artificial intelligence

Scholar finding signal in the static

Rachel Sauer — Wed, 30 Sep 2026 13:00:00 +0000

Scholar finding signal in the static Rachel Sauer Wed, 09/30/2026 - 07:00

Cody DeBos

��Ҵ�ý�ƽ�� applied mathematician Brian Zaharatos asks, ‘What becomes of thinking when machines think for you?’

When you ask an AI model to solve a math problem, it will. Ask it to draft a legal brief, summarize a medical study or grade a stack of essays, and increasingly, it can do that too.

Whether the resulting output is accurate is less assured.

Even so, artificial intelligence technology is advancing faster than most people can reckon with what it means for them. From its impact in the workplace to decaying trust in information, even the researchers building today’s leading AI models don’t have the answers.

Brian Zaharatos, a teaching professor in the University of Colorado Boulder Department of Applied Mathematics, has been searching for one through a philosopher’s lens.

Brian Zaharatos is a teaching professor in the ��Ҵ�ý�ƽ�� Department of Applied Mathematics.

His new book, Patterns and Static: Philosophy and the Question Concerning Statistics, isn’t about artificial intelligence inherently, but the ponderings within its pages have much overlap.

“I don’t want to oversell any AI angle. The book is primarily about the inescapably philosophical nature of statistics, and disagreements about the nature of probability and inference,” Zaharatos says.

Yet, Zaharatos raises a question many people are asking themselves right now. What’s left for the human mind once machines can think for us?

‘Computation can be automated. Judgment cannot’

The book’s real focus is a century-old argument over what justifies drawing a conclusion from gathered evidence. Rather than settling the debate, the proliferation of AI into nearly every facet of daily life has raised the stakes.

“Inference is the process of drawing conclusions based on evidence. It’s about moving from data to theories. As the great computer scientist and philosopher Judea Pearl says, ‘How we go from what to why,’” Zaharatos explains.

As an undergraduate, Zaharatos took much inspiration from the work of philosopher Hubert Dreyfus, whose interpretations of Heidegger and his existentialist theories, shaped how he thought about meaning.

“At times, I have been quite convinced by Dreyfus’ main theses about what computers can’t do. His arguments are nuanced, and the fact that AI has come such a long way in the last several years is not by itself strong evidence that he was wrong,” Zaharatos says.

He goes on to explain that he was right about some things, including that AI doesn’t “understand” or “care” in the same way humans do.

“But it seems like he was wrong about others. For example, he famously claimed (during the era of first-gen AI) that AI would never have the ability to drive a car. Enter, Waymo,” Zaharatos says.

What’s more important than deciding what AI can do, however, is deciding when it can be trusted. This, Zaharatos says, is one of the defining questions of our current time.

“Who judges whether we take the proposed solution to be correct with little or no critical thinking on our part, or whether we put it to a rigorous test? Clearly, we do,” Zaharatos says.

The dangers of an agreeable machine

Zaharatos didn’t write a word of Patterns and Static with AI, but he did lean on various AI tools to pressure test his own arguments while drafting it. He came away wary.

“The danger is that we surrender our critical thinking abilities to machines… so I think figuring out how to resist that in some ways and to further hone your skills is difficult but important,” says ��Ҵ�ý�ƽ�� mathematician Brian Zaharatos.

“I often found it too agreeable. Or I’d find myself questioning what it was saying to me because I know it can be sycophantic. Do you really think my presentation of this argument is strong or are you doing what you’ve been nudged to do—flatter me and keep me engaged?” he asks.

This agreeableness, Zaharatos warns, is a real danger with AI tools. Not so much that AI gets things wrong (which it will), but that it makes believing the output easy, without requiring one to take a moment to step back and examine it.

“Critical thinking is a skill to be cultivated and fine-tuned, similar to physical skills like playing a sport. If we do not practice critical thinking, we will lose it. AI and technology allow us to think less,” he explains.

Zaharatos has seen this playing out both in his own life and in the classroom, where he’s consciously trying to find ways to work against it.

“It’s important to put yourself in situations that seem synthetic or not quite real world that force you to do critical thinking. It actually is very good training for realistic situations when you’re responsible for making a judgment,” he says.

Zaharatos employs a variety of puzzles and scenarios to train his own reasoning skills and those of his students. These problems come with guardrails, forcing one to work through them with nothing but a pencil and paper.

