Background

Poetry is difficult to define, but most scholars, poets, and laypeople can agree that poetry consists of expressive language with the intention of creating meaning. Since the advent of the written word, poetry has been an incredibly important part of human cultures and history — in fact, even before that, ancient civilizations were orally passing down epics and legends in the form of poetry, keeping their culture alive. The puzzle of generating poetry with artificial intelligence, too, is not a new mystery.

Perspectives

I’ve tried to write poetry with AI, and I’ve never been able to use the output. It’s hard to prompt it in a way that captures the vision in my mind but I can gather inspiration that gets me where I want to go.

Shaadaan, 20

The point of poetry is to reach someone, and you do that by creating an experience for the reader that involves them. Poets ask, ‘So what?’ AI can’t do that.

Lisa Russ Spaar, Professor of English and Creative Writing at UVA

What AI has already accomplished is spectacular, and its further advances will continue to change the world. But for all the functions an AI can potentially achieve — an intelligent program will fall short of authenticity as a poet. AI lacks what is most needed to place the footprints of its own consciousness on another mind: inner experience.

Keith J. Holyoak, Distinguished Professor of Psychology, UCLA

It didn’t have the context to be able to help me convey the specific style / sound when writing poetry, and I wanted it to sound like myself.

Interviewee, 20

I think that the separation between large language models and human writers is unhelpful. A large language model is not a writer, but a writer can write using a large language model. The human versus AI polarity is encouraged by big tech as it makes large language models seem more powerful and creative than they are — it’s a marketing ploy.

Hannah Silva, poet

Demo

Credit to Huayuan Song for Haikool!

We have attached an example of an AI system that is able to generate poetry in the form of a haiku, a Japanese form of poetry with a specific amount of syllables in each line (5-7-5). As you will see, hallucinations are common in the form of misunderstanding how many syllables each word has, and these haikus are often much less interesting than human-generated ones from the past. However, we hope that you will play around with this demo and see what you think about AI-generated poetry.

Background

Visual art is one of the oldest and most common mediums humans have used to express themselves. Originating all the way from prehistoric cave paintings, evolving into the masterpieces studied in the Renaissance, to the first animations and their integration with digital technology, visual art has evolved and continued to be an important way people express culture, emotion, and ideas throughout history. As AI initially developed, it was a natural question to see what it could do for such an important section of human art.

Perspectives

If what AARON is making is not art, what is it exactly, and in what ways, other than its origin, does it differ from the 'real thing?' If it is not thinking, what exactly is it doing?

Harold Cohen, creator of AARON

Demo - Black Forest Labs Image Generation

We have attached an interactive demonstration of an AI image generation model created by Black Forest Labs. You can input detailed text prompts or upload example images to merge or edit. As you experiment, pay attention to how styles vary based on your inputs. How specific do you need to be to get a certain output? Pay attention to finer details like lighting and textures. Does prompting in a certain way give a more desirable result? Do the resulting images accurately reflect your vision for the prompt? Does it contain your soul and the emotions you wanted to convey? And ultimately, is it art?

Background

The definition of what is considered music has been a subject of constant debate. Oftentimes, pieces that were ridiculed in the past become works studied deeply in the present. Music has been a cornerstone of humanity throughout history. Whether used as a core part of culture or just as entertainment, it has never ceased to be a key part of society. Developing from hand-made wooden instruments to symphonies, and now entering the digital age, music has constantly evolved alongside humanity. Despite how subjective music is, it is often quite mathematical in nature. With its foundation in rhythm, frequencies, harmonies, time signatures, and even the overtone series, it is no wonder that music has a history of attempting to blend with computational and algorithmic advances.

Perspectives

I am often told ... about soul in music. But I look at the notes on the page and listen to the music the performers play from those notes and I am struck by a single, clear and resonant thought: the soul I hear, if indeed I hear one, is my own.

David Cope, Composer, Professor of Music, Early AI Adopter

I came across a band that got over a million streams on Spotify, and it was an AI band. It made me annoyed and sad that it is possible for this to happen organically. I feel like it is expanding into a space that I would like to instead be filled by a young original artist.

Jack, 22

The power of a great artist is their ability to create music that inspires others, to move them and spark flames of creativity. Covers by human musicians are a way to pay tribute and express appreciation; AI covers felt like cheap imitations and mockery by comparison.

Katelyn Chedraoui, AI reporter for CNET

I feel like no matter what type of music genre, artist, and intent, if it is made by a human being, there is some kind of realness to it, even if it is good or bad, and AI music lacks the intention that is needed to make good music.

