Artificial Intelligence (AI) has undergone dramatic transformations over the decades, from early neural network research to today’s powerful large language models. In this insightful discourse, Prof. Christopher Kello, Chair of the Cognitive Information Sciences Department, shared his thoughts on AI’s evolution, its role in cognitive science, and its implications for academia.
A Journey Through AI
Research
“I’ve been working on neural network research since my graduate school days,” Chris explained. "Back in the ‘90s, these were called connectionist models, and they were simplistic representations of human cognition. The computational power then was nowhere near what we have today." Despite these early limitations, progress was made. "Geoff Hinton, who was working as a cognitive scientist at the time, has now won the Nobel Prize for his contributions. The evolution from those early models to today’s deep learning and AI is incredible."
However, AI's development was not always linear. "I lived through the third AI winter. Neural networks first emerged alongside the development of computers in the ‘40s and ‘50s, then experienced cycles of hype and disappointment. By the time I entered graduate school, AI had gone through two previous hype cycles. Each time, the technology didn’t quite meet expectations, leading to funding cuts and skepticism."
The lack of computational resources and data in the past hindered progress. “My own models became unwieldy, and the available compute power wasn't enough. Many researchers, including myself, pivoted into fields like neuroscience and complex systems. AI, as we see it today, is very different from what we envisioned then.”
AI’s Changing Definition and Capabilities
Reflecting on how AI has been redefined, Chris noted, “When we thought about AI decades ago, many envisioned an intelligence capable of autonomous reasoning and self-improvement—an artificial brain that learns and grows like a human. Today’s AI, however, is more like an advanced pattern-matching machine. We initially believed that solving symbolic reasoning problems, like playing chess, would naturally lead to broader intelligence. But it turned out the hardest problems were actually the easiest ones to crack, while tasks like physical movement and decision-making in the real world remain incredibly difficult." One of the ongoing challenges with modern AI models is their opacity. "They were always black boxes," Chris said. "Even in the ‘90s, neural networks had complex, nonlinear processes that we couldn’t fully interpret. But now, the scale of this problem is vastly greater. We have to study these models just to understand what they’re doing."
AI in Research and Academia
Despite the challenges, AI has proven useful in research. "We’re studying how large language models adjust their language use based on the audience. We’re essentially using AI to analyze AI-generated data, which is a strange meta-experiment in itself." However, he acknowledged the need for validation. "We cross-check results using traditional methods—counting proper nouns, conducting spot checks, and using statistical analysis. AI can generate insights, but it’s not infallible."
AI is also making its way into the classroom. “I use it to curate readings and generate tailored course materials. I write prompts carefully to generate the material, and then I provide students with prompts and responses so they can see how to engineer prompts to get good results.” When asked about AI’s impact on writing in education, he observed, “One of my freshman students told me, ‘My writing is starting to sound like ChatGPT.’ That concerns me because students are now learning from AI rather than from great human writers. This raises the risk of intellectual stagnation—what some researchers are calling ‘model collapse.’"
Academic Integrity and AI
With AI's growing presence in academia, issues of academic integrity are at the forefront. "I draw the line at students presenting AI-generated writing as their own. But what about students who use AI to refine their grammar? That’s a gray area. I am okay with students improving their grammar by getting feedback from chat bots, but instructors need to decide for themselves.” AI policies remain a moving target. "Funding agencies now say you can’t use AI to review proposals, but you can use it to write them. Journals ask for disclosure if AI was used. There’s no universal standard yet, and that makes policy-making difficult.”
The Future of AI in Academia and Beyond
Despite the rapid advancements, Chris emphasized that no one anticipated AI would progress this quickly. "None of us saw this coming – at least I sure didn’t." Looking ahead, he sees both promise and peril. "AI will be a fundamental tool in everything we do, just like the internet or smartphones. But we must be careful about over-reliance and ensure that human creativity and critical thinking remain central."
Chris explains it brilliantly: AI is transforming research, teaching, and writing at an unprecedented pace. While it offers powerful tools for knowledge creation and learning, it also presents new challenges in understanding, ethics, and academic integrity. As academia navigates this evolving landscape, the key will be striking the right balance between leveraging AI’s strengths and maintaining rigorous human oversight.
Interested in contributing how you AI? Email sghosh@ucmerced.edu