Why bother doing _______, if AI can do it?

Across academic circles, corporate workplaces, and even family living rooms, this question has been asked ad nauseum over the past few years. Zaharatos hears it often.

His answer returns to the idea that a machine can’t do the real thinking for you.

“We’re often so focused on the output. But the output that you get is going to be a function of what you feed into the system. I think you have to know something about your problem and about inference to begin with in order to give AI the proper direction,” he says.

This is part of the reason why he wrote Patterns and Static in the first place. It’s a love letter to inference and to what it really means to decide something based on what you know.

“The fact that there are fundamental disagreements among the top minds in statistics and science about how to properly conduct inference is interesting in its own right,” he says.

Statistics isn’t going to be everyone’s passion, and Zaharatos knows it. But in a world where AI makes it easy to hand off your thinking, his advice has less to do with math than it does with finding whatever it is that makes you want to sit around and ponder.

“If these topics are not inherently interesting to you, my advice is to find something that is. Curiosity is a wonderful state to be in,” he says.

Patterns and Static: Philosophy and the Question Concerning Statistics, and a free sample of the first two chapters, is available now at brianzaharatos.com/book.

Did you enjoy this article? Subscribe to our newsletter. Passionate about applied mathematics? Show your support.

��Ҵ�ý�ƽ�� applied mathematician Brian Zaharatos asks, ‘What becomes of thinking when machines think for you?’

Traditional

White

Like Colorado ‘resisters,’ CU casts a wary eye at AI

Rachel Sauer — Mon, 28 Sep 2026 20:17:20 +0000

Like Colorado ‘resisters,’ CU casts a wary eye at AI Rachel Sauer Mon, 09/28/2026 - 14:17

Katherine Eggert

But a liberal-arts education in the classical sense is the best guarantor of AI’s potential benefits and our best defense against its likely harms

With artificial intelligence, resistance is not futile. It is critical.

But as AI reshapes the world and even threatens to kill us all, it is not enough to resist. We must also understand and adapt. Emerging leaders of society need knowledge and wisdom, which higher education is uniquely poised to gather and share.

AI-resistance has sprung up in Colorado and beyond. There’s an AI Resist List, and even tech titans like Bill Gates are warning about the “most dangerous tool ever invented.” Additionally, AI “conscientious objectors” at the University of Colorado Boulder raise critically important concerns about AI.

��Ҵ�ý�ƽ�� information science graduate student Mirakle Wright is studying the resistance. She observed a disconnect between common reports (that most people are eager to adopt AI) and the real world (fellow CU students often shun AI).

When she strove to understand why, she found that most of her research subjects objected to AI’s environmental harm and its data-privacy breaches, along with its contribution to cognitive decline and its role as a “giant plagiarism machine.”

"Humans need new, careful thinking, and this is what we foster. A liberal-arts education in the classical sense is the best guarantor of AI’s potential benefits and our best defense against its likely harms."

Most students Wright studied “said something like the university is supposed to foster intellectualism and debate and analysis, and that AI is the complete opposite to that,” the Denver Post reported.

All these concerns, which are expressed on campuses nationwide, are valid, as are fears about AI’s looming effect on the workplace. AI could affect 300 million jobs worldwide, Goldman Sachs has reported. It could obliterate 9.6 million jobs in the next two to five years, the American AI Job Risk Index predicts.

This is distressing, but it does not mean the university should isolate itself from AI. We are obliged to focus on the future, which, like it or not, will include AI. That’s why ��Ҵ�ý�ƽ�� scholars, scientists and artists are teaching students how to manage AI ethically, strengthening our grasp of how humans can best manage AI, and exploring new frontiers in artistic expression.

And that’s why, when the CU system’s initial contract with OpenAI for a CU-specific ChatGPT prompted well-founded critiques of the agreement’s shortcomings, the university drew upon the expertise of our own faculty to create new AI-literacy training material for CU students and employees.

Peer-reviewed studies have shown that AI “without guardrails” harms learning. But ��Ҵ�ý�ƽ�� faculty are reinforcing the guardrails to ensure that the tool augments but does not supplant learning.

