鶹ƽ

Skip to Main Content
鶹ƽ College
First-Year Experience

“In AI we trust?”

July 14, 2023

Around the time when was making headlines last fall, I was teaching an upper-level physical chemistry course. I asked ChatGPT one of my homework questions, and it got it wrong. It conflated the entropy of the universe and the system, a common student error, but it also claimed that more ordered ice molecules had higher entropy than liquid water molecules, which any of my students would know is false. Disturbingly, however, the confidence and clarity with which the AI responded made me almost start to doubt my own understanding of the topic.

In Kazuo Ishiguro’s novel Klara and the Sun, we become intimately acquainted with an AI that surpasses the capabilities of today’s technology and approaches the ability to resemble or imitate a human. While characters in the novel discuss and disagree about Klara’s ability to act human, they seem to take for granted the accuracy of the things Klara tells them. Rick and his mother ask Klara to help tutor him in school. Both Rick and Paul, Josie’s father, are easily convinced to help Klara in her task to persuade the Sun to send Josie special nourishment for her recovery. Maybe it is human nature to not question something when you are told it might help someone you love, but I also think Ishiguro captures our tendency to give the information we receive from a computer a special aura of certainty and objectivity. As we proceed into an era where our interactions with AI are more and more ubiquitous, we must remember to doubt what computers tells us and inquire about the data used to train these algorithms.

AI can get things massively wrong. There is the trained on current employees and asked to rank resumes in a stack of applicants. It penalized resumes that mentioned women’s colleges or participation in women’s clubs and sports, as these were not commonly found among Amazon’s current employees. AI asked to produce an image of a group of computational scientists generates one containing mostly or all white men. Automatic face detection software has . An AI is only as reliable as the data it was trained on, and it reproduces biases and subjectivities of its human creators. This is inevitable, but problems arise when we start to treat computers as infallible. Klara’s capabilities are impressive, and I am curious about the algorithm that taught her to understand human emotions. But, her information on certain topics is flawed. Using multiple AIs designed by different people might provide more reliable results. When I asked ChatGPT how to help Josie, it responded “While the sun is an astronomical object and does not possess consciousness or the ability to make decisions, encourage Josie to get regular sunlight exposure, engage in outdoor activities, maintain a healthy diet, consider vitamin D supplements, and foster a positive mindset to support her well-being.”

Now the question is, which AI do you trust more?

 

HUMAN or AI?

Aurelia Ball
Associate Professor of Chemistry