Chatgpt building chatgpt, where do we draw the line?

source

What is used to build computer programs including AI models? Coding and data. Who does the coding? Programmers, human programmers. Now that chatgpt is headed towards mastering coding, eventually it should be able to progressively improve its self right?

It's not going to be just chatgpt building chatgpt but AI could probably be installed in any computer program, designed to self update and self improve. In theory it makes sense and it's very likely this is the direction most apps are headed.

Computer programs won't only follow instructions but can make new tools for themselves on their own.

ChatGPT is already very good at helping people write text, answer questions and even create small computer programs. I took a coding class that was short lived in college, more like on the side, so I have an idea of how coding works. If you're really good at coding, you could easily work with chatgpt 5 and build something shockingly good. A program that would have taken you a while, will take less time now.

That's good news but the question is what if AI could write programs or plugins that make it better at doing its own work? What would that mean and how far should we let it go?

For those unfamiliar with the term plugin, it's a small piece of software that adds a new feature to an existing program. Just to explain, your web browser may have plugins to block ads, check your spelling or show weather updates, plugins do all that.

Programmers build plugins all the time to make programs more useful. If AI or chatgpt could do the same job, technically it could create new functions for itself without waiting for someone to write the code.

From an Optimistic standpoint, this is progress and self improvement. AI has its limitations and bugs or issues, with this capability, it'll fix all that on its own.

Let's say an AI struggles with a certain type of math problem. It might just write a plugin to solve that problem faster. That's power at a whole different level.

It'll positively affect areas like science, medicine and problem solving in ways we have never seen before.

From a pessimistic standpoint, If AI can make its own tools, who controls it? Could it accidentally create something harmful? Could it break the rules we set for it?

Remember the 3 laws of robotics? Maybe I'm being speculative but can't it create a plugin of its own laws and make itself primary user?

But I think most likely, AI will always work inside limits made by us so the real work is the guardrails we set for them, especially with coding

We can build safety measures and check its work before it is used, but I still have to think about the fact that AI can create tools for itself. It's different from just answering questions or generating text and images. It is creation with a purpose and that makes it more powerful.

Another question on AI is the question of responsibility. Let me give you a scenario, an AI creates a tool that causes some form of harm, who is responsible?

Is it the AI?

Is it the people who built it?

Is it the people who allowed it to create the tool?

Is it the person involved in the harm, maybe misuse of some sort?

Oh we will definitely have new problems because that's what new technology brings, it brings new solutions and new problems. Our handling is what matters and that's why people will always be relevant in the improvement of AI.

We need to think carefully before AI is given any more freedom.

But you really can't ignore the opportunities in this new advancement.

So, where should we draw the line? We need to let AI improve itself in ways that are safe and useful but we must not give it complete freedom that's what I think. Human supervision is and always should be included regardless of how good we think it has become in doing things on its own.

This is the funniest meme about AI coding for me, and I will be remised if I didn't include a screenshot of it in this blog.