What you want to know about GPT-3

Typewriters that write by themselves still are typewriter.

What you want to know about GPT-3 /img/gpt3-face.jpg

GPT-3 (“Generative Pre-trained Transformer”) is software that automatically produces, or analyse, human-like texts, in ways that may have a huge impact on society. This posts summarizes an article on GPT-3 nature, and effects, published in November 2020 by L. Floridi and M. Chiriatti.

First, the basics: who mowed that lawn, and when should it matter?

Robots are nothing else that software, or “Artificial Intelligence” if you will, powering something that looks like a human being, or some part of it.

It is impossible, looking at a perfectly mowed lawn, to know for sure whether its mower was a human gardener, or some robot. But this does not mean, in any sense, that that robot has or should have the same nature, dignity, behavior or rights as the human gardener.

In other words, what “Artificial Intelligence” really delivers is an ever growing decoupling of the ability to get something done as specified, from any need to be (humanly) intelligent to do so. I already covered this fact, explaining why robots can NEVER dance. GPT-3 is the latest, text-specific development in this decoupling process.

What can we do with GPT-3?

GPT-3 is as eay to use as any search engine. First, you give it some input text, called “prompt”. Then, after comparing the prompt to the huge quantities of text it has already analysed, GPT-3 starts writing as much “related” text as you want. GPT-3 has been already used with good results to a “wide range of use cases, including summarization, translation, grammar correction, question answering, chatbots, composing emails, and much more."

Is this human? Is this artificial REAL intelligence?

No, of course. See above. Just like the robot mower, or the robots that SEEM to dance, again. It is not what is achieved but how it is achieved that matters.

And GPT-3 works without the slightest real understanding of what the heck it’s writing about. What professor Lister said in 1949 still applies:

“Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain - that is, not only write it but know that it had written it."

Three Tests, one conclusion

Floridi and Chiriatti ran three tests to check how well GPT-3 performs with logico-mathematical, semantic, and ethical requests. In the first two cases, they concluded that you’d better not use GPT-3 to solve math problems, or understand some text. The ethics test went exactly as expected. Since GPT-3 “learns” from human texts, it “reflects some of humanity’s worst tendencies”, whatever stereotype you give it to work on:

What you want to know about GPT-3 /img/gpt3-about-black-people.jpg

The conclusion is quite simple: “GPT-3 is as intelligent, conscious, smart, aware, perceptive, insightful, sensitive and sensible… as an old typewriter." Any interpretation of GPT-3 as the beginning of… a general form of artificial intelligence is “merely uninformed science fiction”.

Good and bad consequences of GPT-3

What you have read so far does not mean that we should ignore software like GPT-3, if nothing else because it does, already, “write better than many people”!

It is reasonable to expect that, thanks to GPT-3-like applications, we will have more possibilities to understand ourselves, and the world around us. At the same time, we will have to get used to things like:

  • The impossibility to always know for sure if something was written by software, or humans
  • “Industrial automation of text production”, from extremely useful manuals or catalogs, to credible, but automatic fake news
  • Political or advertising web pages, rewritten from scratch by software in real time, every time, to match the current reader

How to get ready for a GPT-3 world

Whatever GPT-3 will do, it seems very likely that:

  • People whose jobs still consist in writing will need to be good at prompting software like GPT-3, and then edit its output. The good news is that this will still require a lot of real human brain power
  • Complementarity among human and artificial tasks, and successful human-computer interactions will have to be developed
  • [In general] Humanity will need to be even more intelligent and critical
  • Business models should be revised (advertisement is mostly a waste of resources)


(this is the actual, untouched warning of the original article. I pasted it verbatim because I really like it, but also as a reminder of why this website needs your support)

“This commentary has been digitally processed but contains 100% pure human semantics, with no added software or other digital additives. It could provoke Luddite reactions in some readers.”

License: as the original article, this post is licensed under a Creative Commons Attribution 4.0 International License. The face image is by Gerd Altmann from Pixabay.