI project is human-centred.

A Phrase Worth Defining

Human-centred AI sounds like something everyone can agree on, which may be exactly why the phrase has become so popular. The problem is that if three people can use the same words while imagining three very different outcomes, agreeing to be human-centred doesn’t necessarily mean you’ve agreed on very much at all.

Perhaps the question organisations need to answer first is simpler: who are we trying to make things better for, and what does better actually mean?