It's Time for Generative AI to Prioritize Understanding Over Communication

Signup Write New Story It's Time for Generative AI to Prioritize Understanding Over Communication by Kelvin Leong by Kelvin Leong | @ kelvinleong Kelvin Leong is a Professor at the University of Chester, U.K. His current research interest is in the interpretive behavior of Gen AI.

Anna is a Senior Lecturer at the University of Chester, U.K. Her current research focuses strongly on human–AI interactions

Kelvin Leong is a Professor at the University of Chester, U.K. His current research interest is in the interpretive behavior of Gen AI.

Past communication tech conquered physical distance and time. But as generative AI becomes part of how we create, receive, and interpret human messages, it must focus not just on speed and ease, but on reducing the emotional and interpretive distance between people.

Communication is one of the most basic human survival behaviors. People communicate to warn others about danger, ask for help, share knowledge, coordinate actions, build relationships, solve conflicts, and make decisions. In modern society, almost every form of collective activity depends on communication: scientific research, business cooperation, education, government operations, family life, and even an ordinary conversation between friends.

Therefore, communication has never been only about transmitting information.

If communication were simply about moving information from Point A to Point B, then a fax machine would already have solved most human problems.

Unfortunately, humans are not fax machines.

The deeper purpose of communication is to meet human needs. We need to understand the environment around us, and we also need to understand the people around us. We want other people to understand our intentions, needs, and concerns. At the same time, we must work with people whose knowledge, interests, values, and life experiences may be completely different from our own.

So communication looks simple, but in reality it is extremely complex.

At the most basic level, the structure of communication is so simple: a sender creates a message, a medium carries it, and a receiver receives and interprets it. For decades, this has been one of the central structures in communication theory.

The problem is that two sides of this process are human. And once human beings are involved, things are no longer simple. The sender has their own background, knowledge, motivation, emotional state, culture, and understanding of the situation. Meanwhile, the receiver brings an entirely different set of perspectives, which rarely align completely. The sender may believe a message is perfectly clear because, in their mind, ten underlying assumptions need no explanation. However, the receiver cannot automatically know those unstated assumptions. A word that sounds completely neutral to one person may carry strong emotional weight for another. Humor can be interpreted as playful or as disrespect. Similarly, silence might signal agreement, hesitation, uncertainty, or rejection.

Communication technology, as a medium, has been most successful in solving is physical separation.

For most of human history, physical distance was the enemy of communication. The speed of a message depended on how fast a person could travel, how fast a horse could run, how fast a ship could sail, or how quickly a document could be delivered.

Modern communication technologies gradually removed this barrier. The telegraph allowed information to cross long distances through electricity for the first time. The telephone allowed human voices to cross distance in real time. Later, computer networks such as ARPANET connected people, machines, and information systems on a larger scale. The World Wide Web then made information stored in different computers and institutions easier to connect and share. After that, email, mobile phones, social media, instant messaging, and videoconferencing made communication faster, cheaper, and more widely available.

Today, people across several continents can work together, attend meetings, exchange information, and even maintain long-term relationships, while the delay may be only one or two seconds. From this point of view, communication technology has been very successful in reducing physical distance. But one problem has never been fully solved. Distance has been reduced, but misunderstanding has not disappeared.

The reason is not difficult to understand: the most difficult barrier in communication has never been only the distance between two places, but also the distance between two minds.

With the arrival of generative AI, the difference between physical distance and mental understanding has suddenly become even more important.

Today, people already use generative AI to write emails, prepare meeting summaries, rewrite messages, translate text, adjust tone, explain ideas, and suggest how to reply to others. This creates a basic difference between generative AI and most earlier communication technologies. Telephones, email, and videoconferencing mainly transmit messages created by people. Their job is to move something from one side to another.

Generative AI is not only a delivery worker, it also begins to help write the letter. It can help create, modify, and even interpret messages. Hancock, Naaman, and Levy define “AI-mediated communication” as a form of communication in which an intelligent agent modifies, improves, or generates messages on behalf of a communicator.

In other words, generative AI can enter both the sending and receiving sides of communication. It can help people overcome language barriers, simplify difficult information, summarize discussions, and identify unanswered questions. These functions are clearly valuable, but this also makes communication more complex.

Social media greatly expanded access to information, communities, and public expression. However, greater access did not automatically lead to greater understanding.

People interpret information through their own experiences, identities, values, knowledge, and beliefs. This individual perspective does not mean selfishness; it simply means that no two people understand the world from exactly the same position.

One example is confirmation bias, the tendency to seek or interpret information in ways that support existing beliefs. Social media can make this tendency easier to express through choices about whom to follow, which communities to join, and which sources to ignore. Recommendation systems can further shape what users see and may contribute to information bubbles and echo chambers. Research shows that both algorithms and user choices can affect exposure to different perspectives, while similar patterns of selective exposure have been observed across other platforms.

It would be too simple to argue that social media algorithms created confirmation bias or individualism. They did not. Human beings already had different interests, beliefs, and social identities. The main lesson from social media is that technology can amplify human bias.

If the problem of social media is that it helps people select information that matches their existing views, then the problem created by generative AI may go one step further: it may make communication itself more self-centered.

