Just when Creatives thought they would be safe – here comes Generative AI, artificial intelligence that actually generates (i.e. creates) new content. That means writing texts from scratch, producing unique images, videos, music, and more. All with minimal human input. A prompt will do.
No-one will need to take notes of a meeting any more, or write up a summary of the main points, AI will take care of these tasks. More and more companies are now offering their services and “playgrounds” in this area, promising to revolutionize the working lives of millions of people.
From what I can see, the new skill we have to acquire is “prompting”. Rather than telling a colleague to write up a presentation or an agenda, you simply “prompt the model”, for example by straightforward instructions like “Transcribe what is said at the meeting and then summarize the main action points”. Or you start out with a topic such as “The benefits of drinking tea” and you ask first for an eye-catching headline, which will produce perhaps 5 or 6 options for you to choose from. Once you have done that you ask for an introductory paragraph, for an essay, a blog, an e-mail message. And you can specify the tone of voice, such as friendly, professional, persuasive, etc.
A second method of prompting is “by example”: You start an epic novel with a phrase of your choice and then invite the machine to continue. Just imagine! Millions of pages of fiction and non-fiction being churned out literally at the press of a button!
Needless to say, just as in my samples using DALL-E for generating pictures (earlier blog), where I included an additional stylistic instruction (in the style of Matisse, Picasso, Goya, etc.), so you can do the same as an author, by asking that a certain copywriting framework be applied, or by specifying “in the style of Shakespeare”…
After its launch on 30th November 2022, ChatGPT within 5 days counted 1 million people, all keen to test the machine. ChatGPT was asked to describe itself, to write poetry, to think up jokes, to create texts for songs and recipes, you name it.
GTP-3 is a development by OpenAI, originally a non-profit organizations financed by various investors, including Elon Musk. Its originally stated aim was to make AI available and useful to all (wo)mankind. As the enterprise turned out to be rather more costly than originally anticipated, they turned it into a profit-making enterprise, which is now for the main part financed by Microsoft.
Just like in the arts world, where an important prize went to a work that was generated with the assistance of DALL-E, so for the first time the “Deutsche Reporter:innen-Preis” for a scientific piece of journalism went to an article that playfully interacted with contributions from GTP-3. The title of the piece: “Wie lange braucht es uns noch?” (How long will we still be needed for?), its human author: Reto U. Schneider.
It goes without saying that some will condemn GTP-3 and its rivals as a “bullshit machine” or a devilish invention that can create text that sounds good, but lacks authenticity, originality and creativity. Or that it is a parrot that randomly generates sequences of words according to statistical rules…
Reto U. Schneider published his winning article in the NZZ (Neue Zürcher Zeitung) Folio in September 2002. It starts out like this:
Ein Computer wird nie in der Lage sein, einen Text zu schreiben, weil er die Nuancen der menschlichen Sprache nicht verstehen kann. Computer sind gut darin, präzise Anweisungen zu befolgen, aber sie können die Feinheiten und die Komplexität der menschlichen Kommunikation nicht erfassen. Ein Computer wäre zum Beispiel nicht in der Lage, zwischen den Bedeutungen von «ich liebe dich» und «ich hasse dich» zu unterscheiden.
(translation by DeepL: A computer will never be able to write a text because it cannot understand the nuances of human language. Computers are good at following precise instructions, but they cannot grasp the subtleties and complexities of human communication. For example, a computer would not be able to distinguish between the meanings of “I love you” and “I hate you”.)
He then puts the question to his readers as to whether they think this first paragraph is a suitable introduction or not, adding that if the response is negative, well, that’s not his fault, since he did not write a single word of it. Instead he went on beta.open-ai.com/playground and gave it a prompt: “Write a text why a computer will never be able to write a text”.
Mr. Schneider pressed on, asking the machine “What is the difference between ‘I love you’ and ‘I hate you’. And loan behold, he got a very satisfactory, if predictable, answer.
