Jay McKinney, Visiting Assistant Professor of Cognitive Science

7 May 2025

Listen and subscribe to The Year of Curiosity: AI podcast on Apple Podcasts or Podbean.

Highlights from talk with Jay McKinney, Visiting Assistant Professor and newly hired Assistant Professor of Cognitive Science, who shares their perspective on how generative AI “thinks,” the likelihood that this technology will lead to a singularity and the coolness of giant robots.

Curiosity

George is curious about the possibilities that AI may allow us to customize mass entertainment – what if we can change the plot of our favorite show or improve the ending of a show we thought ended badly, like Seinfeld, The Sopranos, or How I Met Your Mother. We could create instant fan fiction. At the same time, that would give us even less common ground to talk about and potentially increase our isolation.

Jennifer has been curious about the wooly mouse. The wooly mouse was presented as a mouse with wooly mammoth genes. The goal is that this company will de-extinct or bring back to life the wooly mammoth by introducing genes to a close living relative, like an elephant. The genes used are most likely genes that alter hair growth, altered them in a transgenic mouse and claimed that they made a wooly mouse. So, the question is – does this show we can make a woolly mammoth or that we can make a hairier mouse? The publication was in Nature, which, in its headline, questioned the claim Meet the ‘woolly mouse’: why scientists doubt it’s a big step towards recreating mammoths. We should also remember the 2015 chicken with a dinosaur’s face, where researchers changed a chicken’s gene for beak development as birds and dinosaurs are related. We should also remember the FGF5’s role in hair growth.

Jay is curious about building very large, Godzilla-size robots. Robots exist but tend to be small in size; what if we had very large, bi-ped robots that could unload freight ships?

AI Origin Story

Jay McKinney

Jay learned about the future of AI and AI dystopias from anime like Ghost in the Shell or Gundam Wing. As a psychology and philosophy double major, Jay cared about the theoretical subject matter of Alan Turing’s work, and how it relates to the kinds of skills that we build or lives that we have. Technology is not just computer-aware technology, it is also things like buildings, doors, or wrenches, all of which augment our lives in fundamental ways across different developmental timelines. A new tool can change how a whole industry works, leading to a kind of dialogue between objects and our species.

Some Highlights

Jay McKinney, currently a Visiting Assistant Professor at Carleton and soon an Assistant Professor, did graduate work at the University of Hawaii, Manoa, and the University of Cincinnati. Their research interests include embodied cognition, games and gaming culture, Japanese philosophy, and AI.

One question about our current versions of generative AI is whether it is thinking or reasoning. This is not a new question about AI but has been one since the 1950s. Jay’s answer to this question is a resounding no. While it is impossible to say this about every single existing variation of generative AI, the theoretical argument Jay shares with students starts with Turing’s 1950 paper Computing Machinery and Intelligence. Turing does not ask whether computers can think; rather, he asks whether humans can be made to believe that the machine can think like a human more than 50% of the time – the Turing test or Imitation Game. The next step in thinking is guided by the paper On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? from 2021. This paper focuses on environmental harm, people attributing humanity to AI, the AI’s inability to represent information accurately. If the benchmark of being human is appearing coherent, and the machine does this more than 50% of the time, does that constitute thinking or being human?

While Turing’s test raises interesting questions about what it means to think like a human, Jay’s point is that today’s AI companies are simply saying that their products think and reason (and therefore they need more money), but what they leave out is the embodied experience that really makes human thinking human, including the social environment. For example, we use language by combining sounds, words, structures – but our language is also shaped through the empathetic environment around us that shaped us as human beings through our youth and beyond.

Jay plays a game in their class, Rocks, Plants, Cats, People, AI, to show students which of these categories can have an eventful cognitive day. Students join teams, one for each category, to decide what kinds of interactions they may be able to experience. When it is the turn of Team AI to talk about its interactions it becomes clear that it, just like Team Rock, has to wait for interaction but cannot start any of its own. People, on the other hand, tend to be constantly in action, even if it just a wiggle or postural sway or micro movements, and this constant movement is cognitively relevant.

Students in Jay’s class get a lot of context and questions that allow them to deal with information or misinformation about AI; they have developed the skills and skepticism to think through claims about and generated by AI. Students learn answers and contexts for the questions that arise from generative AI and the theories around it. This grounding allows them to then explore further, a strategy similar to other introductory courses at Carleton.

Jay reminds us that we do not really know how thinking works. One question is then, will students come in as more passive humans because they are relying too much on generative AI? Will they look for reasons to make AI use okay or are they more ready to approach it critically? Some students are tech-optimistic and think that they can use AI to save a societal problem. This is a good place for them to be and then to learn that life is more complex. Other students recognize the limitations that exist (right now). We also recognize that we will have more and more trouble with arcane technologies like books and finding them accessible to us.

José Bowen draws the distinction between cognitive technologies that make you smarter and cognitive technologies that make you dumber. For example, a map requires you to think and learn and process in a way a tool like MapQuest does not. Generative AI, because of its parroting that rarely repeats itself, makes it difficult to predict outcomes and thus become truly skilled at using it – and repetition may water down the outcome. This also means that having a conversation with generative AI that builds on a series of prompts will lead to a degradation of output and to a degradation of coherence, especially if the input comes from generative AI.

One struggle with generative AI is to find fitting metaphors for it – as metaphors help humans understand a concept and relate to it, not finding a metaphor that works makes explaining generative AI difficult.

Tools/Recommendations

Return to AI at Carleton Blog