Bushra, 21

Music is a verb. It's not a noun. It's not a thing sitting on a table. It's something we do and we can express this 'doing' in a lot of different ways, and music takes a lot of different activities to make it happen. It's all part of the musical ecosystem.

Richard Randall, Associate Professor of Music Theory, Carnegie Mellon University

As far as I can tell, nothing about the song sounds "fake", more than how all recorded music is fake to varying degrees, especially now. There are no analogues of the "extra fingers" that sometimes tip you off that an image is AI-generated.

Joseph Pater, phonologist, UMass Amherst

Demo (Tencent Song Generation)

We have attached an interactive demonstration of an AI music generation tool created by Tencent. You can write your own lyrics, choose a genre, input a text prompt, or input an audio prompt to create a song of your choice. Pay attention to the quality of vocals and instruments, and more importantly, the form and uniqueness of the music. Think about how you feel using this tool. Do you feel like you're creating music? Does the output feel personal to you? How much control do you have over the creative process? And most importantly, is it art?

Executive Summary

Every participant reported using AI tools in some form, and nearly half described daily or near-daily use. Even so, opinions about AI art remained deeply divided: 4 participants said AI can produce real art, 4 said no, and 3 gave a conditional or mixed answer.

The strongest pattern across the interviews was a separation between classification and value. Several participants were willing to call AI output "art" in a broad sense, but still valued human-made work more because of intention, effort, backstory, craftsmanship, or emotional depth.

The identification activity reduced confidence more often than it increased it. Of the 10 interviews that stated pre-activity confidence, 5 were low, 4 were medium, and only 1 was high. After the activity, 7 of 11 participants said the conversation changed their thinking — usually by making them more skeptical of their ability to spot AI reliably.

Where numeric activity results were recorded, performance was mixed rather than dominant. Recorded percentiles ranged from the 9th to the 70th percentile, and the three transcripts with raw scores averaged 60.0% accuracy on the 50-image test.

Metric Summary

Interpretive Write-Up

The interviews show a clear tension between everyday AI use and resistance to AI authorship. Participants used AI for studying, writing, coding, summarizing, recipes, trip planning, or image generation — yet many still treated art as a specifically human domain. The strongest recurring language centered on the "human element": respondents repeatedly described art as tied to soul, struggle, effort, backstory, craftsmanship, or emotional truth.

At the same time, the interviews also show that people do not draw a simple line between "art" and "not art." Several participants accepted that AI output can count as art in a broad category sense, especially when someone prompts, directs, or iterates on the result. But even many of those respondents still valued it differently. In practice, the question was not only whether AI output qualifies as art, but whether it deserves the same cultural value as human-made work.

The identification activity pushed participants away from certainty. Many expected texture, brushstrokes, flawed anatomy, shadows, or a vaguely "computeristic" feel to expose AI. Instead, several participants described being fooled by pieces they were sure were human. This matters because the interviews suggest that the social meaning of AI art is now shifting from simple detection to trust, labeling, and context.

Another important pattern is that authorship changed the interpretation. When participants learned that a beautiful work was AI-generated, they often said the image felt flatter, less meaningful, or less worthy of deep analysis. When they learned a work was human-made, they often immediately restored value through labor, biography, history, or intention. This suggests that in this sample, authorship actively shapes how viewers decide whether a work deserves attention, emotion, or respect.

Overall, the interviews point to a strong class theme for AI & Humanity: people are not just evaluating outputs — they are evaluating relationships. The participants kept returning to who made the work, what process produced it, how much effort was involved, and whether human originality is being displaced. The debate over AI art in this sample is less about surface quality alone and more about authenticity, attribution, and what kinds of creative labor people want society to value.

Dominant Themes

The following themes were coded across all interviews (multi-select; one interview can contribute to several themes).

Method & Coding Approach

The semi-structured interviews were converted into a coded dataset so responses could be compared consistently across participants. Each participant was coded on background characteristics, AI usage frequency, whether they believed AI can produce real art, pre-activity confidence, post-activity change, activity results when available, and recurring themes.

Because the interview format was qualitative, not every transcript included the same level of detail — some metrics use smaller denominators. For example, pre-activity confidence was explicitly stated in 10 of the 11 interviews, and percentile results were only reported in 4 interviews. Theme counts are multi-select researcher codes, meaning one interview can contribute to several themes at once.

Appendix: Coding Scheme

Six artworks are shown below. For each one, guess whether it was made by a human or by AI. Click a button to reveal whether you're right! All images are from Scott Alexander's AI Art dataset.