How AI is Changing Human Communication

In 2013, the science-fiction drama Her questioned whether communication with artificial intelligence could be indistinguishable from communication with humans. While this day is yet to come, AI is increasingly facilitating human-to-human communication. This phenomenon is AI-Mediated Communication (AI-MC), which Jeffrey T. Hancock, Mor Naaman, and Karen Levy define in their recent paper as “interpersonal communication in which an intelligent agent operates on behalf of a communicator by modifying, augmenting, or generating messages to accomplish communication goals.” Despite the growing prevalence of AI-MC, there is little discussion about its societal implications. How should we think about how AI-MC will change human communication, and what is the role of public policy? Hancock et al. lay out key issues that society will grapple with as AI takes a more prominent role in how we communicate. 

In Her, the main character, Theodore, works as a writer at a company called Beautiful Handwritten Letters. His job is to write letters that believably sound like someone else — AI systems in communication work similarly to Theodore. From auto-correct and predictive text to auto-completion and smart responses, we rely on AI to facilitate our communication without disclosing its involvement. As Hancock et al. explain, AI does this in various ways and magnitudes. For example, AI will perform different functions depending on whether it seeks to optimize a message’s trustworthiness or an image’s attractiveness. It can perform small functions like correcting spelling and suggesting replies or large functions like generating an entire message and correcting appearance in videos. Moreover, AI’s abilities are not limited to message suggestions — it can also carry out conversations on its own. These roles can be adopted on behalf of the sender or the receiver. 

The differing levels and modes of potential AI involvement are simultaneously alarming and exciting. When thinking about the implications, Hancock et al. suggest we start small by considering the impact of design choices. These decisions can influence our awareness of AI involvement in the messages we send and receive. Such awareness, in turn, can affect our perceptions of how much agency people have over their communication. Such perceptions may change how content is valued — it is possible that humans are less interested in content thought to be generated “artificially.” Setting matters too. While people may accept AI’s use for translation or autocorrect, would they do so for optimizing social media profiles and self-presentation? 

From there, the authors argue we should think about the longer-term social and cultural implications. AI-MC has enormous potential to impact our use of language and thought processes. When Gmail suggests three options for responding to an email, it is framing how we employ and understand language. A study of smart replies in text messaging found that AI’s suggestions are more positive in tone than replies typically made by humans. This finding raises interesting questions. Can these positive suggestions slowly teach us to use more positive language? Would they make it easier to understand each other? Hancock et al. note that, at the scale of Gmail’s operations, these suggestions have the potential to shift social norms and change language in the long term. They also point out that by allowing for variations in linguistic styles across groups, AI-MC could break down social barriers or strengthen them. 

Awareness of AI in human-to-human communication will influence both how people present themselves and how they perceive others. For example, an earlier study of Airbnb profiles found that people are less trustworthy of AI-involved messages only when they appear alongside messages written by humans. Hancock et al. term this the “replicant effect”, AI-MC’s influence changes depending on its presentation context. Perhaps Airbnb hosts who use an AI-generated profile are perceived as putting less effort into their role when compared to human messages. With this in mind, AI-MC could have similar effects in more intimate relationships. Would we trust and connect more or less when AI is influencing what we say?

This debate matters for society at large because secure and meaningful relationships among individuals translate into well-functioning communities. The nature of communication directly impacts these relationships and thus affects pro-social behavior and altruism in society.  

AI-MC can either bring us together or drive us apart. The changing nature of human-to-human communication will not be without repercussions in policy and political decision-making. Research must guide how AI-MC is designed, regulated, and interacted with at the individual level. Even if AI is increasingly doing the talking, we must not forget that our own choices will play an active role in the future of human thought and communication.