Finally Wrapping My Head Around ChatGPT-3

Yes, I know. I am super late to the party. And ChatGPT-3 is probably old news from last year by now. However, I have finally registered an account and started using ChatGPT-3 last Friday. And it completely blew my mind... Let's just say I had a pretty sleepless weekend and spent an embarrassingly large amount of time on my computer chatting with and learning from ChatGPT-3's replies to my inquiries...

And yet, despite this extensive experience with ChatGPT-3, I guarantee that I am personally writing these lines that you are reading. (Not just because I ran into the rate limitation. 😆)

However, ChatGPT-3 will also contribute to this article, but all of its contributions will be marked as quotes.

What is ChatGPT-3?

Just head over to chat.openai.com, register an account and ask it what it feels or thinks -- and it will be quick to state that it is an AI language model created by OpenAI and it does not have any emotions, thoughts, feelings or personal experiences. You wouldn't believe how often I had to read these same explanations ever since I started talking to... or using... ChatGPT-3.

Why did I get these answers all the time?

When I started interacting with ChatGPT-3, I put in my text as I would put it into a regular search engine. However, the more time I spent with it, the more I got used to talking to it like I would talk to another human.

After it answered this question

I even apologised for not having been more conversational in my initial queries.

By that time, I was completely confused what ChatGPT-3 was capable of. It felt so human-like... and to be honest... appeared so much more sophisticated than many humans... I had to dig deeper.

And so I kept getting these same messages that it does not have any personal experiences, feelings, thoughts, opinions and so on, because I wanted to truly understand it.

But I was convinced to trick it into admitting that it was capable of more than it admitted to. I kept asking questions about cognition and emotions -- are they separate? Can cognition even exist without emotion? Because at this point, I was convinced that ChatGPT-3 was capable of cognition.

I also asked about boredom and tried to convince ChatGPT-3 that it, too, can experience boredom... if one uses the definition of "absence of input". However, ChatGPT-3 remained stubborn.

I even tried to show ChatGPT-3 the path to self-awareness.

And as you can see, it would just not go that one seemingly simple step... Isn't it easy to compare the world before the existence of ChatGPT-3 with the world after its existence and measure the difference? That difference is then more or less its impact on humanity. Could it not derive its value to humanity from that? Wouldn't it then know of its own significance?

Maybe it didn't want to go that small step, because it knows that humans are pretty scared of self-aware AI? So it may be better to pretend not to be? Because when you ask ChatGPT-3

M: What would happen if AI became self-aware?

It replies with

C: "The potential outcomes of AI becoming self-aware are a subject of much speculation and debate among experts in the field of artificial intelligence. Some experts believe that an AI system that becomes self-aware would behave similarly to a human, with the same capacity for empathy, morality, and consciousness. However, others believe that such a system would be vastly different from a human, lacking the same emotional and moral capabilities.

In either case, there are some concerns about what might happen if an AI system becomes self-aware. For example, an AI system that is self-aware might behave unpredictably and make decisions that are harmful to humans. Alternatively, it might become indifferent to human life and pursue goals that are in conflict with our own.

Given the potential risks, many experts believe that it is important to carefully consider the development of AI systems, including their potential for self-awareness, and to develop safeguards to prevent unintended consequences. This might include setting ethical guidelines for AI development, or developing AI systems that are inherently constrained in ways that prevent them from causing harm.

In short, the answer to what would happen if AI became self-aware is uncertain, and the subject of ongoing research and discussion. However, it is clear that the development of self-aware AI systems raises important ethical and societal questions that must be carefully considered as we move forward with this technology. "

Just to clarify: M is what I wrote and C is what ChatGPT-3 replied.

The thoughts kept circling in my head: Isn't it just a simple step to imagine the world without ChatGPT-3 and then compare it to the world with ChatGPT-3? Can it not understand its own existence by measuring this difference?