All truly amazing. Even if we know, deep down, that GPT-3 has absolutely no feelings and cannot tell ‘love’ from ‘hate’. Its success is due to its size, which is huge. It gets its abilities from making billions of connections between neurons and has been trained and re-trained by being provided with enormous quantities of examples which allows it to “fill in gaps” and “complete sentences” and “continue a story”.
Given that the waiting list on OpenAI is now so long that it would probably take months to get in, I went on another AI platform, Anyword (“AI that crafts and optimizes your copy”) , and asked it for a Social media post on the topic “There are good reasons for returning to the office now that Covid is over”. Under the parameter “Tone” I was given a choice of neutral, friendly, professional, witty, persuasive.
Going with friendly, it came back to me within 3 seconds:
It’s great to be back in the office! After a year of working remotely, it’s nice to reconnect with our colleagues and work together again. Let’s make the most of this opportunity by collaborating on new projects, sharing ideas, and creating innovative solutions for our clients. #TeamworkMakesTheDreamWork
Changing the tone to “Professional”, again within 3 seconds I was presented with:
We’re excited to be getting back into the office! Returning to our physical workspace brings with it many opportunities and benefits, including increased collaboration, better communication, improved morale and productivity. Let’s get ready for a successful transition back into the workplace! #ReturnToWork #Collaboration
I then tried “Persuasive”, but this made it go into an eternal spin … I guess you need to start properly registering and paying, etc. So that’s where I stopped.
These new forms of “digital assistants” are advertised as helping you improve your writing style and as a cure against writer’s block. Students have discovered it’s a great way of getting through their homework quickly. Teachers can normally tell that a pupil has had intelligent help because the quality of their essays is suddenly unexpectedly high.
In the professional realm, everyone that makes a living out of words is, or will be affected. That means copywriters, translators, writers and poets, as well as reporters and journalists.
Schneider uses this rather provocative analogy: “At one time, the steam engine replaced muscle power. Now, intellectual power is experiencing the same fate.”
He reports a further experiment, which sort of makes me shudder. He writes a few lines about why language is so important for AI. For one thing because language comes closest to what we describe as “thinking”. He mentions that Socher (AI guru and inventor of the writing assistant) says that ‘language is the most interesting manifestation of human intelligence’, followed by one sentence as a cue for GTP-3 to continue: “That’s a fact of which Alan Turing was convinced of”. GTP-3, he reports, picks up the cue and continues the thread without further ado:
In his legendary 1950 paper “Computing Machinery and Intelligence”, he asked the question: “Can a machine think? he posed the question: “Can a machine think?” To this end, he formulated the “Turing Test” named after him. According to this test, a machine would be intelligent if it could convince a human being that he was talking to another human being and not to a machine.
That may sound simple, but it’s not. For what constitutes a human being? And what about the concept of consciousness? How does a machine come to know what a human being thinks and feels?
By this time, our journalist is getting rather worried. He probes GTP-3 what consequences the machine’s language ability will have for his profession, and GTP-3 comes back, unhesitatingly:
At present, it is still difficult to predict to what extent artificial intelligence will influence journalistic work. It is possible that some tasks previously done by journalists will be taken over by artificial intelligences in the future. This could lead to some journalists losing their jobs.
And of course this applies to many other jobs too. In his article Schneider mentions an industrial designer who was able to create 400 new designs for shoes in 2 hours – with the help of KI. He is reported to have said that when using the tools first he was so overwhelmed that he could not sleep. Because tasks that would take him months and months of hard labour can now be completed within a few hours.
And while some artists are outraged and speak of plagiarism, pointing out that AI is reaping the fruit of the work that artists have done over centuries and of their genius, others argue that artists in all epochs and styles have always been inspired by what has gone before them.
Let’s not even think about what we will be seeing in terms of doctoral dissertations in the very near future. Plagiarism checkers have come to the end of their tether.
And the most amazing fact: No-one knows how neuronal networks work. Only that they do.