Skip to main content

Would AI lie to you?

 

The short answer is yes. But not for the reason that you think. When artificial intelligence answers your questions using the GPT method, which is what basically ChatGPT is, it does so by assuming or figuring out what is the most likely answer. It uses probability. So essentially AI is giving you what's probably the answer. And sometimes that may not be the truth. 

AI experts call this hallucinations. The AI brain thinks or it hallucinates that an answer is correct. Based on the probability, it would be likely correct. Let's try an example. Let's say you are asking about about something, like a bicycle, not just any bicycle but a specific bicycle. The more detailed your questions become, more the more specific the answer should be. This will lead the AI to build answers which may not be truthful or 100% factual but likely are truthful. The AI model may might not know the specifics of that bicycle, the Cyclemaster 1000, but it may assume that the answer that you're looking for is probably similar or even the same to that of another bicycle, the Windbreaker 5000, due to their similarities. Either bicycle may not be the same or have the same attributes, but it would have a lot of common attributes. How extensive the AI model is, will prevent it answering incorrect answers about the CycleMaster 1000. But if the question gets too close or too similar, the AI model will start answering questions using information about the Windbreaker 5000. Which may not be true or factual.

So The AI brain doesn't intend to lie. What it intends to do is to give you an answer which is most probably correct, even though it may not have the exact information. 

Much like your best friend, who is winging it as a wine connoisseur. 

Comments

Post a Comment

Popular posts from this blog

Should we ask AI to destroy humanity?

I wasn't sure of whether to share this. Not because what it is but because I do not want to contribute to what I don't want to happen the most. But as you will see at the end, my attitude changed and this is the result.  I was asked recently whether killer robots could come and kill us all. Not wanting to lose an opportunity of teaching, I asked the question. How would you build a killer robot? The reply I got was I don't know. Well, I said, let's break this down to what AI we already have. Basically, we need AI to identify the humans and to move the robot arm holding the machine gun to shoot at us. We have vision AI to identify the target, the human. Vision AI currently is sufficient to identify humans and non-humans fairly accurately. This would lead to a high accuracy of recognition by the robot. The robot would be able to identify the human when it sees one. Machine learning would be required to find where humans are likely to be hiding. So an AI robot would need th...
Post a Comment
Read more

In the Beginning

What is artificial intelligence? It is a question that everybody seems to be asking. I would like to answer this in the most indirect fashion ever. Let's not think about what is artificial intelligence. But, let's refocus that to "what does artificial intelligence means to me"? Let's look at something very similar. Something that has become part of our daily lives, something that artificial intelligence will eventually be. Let's look at the internet. I was lucky to be at the beginning of the internet when it became more widely known to the public and became that much more useful. If you ask me then what was the Internet then and then you asked me now what is the Internet, I would have given two very different answers. Largely because I am a technical guy and I basically understand the various components and how the internet works. But what the internet is now, means very different. What the internet means to me. I mean that in a way that we will think of AI in...
Post a Comment
Read more