Does AI understand us?

By now, I’m sure you’ve heard of AI tools like ChatGPT and how they can generate human-like responses to complex prompts and questions. When we get a human-like (and relevant) response from an AI it seems as if it understood what we asked. It seems to have understood the language we used.

Maybe this isn’t so strange. After all, AI models are trained on billions of web pages, books, images and data sets. In a way, they know more than any human ever could. But can an AI really understand us?

The gift of knowledge

The data sets that an AI uses to generate responses are incredibly large. Imagine being asked a question and scanning an index of 800 billion web pages for the answer – kind of a Google search but in your mind. That’s akin to what is happening with AI. And not just webpages but millions of books, maps and images, and numerical databases with millions and millions of rows of data – all indexed, processed and connected in an enormous web of connections.

An AI that uses these huge amounts of data to provide answers in language we can understand is called a large language model (LLM). How it constructs its answers is to predict the next word in each sentence, based on what it knows of every sentence it has ever read.

It works, word by word, to predict the most likely next word. This is why you may have heard some people describe AI as being a very clever auto-complete.

The difference here is scale, speed, and its ability to change what it has written as it progresses with the answer. Learning as it goes.

So, does AI understand us?

The short answer is, not yet. Here’s why…

The Chinese Room

Alona Fyshe gives a fantastic analogy of how AI works in her TED Talk – Does AI actually understand us? She calls the analogy the Chinese room.

Imagine a room with no windows and inside sits someone who can’t understand Chinese. Instead, they have with them some instructions on how to respond in Chinese to any Chinese sentence.

From outside, a person posts a card with a question written on it on it through a small slot in the wall. The person inside matches these symbols with those on the instructions, and they write the corresponding reply on the card and post it back through the slot.

Can the person outside the room tell if the person inside the room can understand Chinese? There is no way of knowing. All that can be measured is the accuracy of the returned response. Because there is no requirement to understand Chinese to complete the task, you cannot use the response to determine what is happening inside the room.

This is how we use AI. To us on the outside, it looks and feels like AI understands us. But it doesn’t – at least, not in the way we use the word ‘understand’.

What does AI say?

If AI understands us then is seems only fair to ask it. So, I asked ChatGPT if it could understand us, and this is what it said:

As an AI language model, I am designed to process and generate human-like text based on the input I receive. I have been trained on a wide range of topics and can understand and respond to natural language inputs in various languages. My understanding of language is based on the patterns and structures present in the vast amount of text data I have been trained on. However, I don’t “understand” language in the same way humans do, as I lack consciousness and genuine comprehension. My responses are generated by identifying and predicting patterns in the text, rather than by truly understanding the meaning behind the words.

I then asked it if it thinks it will ever understand us?

As an AI language model, I am a product of programming and training, and I do not possess the ability to “think” or possess self-awareness. My abilities are limited to what I have been designed and trained to do, which is to process and generate human-like text based on input.

So, definitely not artificial intelligence as it is portrayed in science fiction novels and films, then.

As for the future of AI, it’s difficult to predict if it will ever truly ‘understand’ language in the same way humans do. Researchers are working on improving AI models to better comprehend context, nuance, and emotion in language. However, whether this will lead to genuine understanding or remain a sophisticated pattern-matching process remains to be seen. As Alona Fyshe says in her TED Talk, the only way to know is to see what’s happening inside the Chinese room and compare that with what happens in the human brain.

Top tips for using AI, from AI!

Given that AI currently doesn’t understand language the way we do, how can we get the most coherent, natural-sounding, useful responses from AI?

I asked ChatGPT if there is anything we could do to help it understand us better?

As a user interacting with me, your input can help me provide more accurate and useful responses. Here are a few ways you can help me understand your questions better:

  • Be clear and concise: Provide enough information in your questions or statements, so I can understand the context and respond accurately.
  • Use specific examples: If you’re asking about a particular concept, providing examples can help me better understand what you’re looking for.
  • Avoid ambiguity: Ambiguous questions or statements can lead to confusion. Try to be as precise as possible when formulating your questions.
  • Be patient: If I don’t understand your question or provide an unsatisfactory response, feel free to rephrase it or provide more context.

While these suggestions can help me generate better responses, it is important to remember that I am an AI language model and my understanding is based on pattern recognition rather than true comprehension.

Then I said ‘thank you’, just as I would for any help from an intelligent being…

The need for meaning

I said ‘thank you’ because it’s incredibly difficult not to see AI response as intelligent. Humans look for familiarity and meaning within our language and world. That’s why we use words like ‘understand’ and ‘intelligence’ to describe something which is essentially a pattern-matching system (albeit an INCREDIBLY sophisticated and impressive pattern-matching system).

Just looking at the input and output of AI it’s easy to be fooled – it responds with human-like answers and so we treat it as human. But, as Alona Fyshe says, to really know what’s going on, we need to see what’s happening inside the room. (This is harder than it sounds, to find out why check out my other post and video on why no one really knows how AI code works).

The best we can say right now is that if AI does understand us, it’s in a very different way to ourselves. What we can do is look at how we can use AI better – both in a way that trains it (helps it improve) and produces more useful and accurate responses.

I often write about marketing tools and techniques that are not ethical and effective, and there will be more content relating to AI and non-AI topics coming up. If this blog has got you thinking, become a subscriber to get my latest content sent to you every Monday.