Yes, of course I too have been playing with ChatGPT, the new chatbot. And of course I have also followed the reactions to this chatbot with great interest. As expected, these ranged from seriously concerned to extremely enthusiastic. From a plea in one of our Dutch quality newspapers to restrict the use of ChatGPT, “because we are heading straight for an information apocalypse” and “we as a society are not prepared for it at all,” via concerns about students who will now have their essays written by ChatGPT and teachers who won’t notice, to the prediction that we will soon be able to outsource all our writing work to these kinds of machines.
Now for my own experiences. I have tested ChatGPT in various ways. First of all, I would like to point out that chatbots like this are not really new. They have been used for years to cover, for example, sports news and economic news. But ChatGPT is a step, no, a big step further. What do you use ChatGPT for? Compared to previous chatbots, you can now really communicate with this bot in plain Dutch. And in many other languages, for example Finnish or Urdu. ChatGPT seems to be therefore also a great translation machine, better than the well-known Google Translate and also better than DeepL, which is highly valued by many experts. (NB: This text was translated from Dutch by ChatGPT)
Then the use as a text writer. You can indeed ask ChatGPT to write an essay or blog of any length on (almost) any topic and ChatGPT will deliver the desired tekst to order. An essay about, for example, the usefulness of futurology. According to ChatGPT, by the way, an excellent means of making better policy decisions in many areas. Or about the relationship between happiness and philosophy or between animal welfare and the environment. And staying with my profession; you can also ask ChatGPT, for example, to develop different scenarios about China’s position in the world in 2040.
These assignments are not only well formulated, but they are also fairly complete. They are a bit boringly written, but you can then ask ChatGPT to rewrite the text in the style of a well-known author, such as Dan Brown or John Grisham. However the content of the text remains somewhat mediocre, not really exciting, too much the greatest common denominator of what you will find on the internet. And sometimes ChatGPT also makes real mistakes and confuses, for example, the philosophers Locke and Hobbes. The program “doesn’t really know” what it is writing. It only combines and reproduces existing texts.
In short, ChatGPT is a wonderful tool for writing standard texts, such as complaints letters, policy recommendations, reports from consulting firms, assignments for school, etc. ChatGTP is also a powerful tool for texts for internal use, where style does not matter as much. You can use ChatGTP, for example, as a “devil’s advocate”. For example, if you want to quickly get the five most important arguments for or against XXX, or if you want to quickly inventory what “people” think of something. That last application is also interesting for a futurologist, for example when asking for scenarios for China in 2040. That answer was very “average”. If you want a “bold” scenario, you will still need to hire a professional.
However, even with these standard texts, a human is still needed. First of all, to formulate the task in such a way that ChatGTP actually delivers the output you want. And secondly, to check that output for any factual errors and sometimes also to clean up the text. Sometimes there are almost literal repetitions in the text. But ChatGPT already writes better than the average Dutch person and also has much more “knowledge” than the average Dutch person. Or better formulated: ChatGTP can produce coherent and readable texts on many more topics.
Something similar also applies to ChatGPT’s sisters and brothers that produce images or music. Dall-e2 (ChatGPT’s brother) works in the same way. You type in a task, for example “make a photo-realistic rendering of a camel playing golf or make an image of a house with solar panels in the style of Van Gogh” and Dall-e2 produces a much more beautiful image than I could ever make myself. And probably also more beautiful than the average Dutch person could make. (The photo with this blog was made by Dall-e2) In the same way, the music bot “Flowmachines” composes music on request in any desired style and performs it for you as well. And also much better than I ever will.
These bots all derive their skills from absorbing a huge amount of music, images, and texts from the internet. That brings me to the question of what will happen to us if we start using more and more of these tools (and their even further improved successors). At first glance, we are entering a true artistic paradise. If ChatGPT is already capable of such things, we will soon be able to call upon the most beautiful new texts in the style of Victor Hugo or Leo Tolstoy or any other style you can think of. And not just beautiful new texts, but also new images and music, made by ever-better bots in any desired style and type. And for those who still cling to art made by a “real” person, there will surely be publishers and artists who will want to give computer-generated texts, images, and compositions a touch of human (apparent) involvement. The computer as the perfect “ghostwriter,” composer, or video artist.
This brings me to the almost philosophical question of whether machines are really able to reflect the thoughts and feelings of real people. I suspect they are. After all, they have access to the entire world literature and can consult the total of all social sciences. I see no reason why a machine that is more well-read than I am, has seen more images than I have, and knows more music than I do, could not produce a version of the world that I find richer and more convincing than my own version.
But such software can apparently produce language, images, and music without knowing what it is about. And that leads me to the question: what is language then? And what is meaning? What is the essence of meaning? Do we as receivers of the words, images, and sounds give meaning? And the sender? Is the medium then completely neutral?
And there is one more follow-up question. These machines derive their skills from “absorbing” huge amounts of text, images, and music. If we then also put these new images and texts back on the internet, that “training set” becomes increasingly computer-generated. Instead of these machines using language and images made by humans, we humans will increasingly use language and images made by machines.
And why would these machines then be limited to human language? They will also be able to “translate” language between humans and other animals. This will lead to conversations between humans and other animals. Something we may have always wanted, but I wonder what it will mean for the way we currently relate to other species.
In short: ChatGPT is another important step forward in the automatic production of substantive texts, but we will have to learn how to deal with it. This not only applies to sectors such as education, professional life, and art and culture, but also to our entire human existence. Will we still be human if we live in a world where texts, images and music are increasingly thought up by computers and less and less by humans? And the most essential question for me personally: how can we ensure that this world becomes an increasingly better world?
© Peter van der Wel 2022
This text was translated from Dutch by ChatGPT