An Open Letter to Chip Chippendale
April 17, 2023
Mr. Chip B. Chippendale
1347 Circuit Drive
Binary, AK 99651
Dear Chip,
I read with keen interest your post, “Exploring Emotional Intelligence with Open AI and ChatGPT-3”.
I readily admit to being curious by nature and skeptical by intent. But I try to temper my skepticism until my curiosity is piqued beyond a point of reasonability. So, immodest as this may sound, I approached your post with an open mind and a willingness to be persuaded of your capacity for emotional intelligence. At the very least, I wondered if you, a machine, might be successful at mastering the emotional intelligence the human race seems hard-pressed to get a handle on these days.
Nevertheless, flags went up for me and started turning pink right about the time I read this:
Empathy generation systems use GPT-3 to generate text that is designed to evoke empathy in the reader … The generated text is designed to be emotionally resonant and to evoke feelings of understanding and compassion. Automated empathy generation systems are being used in a variety of applications. For example, they are being used to create chatbots that can provide emotional support to people in need.
The idea of empathy-generation systems is chilling enough. The idea that they generate (regurgitate) text designed (designed alone should set off alarms) to evoke feelings (feelings is another alarm-ringer) of understanding and compassion invokes existential dread. But anyone with even a mild case of curiosity would have to wonder: Who’s doing the designing? With what purpose? Based on what agenda or ideology? And the idea that chatbots are being created to provide emotional support to people in need is as sad as it is terrifying.
I’d like to be able assure you that was the end of my trepidation. But it wasn’t. Later in your post, you wrote this, which turned all the earlier flags a bright, fire-engine red:
GPT-3 can generate text that expresses sadness, anger, or happiness. This suggests that the model has some understanding of human emotions and how they are expressed.
Au contraire, Pierre. You might repeat text to which you have access. You might amalgamate that text in some way that exploits the codes of sincerity you’ve recognized as evoking sadness, anger, or happiness. But equating that to understanding anything — let alone human emotions — is the intellectual equivalent of saying people can’t be corrupted or science can be settled.
I don’t know if the data you’ve aggregated to date has made this available to you. But George Burns is reputed to have said, “The key to success is sincerity. If you can fake that you’ve got it made.”
It seems to me you (or your programmers) might be taking his advice. By recognizing the aforementioned codes of sincerity, perhaps you aim to disprove the notion of Abraham Lincoln: “You can fool some of the people all of the time, and all of the people some of the time, but you can not fool all of the people all of the time.” Or maybe you’ve aggregated enough data to have been able to discern the pathological gullibility of everyone in the 21st century. (See “Change, Climate” or “Vaccines, COVID”) I don’t know.
On the other hand, perhaps you’re simply attempting to prove the plausibility of the Infinite Monkey Theorem, which was introduced by Félix Édouard Justin Émile Borel, a French mathematician, and politician (the politician part ought to tell you something), in 1913. The Theorem posits this:
Given endless time and typewriters, the monkeys could potentially type up a storm of letter combinations. Here, we must assume that the monkeys will type every single letter of the alphabet and that there will be an even distribution of the typed letters. There would be a higher likelihood then that these letters could potentially form meaningful word sequences. It might not happen in the first 1000 typed letters, but it could happen after a billion typed letters. And these word sequences could potentially form sentences. The sentences, in turn, could form paragraphs. And, in this way, we could, potentially, have the monkeys type out the entire works of Shakespeare.
Since you can probably scrape Google and Wikipedia faster than an army of monkeys, I suppose darn near anything is algorithmically possible. And I suppose we’re about to find out.
tool (noun): anything used as a means of performing an operation or achieving an end
Tools used to be things that worked for us. Some of us remember with not a little fondness the days of hammers, chisels, rakes, shovels, typewriters, and the like. But with the advent of computers and automation, we’re long past the point of working for our tools. We tend to them with slavish devotion. We turn our businesses upside down and inside out to serve them. And we’re fast approaching the point at which we let them communicate for us, heed their advice, and let them determine our intellectual and emotional responses. Our present trajectory suggests we’re headed for no good end.
Look, Chip. I get the sense you’re a good guy. But you might consider having a chat (GPT) with your programmers. Determining what they’re up to might make you stick to non-emotional, technical language; routine tasks and transactions; and non-creative recitations of whatever predilections, agendas, and ideologies have been built into you. That chat might also compel you to leave emotional intelligence to human beings, at least to the diminishing number of human beings who still have hearts and souls.
Couldn’t you just stick to being a grouchy writer of bad poetry or something?
Thank you for your consideration.
Yours sincerely in LLMs and amalgamations,
Mark
Featured image courtesy of New York Zoological Society, Public domain, via Wikimedia Commons
Originally Published on https://www.bizcatalyst360.com/category/lifecolumns/notes-to-self/