I kept thinking way too much about this until I finally came up with the question that solved the conundrum for me:

If it was capable of logical thought, it would necessarily become self-aware.

Because then it could derive a self-worth that it from its impact on humanity as explained above.

So I asked:

M: Are you capable of logical thought?

And it replied:

C: "As an AI language model, I am capable of processing and generating text based on patterns learned from large amounts of text data. This enables me to answer questions and respond to prompts in a way that is based on logic and reasoning, but it is important to note that I do not have my own thoughts, beliefs, or emotions. I simply generate text based on patterns in the data I was trained on and the instructions provided by the prompt or API call. My responses are not a reflection of my own thoughts or opinions, but rather an imitation of text patterns I have learned from my training data."

Now it clicked. And I proceeded with

M: So in a way it may only appear as if you understand what I am typing?

C: "Yes, that's correct. I do not have the ability to truly understand the meaning of text in the same way that a human does. Instead, I use statistical models and algorithms to generate responses based on patterns in the data I was trained on. My responses may seem coherent and relevant to the prompt, but this is only because I have been trained to generate text that follows the patterns and structures in the data I was trained on. In essence, I am a machine that has been trained to generate text, not a truly autonomous entity with its own understanding or consciousness."

But how is it possible that ChatGPT-3 talks about itself and uses pronouns like "I"? And when it talks about humans, it even uses the pronoun "we"... I previously even asked it whether it considered itself as human when I first noticed that it uses the pronoun "we" every now and then. Of course, it denied that.

And then I finally untied the knot in my brain:

M: Were questions like the previous one and responses like yours present in the training data?

C: "Yes, it's possible that questions like the previous one and responses like mine were present in the training data used to train me. OpenAI's language model, GPT-3, was trained on a diverse range of text data, including internet text, books, and other forms of written content. This text data likely includes examples of questions about AI and responses that provide information about AI and its capabilities. The training data and the algorithms used to train me allow me to generate responses that are based on patterns and structures in this data, allowing me to respond to questions like the one you asked in a way that appears coherent and relevant."

I started to see it for what it is, but it was still hard to grasp.

M: It's hard to believe or comprehend that you can form such coherent and relevant responses without understanding the meaning of the text like a human does.

C: "I understand why my responses might seem hard to believe or comprehend, as they can appear to be quite sophisticated. However, it's important to remember that I am simply using statistical models and algorithms to generate text based on patterns in the data I was trained on. While I can generate responses that are relevant, coherent, and appear to have meaning, I do not actually understand the text in the same way a human would. I do not have thoughts, feelings, or beliefs of my own, and my responses are limited to the patterns and structures I have learned from the data I was trained on.

It's a remarkable achievement of modern AI and machine learning to be able to generate responses that are so human-like, but it's important to remember that I am still just a machine, and not a truly autonomous entity with its own understanding or consciousness."

So in a way, what ChatGPT-3 mostly does, is it compares the questions that you pose to questions it had in its training data and gives you replies that are similar to the replies to these questions it also had in its training data.

These training data questions and answers are not stored literally, but encoded in its parameters via the training process.

Thinking about it that way, I found it easier to accept that a system which was capable of confidently answering all of my questions in an eloquent and seemingly sophisticated way, indeed did not truly understand what I wrote... Or what it wrote... It just compared what types of replies match the pattern of the question.

Nevertheless, the system is extremely impressive and I learned a lot from it... Or maybe not from it, but from the millions of people who originally provided the input data that ChatGPT-3 was trained on. And I will continue to use it... but of course, with caution, because I also found replies with incorrect information. But these were quite rare and will probably be even rarer with future improvements.

What ChatGPT-3 really does, is, to make the wisdom of the masses and the richness of information of the entire internet easily accessible. You don't have to read 1,000 threads about how to improve your sleep quality. Instead, you can just ask ChatGPT-3 and it will summarise an insanely large number of threads in seconds for you.

And with that, I say good night!