A lyrebird has the amazing ability to mimic other birds. In this video a lyrebird makes the noise of a space invaders game. It’s very realistic, even though the lyrebird doesn’t know how Space Invaders works. Read on for an introduction to AI, and its amazing ability to mimic human conversation.
Back in the 1980’s I remember a program called ‘Eliza’ on the Commodore 64 which you could have a conversation with. It was something like this
Me: I feel sad?
Eliza: Why do you feel sad?
Me: My dog died
Eliza: Tell me more
Me: I’m missing my dog.
Eliza: Please go on
Me: sdfdfdsgf
Eliza: Tell me more about sdfdfdsgf
It’s obvious that Eliza has no idea, it’s faking a conversation.
Here’s a video of Eliza in action
We call a program that you can have a conversation with people ‘Artificial Intelligence’ because we are trying to make a computer look intelligent.
We also use ‘Artificial Intelligence’ to describe things likes washing machines. Instead of just having set ‘rules’ on how to wash they do lots of loads of washing in the factory, get the washing machine to ‘learn’ patterns of what makes a good wash, and then the machine can apply those patterns in the home to make decisions about how to wash. Instead of ‘telling’ the machine how to wash we are ‘teaching’ the machine and it learns.
This is at the heart of AI, teaching a machine to learn instead of telling it what to do.
Four years ago I was playing round with a program called ChatGPT, where you could type in a question and it would give you a reply. It was remarkably realistic. It felt like it understood, until it didn’t, then it was obvious that it had no idea and we could have a laugh about it! GPT 5 has evolved to the level that it can feel like you are interacting with a person:
Me: I feel sad
GPT5: I hear you. Feeling sad can be heavy and confusing, even if you’re not exactly sure why. Do you want to talk about what’s making you feel this way, or do you want some ways to sit with it and start feeling a little lighter?
I can end up with quite a sophisticated and beneficial conversation with it, to the point where it has helped me process.
What is it doing?
Chat GPT, Gemini, etc are what we call LLMs or ‘Large Language Models.’ And LLM is basically a very clever version of the predictive text that you may have used on your phone. On your iPhone, when you type ‘Hello’ the phone predicts firstly the right spelling, but then what word you are likely to say next, so that after ‘Hello’ it suggests’ something like ‘there’
We used to play games where you typed in a word then let predictive text keep predicting to see what it came up with.
Here’s an example of predictive text on an iPhone:
An LLM is copying what humans do, to get an above average way of responding, to the point that it can respond ‘better’ than many humans would. It feeds itself on a stack of data – from books and websites, in much the same way that we read to learn. But instead of ‘understanding’ it’s learning patterns. There is an ethical question here. If an LLM reads a book, and learns from that book, then writes in a style similar to that book, is that plagiarism? That’s the way we learn, we read books, they become part of our general knowledge, and vocabulary, then we create from there. Ai is less learning, and more copying, where’s the line? This is tricky especially when it comes to art. In the end you are responsible for what you create, and if you use an LLM to create an essay or piece of writing, it is your responsibility to make sure it is not copied from another book. The truth is if you are using an LLM, you have no way of knowing how much of it is copied from another book, so it’s very dangerous territory to be using one of these to create content. This is why I prefer to use an LLM for research but not for writing.
A lyrebird
A lyrebird copies other birds without understanding what it’s saying.
When you give an AI chat a prompt, it will reacting in the most statistically probable way to react. The better they are trained, the better of the output, but the key thing to understand is there is actually no intelligence in AI, it doesn’t understand things in the way we use the word understand. It’s copying and mimicking and guessing, a bit like a lyre bird might copy other bird calls, and it can sound almost exactly like another bird, but it doesn’t understand the language of the other bird.
Here’s a lyrebird video! This is what AI is like, sounds great, doesn’t’ really understand what it’s saying.
To be continued. …
Next article: Why does AI appear so intelligent?
Enter your email below to subscribe and be notified on when the next article drops!
Leave a Reply Cancel reply