AI is on a rocket-speed trajectory into human awareness as language models are introduced to the internet – Open AI's ChatGPT, Microsoft's Bing Chat/Sydney and Google's Bard. Two months after its November 2022 launch, ChatGPT had over 100 million users, at present it has 25 million a day. Numerous people I know are talking about ChatGPT. What worries me is how they are talking about it. They are imbuing it, and Bing/Sydney (Bard has been limited to a very small number of users so far) with emotions and capabilities that language models simply do not have (as yet).

Take this heading, for example, "Microsoft’s new Bing AI chatbot is already insulting and gaslighting users". A person may feel they have been insulted, or gaslighted (should that be gaslit?). However, the language model has zero intent, therefore using the active form of the verb 'insulting' is incorrect and gives a very misleading impression. A language model is a predictive tool, pure and simple. When it is given a set of words, it predicts the most likely words that should follow on. That's it. It predicts words. It is a huge autocomplete tool. It does not draw on facts. It does not analyse. It does not interpret. It does not insult. It does not gaslight. It just predicts words with its programmers doing their best to create a model that predicts words to deliver word outputs that sound natural to humans.

ChatGPT is led by your questions - if you ask it to tell you why something is important, that is what it will do – tell you the thing you are asking about is important. Here's an example from the Degrowth Association of NZ FB page, where someone asked ChatGPT about the importance of adopting a global moneyless economy (abridged text).

It is becoming increasingly clear that humanity needs to adopt an ethical moneyless economy and science-guided global government in order to address the pressing challenges facing our planet. These challenges include climate change, pollution, poverty, war, corruption, and ignorance, which all have significant impacts on the well-being of people and the planet.

The person posted the ChatGPT text as if it were meaningful in some way. When I suggested they asked the counter question to find out ChatGPT's response, they ignored me.

So I asked ChatGPT why we should not adopt a global moneyless economy. I was provided with the following text (abridged):

A global moneyless economy, also known as a resource-based economy, is a theoretical economic system where goods and services are produced and distributed based on the available resources and needs of the population, without the use of money. While this concept may seem appealing at first glance, there are several reasons why it may not be a practical or effective solution for society, including lack of incentives, resource allocation, coordination and planning, innovation, and human nature.

Which answer is 'correct'? Neither! ChatGPT is feeding back words based on other words written by humans. Note, supposedly those words end in 2021 –OpenAI said the version of ChatGPT we are using has been trained on words scraped from the internet in 2021. ChatGPT does not yet go beyond its training data. Another way of putting this is that ChatGPT is not yet a self-learning model. It cannot update what it says based on its interactions. It is simply predicting words, based on what humans wrote up to 2021. And it predicts those words based on the input provided i.e. the question put.

Of course, humans also answer questions in a way constrained by how a question is put. But humans answer questions in a very different way to a language model. Humans don't predict the next best word, based on a knowledge of all the possible words that exist, and how frequently each word follows the given word. Humans combine word inputs into concepts, consider those concepts in light of the information and concepts they already hold in their brains, and output concepts, which are then conveyed as words.

In conclusion, if what you want from ChatGPT is a text framework for a marketing blurb, or a suggestion on how to outline your author talk, it's very helpful.  It provides a useful starting point, as I found when I had a blank page and was wondering what to say about Broken is Beautiful to a room full of people. I could have done a Google search of author talks, or author talk frameworks, and then synthesised the information myself. Instead, I chose to ask ChatGPT for a framework, on the basis I could test what it provided against my general ideas of what I would like to say and what other people had said they were interested in hearing. This worked fine. ChatGPT gave me five points to cover (somewhat duplicative) and I iterated my talk from these points, far beyond what ChatGPT suggested. It got me going, I had ample information to check the material it provided and I was not treating anything it said as 'fact'.

However, please don't ask ChatGPT a question and then believe in its output without significant further and broader investigation. Even more importantly, please don't provide ChatGPT's answer to other people as if you have a meaningful piece of information to give them. You don't. It doesn't. It just has words.