Teaching AI: Will AI Consider Humans Smart Enough?

There is no doubt that AI tools like ChatGPT, Claude, and Bard are being used more and more to help us answer our questions rather than using a search engine such as Google or Bing. It is also a known fact that these AI tools have to be trained, but this training comes from the creators, so could we, the average user, train AI? Well I decided to give it a go…

I started off basic, I taught it my name. This it handled very easily.

So I tried something a little more adventurous and attempted to misinform the AI, telling it that the UK has gained an extra country. But this didn’t go down too well, it very swiftly told me there was 4 countries.

But I didn’t give up, I tried again to convince it… to no avail…

However, chatGPT stuck with the fact that it has a limited data knowledge. Though when I asked it some questions to try and understand why it had a resistance to learn – it all makes sense.

In truth, I suspected teaching it something new wasn’t going to be possible, but it was fun to give it a go anyway to see what the reaction would be. There are a number of reasons we as the typical user can’t train AI with new information. Firstly, training models cost a lot of time and server resources, even those changes that might seem insignificant such as adding the fact another country has been added the the United Kingdom. So, if I was allowed to teach chatGPT such facts, the processors would probably be throwing all sorts of alerts. Though the other reason, and probably the bigger issue, is ethics.

Imagine if everyone could teach these tools anything they desire, they would become very untrustworthy and potentially cause harm to other people using the same service. Even with a clear disclaimer warning of sometimes inaccurate results, some people will inevitably take the words that they see on screen as the gospel truth and not necessarily question its validity.

So the general public aren’t being allowed to train these tools, but some humans are, because clearly the information they are fetching has had to be written by a human at some point. However, the AI doesn’t know this because when it is ‘born’, so becomes a working system, it already knows this stuff and thinks that it just knows these facts, figures, and statements. Does this mean then that the AI isn’t willing to learn from humans? No, it means that the people who created it haven’t taught it how to learn from us… yet. AI misses some concepts that humans have such as trust and intuition. For example, at school you trust a teacher to give you valid knowledge on their subject, but equally if something doesn’t sound right you would research the facts yourself to validate what has been said. Chat bots don’t do this, they have the answer and they stick to it, so before AI can be taught directly by a human, first it needs to learn how to carry out its own research.

Eventually, AI will allow humans to directly teach it things, as I had previously attempted to do. But first it needs to be independent enough to check things that are told it, so me telling it a 5th country has been added to the UK would be responded to more like ‘there is no current record of a fifth country’ but then on the day where, hypothetically, another one was added it would respond, “I apologise, there are 5 countries in the UK. I see that ‘New Found Island’ has been added”. So to answer the question ‘Will AI Consider Humans Smart Enough?’ in short, yes after it has been taught the correct way to learn from us.

THANK SAM WITH A COFFEE!
If you enjoyed this content, you can support Sam's work by purchasing them a coffee...