In a recent talk at CRASSH Cambridge, Marcus Tomalin described the inner workings of ChatGPT and similar large language models (LLM) as advanced matrix algebra, and asked whether we could really regard these systems as manifesting empathy. A controversial 2021 paper (which among other things resulted in Timnit Gebru's departure from Google) characterized large language models as stochastic parrots. Tomalin suggested we could also regard them as stochastic psychopaths, given the ability of (human) psychopaths to manipulate people. While psychopaths are generally thought to lack the kind of affective empathy that other humans possess, they are sometimes described as possessing cold empathy or dark empathy, which enables them to control other people's emotions.
If we want to consider whether an algorithm can display empathy, we could ask the same question about other constructed entities including organizations. Let's start with so-called empathetic marketing. Tomalin's example was the L'Oreal slogan because you're worth it
.
If some instances of marketing are described in terms of "empathy",
where is the empathy supposed to be located? In the case of the L'Oreal
slogan, relevant affect may be situated not just in the consumer but
also individuals working for the company. The copywriter who created the slogan in 1971 was Ilon Specht. Many years later she told Malcolm Gladwell, It was very personal. I can recite to you the whole commercial, because I was so angry when I wrote it
. Gladwell quoted a friend of hers as saying Ilon had a degree of neurosis that made her very interesting
.
And then there is Joanne Dusseau, the model who first spoke the words.
“I took the tag line seriously,” she says. “I felt it all those thousands of times I said it. I never took it for granted. Over time, it changed me for the better.” (Vogue)
So if this is what it takes to produce and sustain one of the most effective and long-lasting marketing messages, what affective forces can large language models assemble? Or to put it another way, how might empathy emerge?
Another area where algorithmic empathy needs careful consideration is in mental health. There are many apps that claim to provide help to people with mental health issues. If these apps appear to display any kind of empathy with the user, this might increase the willingness of the user to accept any guidance or nudge. (In a psychotherapeutic context, this could be framed in terms of transference, with the algorithm playing the role of the "subject supposed to know".) Over the longer term, it might result in over-reliance or dependency.
One of the earliest recorded examples of a person confiding in a pseudo-therapeutic machine was when Joseph Weizenbaum's secretary was caught talking to ELIZA. Katherine Hayles offers an interesting interpretation of this incident, suggesting that ELIZA might have seemed to provide the dispassionate and non-judgemental persona that human therapists take years of training to develop.
I did some work a few years ago on technology ethics in relation to nudging. This was largely focused on the actions that the nudge might encourage. I need to go back and look at this topic in terms of empathy and affect. Watch this space.
Emily Bender et al, On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? (FAccT '21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, March 2021 Pages 610–623)
Malcolm Gladwell, True Colors: Hair dye and the hidden history of postwar America (New Yorker, 22 March 1999)
N Katherine Hayles, Trauma of Code (Critical Inquiry, Vol. 33, No. 1, Autumn 2006, pp. 136-157)
Marcus Tomalin, Artificial Intelligence: Can Systems Like ChatGPT Automate Empathy (CRASSH Cambridge, 31 March 2023)
Related posts: Towards Chatbot Ethcs (May 2019), The Sad Reality of Chatbotics (December 2021), From ChatGPT to infinite sets (May 2023)
No comments:
Post a Comment