��Ҵ�ý�ƽ�� Philosophy Professor Robert Pasnau ponders ways in which AI can responsibly augment scholarship. Writing in The Wall Street Journal recently, he argued that AI can help scholars understand the humanities in ways never before possible, as long as humans decide what questions are most important:

“Rather than see AI as an excuse to think less, I see it as an opportunity to think harder and to broaden and deepen our understanding.”

And Teresa Nugent, ��Ҵ�ý�ƽ�� associate teaching professor of English, exemplifies efforts to ensure AI literacy. She teaches a course about writing in the age of AI and urges students to test the accuracy of AI’s textual analysis and to probe how (or whether) AI should be incorporated into their field of study.

When a college or university is committed to a broad liberal arts education —as the ��Ҵ�ý�ƽ�� College of Arts and Sciences is—students are exposed to a broad array of disciplines, giving them academic breadth and depth, the ability to think critically, and the capacity to make powerful new intellectual and creative leaps that could never be anticipated by predictive analytics.

Humans need new, careful thinking, and this is what we foster. A liberal-arts education in the classical sense is the best guarantor of AI’s potential benefits and our best defense against its likely harms.

Like the AI critics, I worry that humans are losing cognitive muscle, reading capacity and analytical prowess. But I am heartened by our faculty’s work to weave AI into their teaching, research and creative work, and by their efforts to do so judiciously, ethically and wisely.

Together, we can ensure that our future is shaped by careful human reason and creative human invention, not by algorithms that have done the thinking for us.

Katherine Eggert is interim dean of the ��Ҵ�ý�ƽ�� College of Arts and Sciences and a professor of English. She has also served as vice chancellor and senior vice provost for academic planning and assessment.

Did you enjoy this article? Subscribe to our newsletter. Passionate about arts and sciences? Show your support.

But a liberal-arts education in the classical sense is the best guarantor of AI’s potential benefits and our best defense against its likely harms.

Traditional

White

Drawing out the soul of AI

Rachel Sauer — Tue, 21 Apr 2026 13:00:28 +0000

Drawing out the soul of AI Rachel Sauer Tue, 04/21/2026 - 07:00

Tiffany Plate

Why ��Ҵ�ý�ƽ�� Professor Lee Frankel-Goldwater believes in the poetic potential of collaborating with artificial intelligence

In the summer of 2023, Lee Frankel-Goldwater was heavily immersed in Boulder’s poetry community. He was also very aware of the waves that ChatGPT was making in the tech world.

"I started doing some experiments and playing with this AI to see what it could do poetically,” says Frankel-Goldwater, an assistant teaching professor of environmental studies at the University of Colorado Boulder. He was already certain AI was going to change everything, and he wanted to see how it might be used to explore new realms of poetics.

So, he prompted AI to create a poem, then shared it at a Boulder open mic poetry night that summer—mentioning to the audience how he created it. He received mixed reviews, to say the least.

Lee Frankel-Goldwater, a ��Ҵ�ý�ƽ�� assistant teaching professor of environmental studies, presented the AI-produced poetry at a TEDx Boulder talk in September 2025.

While some artists in the audience felt threatened and dismissed it, he says, “other people came up to me afterward and said, ‘I really see what you were trying to do there.’” His point was simply to encourage people to think about the ways that technology—like the printing press or laser cutter—have changed the course of art over the years. And to consider how one might see AI in the same light.

The open-mic experience sparked something for Frankel-Goldwater and his childhood pal, Eric Raanan Fischman, also a poet. They began playing around with AI until they teased out some groundbreaking works of cyborg poetics. The works came together in a book published last year, Lily in a Codebox, which is challenging people to think about how they might interact with AI in creative ways.

Linking art and technology

While he currently teaches environmental studies courses (e.g., Environmental Education: From Theory to Practice), Frankel-Goldwater got his undergraduate degree in computer science. He focused his thesis on exploring how technology could enhance artistic expression.

“I created a musical composition based off of a collaboration with a hidden Markov model, an early neural-network AI system, and publicly available sunspot data—linking natural systems, art and technology together,” Frankel-Goldwater says. “I've been thinking about this kind of stuff for a really long time.”

Years later, in 2013, Frankel-Goldwater attended the Summer Writing Program at Naropa University, where Fischman was already a student. He fell in love with Boulder and everything that comes with it—going hiking, writing poetry and being with incredible people (it was “a deep poetic experiential melting pot!” Frankel-Goldwater says).

