How AI could use language to shape our reality

Throughout human history, we have used language to define our experience of the world. But what happens if AI develops this capability?

Language reflects how we live and what we do. "Influencer" and "selfie" are terms most of us understand, but, go back 30 years, and nobody would know what they meant. They describe concepts that did not quite fit into any existing language that preceded them, and our adoption of them is a shared acceptance that they’re different enough and important enough to be worth learning.

These concept terms are known as categories. For most of history, the creation and dissemination of new categories has been a solely human activity. In recent years, artificial intelligence (AI) has evolved to such a point that it is now fully capable of doing the latter – teaching us about categories we, as individuals, may be unfamiliar with.

But will AI someday be capable of the former too? Will it be able to propose that there are things that we, as human beings, have not yet said or even appreciated about the world, and create new language that allows us to see reality in a different way? Our research explores this question.

Categories shape our reality 

Category work is one of our apex functions as human beings. Throughout our history, the development and sharing of knowledge – which involves inventing labels for patterns and things we experience in the world – is something that has in itself changed the world. In other words, category creation is not just reactive – it actively shapes reality.

For example, we set the Industrial Revolution in motion, we learned how to shape a new world by burning fossil fuels at scale to generate power, and now we have global warming because this has put too much greenhouse gas into the atmosphere. All of this is the process of category creation.

We may... see AI move into other things we currently consider deeply human

Innovators and creatives are constantly engaging in this process, trying to change the world – and they have already started working with AI. Right now, AI is a tool that can help them organise their ideas and plans. But it’s arguably capable of participating more actively too. 

Previous research has found AI capable of participating in the game of categories, so it’s logical to assume it will continue this journey to become an equal collaborator in future and – beyond that – an innovator in its own right. This would see AI join human innovators as a creative partner, interacting with them like a real person and performing traditionally human roles.

We are still a long way from this potential future, and it’s impossible to know now what categories or types of categories AI may produce down the line. If we could know what they are, they would already exist. But we can take a guess, based on the aforementioned trajectory of AI from tool, to participant, to collaborator, to innovator. 

Virtual therapists

In his book How To Know a Person: The Art of Seeing Others Deeply and Being Deeply Seen, David Brooks argues we as humans are not as good at truly seeing one another as we like to think. The key, according to Brooks, is to shine a positive light on others by taking a genuine interest and really listening to them. Doing this helps us to really know a person and, in the process of being known, helps the other person to better know themselves.

Most people are not very good at this. It takes attention, full engagement, a willingness to accompany someone, and the ability to engage. But what if generative AI were able to interact with people in that way? What if it could help us see ourselves in a new or better way?

Category creation is not just reactive – it actively shapes reality

Most people right now would baulk at the idea of a virtual therapist, but at the same time many institutions – universities, for example – lack sufficient therapy resources to support the scale of mental health challenges they face. With this in mind, a first-line option of an AI therapist might not be a bad idea.

And, if this were possible, we may also see AI move into other things we currently consider deeply human and utterly removed from the sphere of computing. For example, caring and nurturing roles that currently require a long time to learn, a great deal of intelligence and expensive resources, but could be done by AI at lower cost and greater scale.

This isn’t a direct prediction, but an area of work that is perhaps ripe for development. In such a future, a supportive AI may be able to alert us to problems, issues and improvements to our lives that we cannot currently even conceive of. This, in essence, will involve the creation of new categories that do not exist today, but which will quickly become part of our social landscape, heralding a strange yet exciting world to come.

This article draws on findings from "Beyond the Turing Test: Exploring the Implications of Generative AI for Category Construction" by Nelson Phillips (University of California), Sai Kalvapalle (Nova School of Business and Economics) and Mark Kennedy (Imperial College London).