Prompt engineering is often described as a way to improve the quality of AI output. This is not wrong but better prompts do not automatically remove the user’s own bias. Instead, they may turn unclear biases, assumptions, and positions into clearer, more precise, and more persuasive instructions. In fact, generative AI can produce a very strong argument in support of the user’s original position, but that does not mean the user understands the other side any better. Research by Cheng and colleagues supports this concern. They found that sycophantic AI systems may excessively agree with users in interpersonal situations, making users more confident that they are right while reducing their willingness to take actions that may help repair conflict.

The word ‘generative’ should therefore receive more attention when we think about AI-supported communication.

If the internet accelerated how fast we deliver thoughts, generative AI accelerates how fast we produce them. Generative AI can greatly improve productivity, but it can also accelerate miscommunication. This problem matters because failures in communication are not minor technical errors.

Many problems in human life can be traced, at least partly, to failures of communication. A misunderstanding between friends can become a long-term conflict because neither person clarifies the original meaning. Repeated communication failure can damage family relationships.

Poorly expressed criticism can cause emotional harm that lasts far longer than the conversation itself. In organizations, communication failures can lead to incorrect requirements, duplicated work, weak coordination, and poor decisions. A senior manager may receive a clear presentation without realizing that different teams disagree about the meaning of the underlying data.

A company may issue a technically accurate public statement that fails to recognize how customers will interpret it, resulting in a public-relations crisis. In scientific and public communication, misunderstanding can be even more serious. Differences between what scientists mean by terms such as risk, evidence, significant, or uncertainty and what members of the public understand from the same words can affect trust and decision-making.

These examples return us to the starting point of this argument: communication exists to serve human needs.

If people communicate faster but understand each other less, a central purpose of communication has not been achieved.

Communication technologies of the past achieved remarkable success in overcoming physical barriers. Distance became less important. Time barriers became smaller. Transmission costs continued to fall.

The next important goal for generative AI should be to reduce emotional and interpretive distance. Emotional and interpretive distance means the invisible gap between what one person truly wants to communicate and what another person finally understands.

By emotional distance, we do not mean that technology should make everyone agree with each other. We also do not mean that everyone should feel the same way. A world without disagreement is not only unrealistic. It may not even be desirable.

Science needs disagreement. Business needs different opinions. Politics is naturally a place of competing views. Personal life cannot exist without arguments. Different viewpoints can improve decisions and create new ideas. Generative AI should certainly not act as a judge that decides who is right and who is wrong. What should be reduced are unnecessary gaps in understanding. In other words, AI should help people recognize that one sentence may have different meanings. It should identify missing background information. It should separate “what the other person actually said” from “what I think the other person was suggesting.” It should preserve real uncertainty. It should also help both sides understand why they disagree, even if the disagreement remains.

Success should not be measured by whether everyone finally agrees. It should be measured by whether both sides truly understand where their disagreement actually lies.

For generative AI, this suggests a different path of development.

First, success should not be measured only by speed, volume, and efficiency. It should also be measured by the quality of understanding. Generative AI can help people communicate faster. But if misunderstanding also increases at the same time, then being “faster” does not mean very much.

Second, generative AI should not become an authority that decides the “correct way to think” or the “correct way to speak.” Generative AI can help people understand how their words may be interpreted by others. But it should not remove reasonable differences in expression simply to create surface-level harmony.

Third, AI should be an objective companion rather than the final judge. Its role should be to encourage reflection and broaden the user’s view, rather than deciding what the user should believe.

Finally, humans should continue to take an active role in important communication. Choosing a word, asking a question, explaining an idea, or taking responsibility for something that has been said may seem troublesome. But these processes are themselves part of understanding.

The history of communication technology is, in many ways, a history of reducing distance.

The telegraph separated information from the speed of physical travel. The telephone allowed voices to cross distance. Computer networks connected machines and institutions. The Web connected information on a global scale. Mobile and social technologies made communication increasingly immediate and continuous.

Each development made the world physically smaller.

Generative AI introduces a different opportunity because it does not only transport human communication. It can participate in creating and interpreting it.

This makes the next challenge more difficult, but also more important.

If generative AI develops without considering the distance between human understandings, it may automate the problems that already exist. It may allow us to produce misunderstandings faster, defend incomplete interpretations more persuasively, confirm our existing views more comfortably, and distribute poorly aligned messages at unprecedented scale.

We may become more connected while becoming less understood.

Shannon, C.E. and Weaver, W. The Mathematical Theory of Communication. University of Illinois Press, Urbana, IL (1949).

Hancock, J.T., Naaman, M., and Levy, K. AI-mediated communication: Definition, research agenda, and ethical considerations. Journal of Computer-Mediated Communication 25, 1 (2020), 89–100.

Nickerson, R.S. Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology 2, 2 (1998), 175–220.

Bakshy, E., Messing, S., and Adamic, L.A. Exposure to ideologically diverse news and opinion on Facebook. Science 348, 6239 (2015), 1130–1132.

Cinelli, M., De Francisci Morales, G., Galeazzi, A., Quattrociocchi, W., and Starnini, M. The echo chamber effect on social media. Proceedings of the National Academy of Sciences 118, 9 (2021), e2023301118.

Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., and Jurafsky, D. Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391, 6792 (2026).