The working paper’s primary authors are business school academics. One has a PhD in psychology, but until AI research includes authors who are licensed practicing psychiatrists and psychologists to inform the work, I find the research hard to trust. We haven’t even figured out how to make LLMs safe. Trusting it with vulnerable populations, even if it is just to lend an ear, while it may yield a reduction in loneliness, imagine the despair a lonely person may feel as they become increasingly dependent on a thing that could never know them, understand them, or empathize.
Hi Krista, thanks a lot for this comment. Your suggestions is absolutely relevant, and I'd look forward to see a similar development. I hope the authors can listen!
To the extent that this is conversation you are having with yourself and your level of neurodivergence, I think "AI" in the form of chatbots can be an amazing tool to get more out of life, and to understand who you are as a person, and who you are against others. This is something you have to experience personally to fully understand.
Hi Riccardo, this is an interesting set of studies!.You noted “Study 5 compared a fully developed AI companion to an AI assistant that lacks empathy and a simpler chatbot that can only handle basic tasks”. (I would have expected a control group which did not include any chatbot.) Does the study report on the number of people in each of these groups? Do they claim that the differences they noted are statistically significant?
Hi Karen, thanks for your questions! The authors recruited 1479 partecipants for the Study 5, and "excluded 98 for failing a comprehension question, leaving 1381". Furthermore, differences were significant: "we ran two-sided t-tests to see whether loneliness reduction in the AI companion condition was higher compared to control and AI assistant conditions. We found that loneliness reduction in the AI companion condition was significantly higher compared to both the control condition (MAI Companion = 8.73; MControl = -0.51; t(920.9) = 7.58, p < .001, d = 0.50), and AI assistant condition (MAI Companion = 8.73; MAI Assistant = 2.10; t(930.4) = 5.79, p < .001, d = 0.38)".
What if we replaced the lack of human interaction with…. human interaction?
That's a great suggestion! 🤣
The working paper’s primary authors are business school academics. One has a PhD in psychology, but until AI research includes authors who are licensed practicing psychiatrists and psychologists to inform the work, I find the research hard to trust. We haven’t even figured out how to make LLMs safe. Trusting it with vulnerable populations, even if it is just to lend an ear, while it may yield a reduction in loneliness, imagine the despair a lonely person may feel as they become increasingly dependent on a thing that could never know them, understand them, or empathize.
Hi Krista, thanks a lot for this comment. Your suggestions is absolutely relevant, and I'd look forward to see a similar development. I hope the authors can listen!
To the extent that this is conversation you are having with yourself and your level of neurodivergence, I think "AI" in the form of chatbots can be an amazing tool to get more out of life, and to understand who you are as a person, and who you are against others. This is something you have to experience personally to fully understand.
Thank you a lot for sharing this!
Hi Riccardo, this is an interesting set of studies!.You noted “Study 5 compared a fully developed AI companion to an AI assistant that lacks empathy and a simpler chatbot that can only handle basic tasks”. (I would have expected a control group which did not include any chatbot.) Does the study report on the number of people in each of these groups? Do they claim that the differences they noted are statistically significant?
Hi Karen, thanks for your questions! The authors recruited 1479 partecipants for the Study 5, and "excluded 98 for failing a comprehension question, leaving 1381". Furthermore, differences were significant: "we ran two-sided t-tests to see whether loneliness reduction in the AI companion condition was higher compared to control and AI assistant conditions. We found that loneliness reduction in the AI companion condition was significantly higher compared to both the control condition (MAI Companion = 8.73; MControl = -0.51; t(920.9) = 7.58, p < .001, d = 0.50), and AI assistant condition (MAI Companion = 8.73; MAI Assistant = 2.10; t(930.4) = 5.79, p < .001, d = 0.38)".