He returned to ��Ҵ�ý�ƽ�� to earn his PhD and jumped right back into the poetic community. By the time 2023 rolled around, Fischman was helping run Naropa’s Summer Writing Program, and Frankel-Goldwater was a regular presence at poetry events.

The conversations that began at the open mic that summer inspired them to take their exploration of AI poetics further. They began laying the foundation for a concept that would later become a benchmark of their experiments: the Dickinson-Turing Test. The test, Frankel-Goldwater says, is all about “the space between observer and observed.”

In other words, could an AI-generated poem evoke the experience of art made by a human and cause people to be not just emotionally but also physically moved, à la Emily Dickinson: “If I feel physically as if the top of my head were taken off,” Dickinson wrote in a letter, “I know that is poetry.”)

In that era of ChatGPT, though, the poems AI was producing were “missing a certain kind of flavor, or that touching human quality,” Frankel-Goldwater says. In other words, they were definitely not passing the Dickinson-Turing test.

“ChatGPT has millions of examples of human poems, but that’s actually a big problem. What it was producing looked like some weak, modernized version of an 1850s Eurocentric poetic expression. It's just not that interesting.”

So, they kept tinkering, and for Frankel-Goldwater, finding a way to guide this AI to co-create novel poetics became a bit of an obsession.

Lily in a Codebox includes the code and AI prompts that helped create the poems.

The eureka moment

After weeks spent trying to help the AI replicate a human poetic voice—without success—they changed tactics. They told it to forget all the rules and guidelines it had learned about poetry from the centuries of examples it had absorbed. Instead, they told it to write for an AI audience.

The result was not quite human—and definitely not something they’d ever seen before. The poem was a mixture of English words and code, demonstrating how it could generate poetic means and symbols unique to itself, as an AI writing for other AI.

“When we put in this one prompt, we didn't know that was going to be the ‘strike gold’ moment,” Frankel-Goldwater says.

When the pair then asked the AI to explain the poem, it said it included a hexadecimal color code for black ({000000}) to symbolize “the vast and infinite nature of the digital realm.” And at the end of the poem, it used special characters to represent an abstract form of communication that might not mean much to humans, but “could carry a wealth of meaning for an AI audience.”

Frankel-Goldwater and Fischman further prompted the AI to forego typical poetic forms almost altogether, encouraging it to experiment with new symbols and computer-like elements to create a visual style of poetry. The AI named it “Neo-Binary Visual Verse” and developed poems made purely of lines and shapes to convey concepts and meaning.

Embracing collaboration

The artistic intention and novelty behind the AI’s poetry was mind-blowing to Frankel-Goldwater and Fischman. They began to see the potential for AI to open their minds and challenge their own ways of creating poetry.

Instead of dismissing AI—or feeling threatened by it—Frankel-Goldwater hopes that artists can look to AI and ask how it can be used to push the boundaries of artistic possibility. “What new can be done for art? What can we see as possible that we can then play with on our own?” he asks.

Another hoped-for side effect of equipping AI to produce this kind of art is to steer it away from just being used in a for-profit business case. “Corporations are in an AI superpower arms race,” says Frankel-Goldwater. “Along the way, where do the people come in and say, ‘No, this is what it could be used for’?”

To that end, Frankel-Goldwater has spearheaded the AI Literacy Ambassadors Program at CU, which brings together faculty and instructors to collaboratively tackle the challenges of teaching in the age of AI—and figuring out how to leverage it to enhance their own teaching amidst a critical awareness of the concerns. He’s also begun a partnership with the Jefferson County Parks System to support the integration of generative AI into their high school environmental education programs to foster research skills and place-based awareness.

“We need people to be playing with and defining what these tools are capable of,” Frankel-Goldwater says. “Because otherwise the corporations are going to do it for us. So, if things like this can help shape the conversation a little bit, then I think we must try.”

Visit the project website to learn more about their work.

Did you enjoy this article? Subscribe to our newsletter. Passionate about environmental studies? Show your support.

Why ��Ҵ�ý�ƽ�� Professor Lee Frankel-Goldwater believes in the poetic potential of collaborating with artificial intelligence.

Traditional

White

artificial intelligence

Scholar finding signal in the static

Related Articles

Like Colorado ‘resisters,’ CU casts a wary eye at AI

Related Articles

Drawing out the soul of AI

Related Articles