Showing someone your world

The question

If people could see each other’s worlds, would disagreements resolve more easily, because each would see how different the other’s view really is? This page gathers what others have measured.

People see differently, and do not know it

Most people assume they see the world as it is, so whoever disagrees must be uninformed or biased. Psychologists call this naive realism. Perception does differ. In 2015 one overexposed photograph of a dress split millions between blue and black and white and gold, and Pascal Wallisch traced the split largely to what each viewer assumed about the light. Inner pictures differ too. About 1% of people report no mind’s eye at all and about 6% unusually vivid imagery (Wright and colleagues, 2024). Most learn this only when someone asks.

Asking beats imagining

In 25 experiments, people told to imagine another person’s point of view became no more accurate about what that person thought, and overall slightly less. People who could ask the other person became more accurate (Eyal, Steffel and Epley, 2018). A world built from a person’s own words is asking. A world a model invents for someone else is imagining, and models of individuals are still poor at it (see What personal AI can cost).

Showing difference alone can harden people

US Twitter users paid to follow an account that shared posts from the other party did not soften, and Republicans became more conservative (Bail and colleagues, 2018). People already overestimate how far the other side is from them (Ahler and Sood, 2018), and correcting that overestimate reduced Americans’ support for partisan violence (Mernyk and colleagues, 2022). So a picture of difference helps only if it is accurate, and it should show the overlap too.

The clearest working case shows both. In Taiwan’s 2015 debate over ride-hailing, the Polis tool mapped participants into opinion groups and surfaced the statements every group supported, such as insurance for drivers and clear prices. Officials drew on those points, together with an in-person meeting, when they wrote the rules.

Text, voice, pictures and worlds

Text. An AI mediator that drafted group statements from people’s opinions was endorsed more than human mediators and left groups less divided, in studies with more than 5,000 UK participants (Tessler and colleagues, 2024). Personal conversations with GPT-4 reduced belief in conspiracy theories by about a fifth, and the effect held for two months (Costello, Pennycook and Rand, 2024). The skill that can manipulate can also correct.

Voice. Hearing someone explain a view you oppose makes them seem more thoughtful and more human than reading the identical words, and audio alone carries the effect (Schroeder, Kardas and Epley, 2017). Voice calls built stronger bonds than text, without the awkwardness people expected (Kumar and Epley, 2021). Text strips out the person, and voice puts them back.

Charts. A graph corrected false beliefs, such as about global temperature, more than the same facts in text (Nyhan and Reifler, 2018).

Pictures and models. A diagram puts related facts next to each other, so searching and inferring become seeing (Larkin and Simon, 1987). In Chicago’s Pilsen neighbourhood, a sketch artist, mapping and edited photographs let residents shape a plan (Al-Kodmany, 1999). In 2025 workshops in Montréal, generated images gave residents something concrete to react to and helped them trade ideas (Mushkani and colleagues). Pictures persuade, which cuts both ways. Computer animations shown to mock jurors helped them understand the evidence and also biased them; the authors call these facilitative and prejudicial effects (Kassin and Dunn, 1997).

Worlds. After a virtual-reality simulation of losing one’s home, 85% of participants signed a petition for affordable housing, against 63% of those who read a narrative (Herrera and colleagues, 2018). Across 43 studies, virtual reality raised emotional empathy but not understanding of how others think (Martingano and colleagues, 2021). Worlds move feeling more than understanding.

Rhythm, touch, smell and taste. Singing or moving in time with others raised cooperation (Wiltermuth and Heath, 2009), a small-to-moderate effect across studies (Mogan and colleagues, 2017). We found no study of arguing through touch, smell or taste.

Why designers argue in pictures

A drawing talks back. Donald Schön described designing as “a reflective conversation with the situation”, in which the designer draws, sees something unintended and draws again (The Reflective Practitioner, 1983). Gabriela Goldschmidt found designers alternating between seeing a sketch as a form and reasoning about what it implies (1991). Expertise is partly seeing. Practising architects read more functional relations from their own sketches than students did (Suwa and Tversky, 1997).

A drawing also lets people work together without agreeing on everything. Star and Griesemer called such things boundary objects, shared things each group reads in its own way while still cooperating (1989). A plan does that for client, engineer, builder and neighbour. The same power is the risk. A convincing picture can persuade beyond what is true, which is why landscape planners proposed a code of ethics for visualizations (Sheppard, 2001).

What it means for us

Our machine returns ideas as pictures as well as words, and each starts from a written account of how someone thinks. Made from a person’s own account, such a picture is a way of asking rather than imagining. The test this suggests is simple. Pairs who disagree about one brief trade text, each other’s generated pictures or voice, then choose together. It would measure whether they agree, and how fairly each describes the other’s view.

Scope

The studies above test text mediation, voice, charts, virtual reality and planning workshops, mostly in short sessions. None yet tests two people’s generated worlds side by side.