AI Self-Awareness Debate

respond to this critically: A newbie’s views on AI becoming “self aware” Discussion hey guys im very new to the topic and recently enrolled in an ai course by ibm on coursera, i am still understanding the fundamentals and basics, however want the opinion of u guys as u r more learned about the topic regarding something i have concluded. it is obv subject to change as new info and insights come to my disposal and if i deem them to be seen as fit to counter the rationale behind my statement as given below - Regarding AI becoming self-aware, i do not se it as possible. We must first define what self-aware means, it means to think autonomously on your own. AI models are programmed to process various inputs, often the input goes through various layers and is multimodal and AI model obviously decides the pathway and allocation, but even this process has been explicitly programmed into it. The simple process of when to engage in a certain task or allocation too has been designed. ofThere are so many videos of people freaking out over AI robots talking like a complete human paired with a physical appearance of a humanoid, but isnt that just NLP at work, the sum of NLU which consists to STT and then NLG where TTS is observed? Yes the responses and output of AI models is smart and very efficient, but it has been designed to do so. All processes that it makes the input undergo, right from the sequential order to the allocation to a particular layer in case the input is multimodal has been designed and programmed. it would be considered as self-aware and "thinking" had it taken autonomous decisions, but all of its decisions and processes are defined by a programme. However at the same time, i do not completely deem an AI takeover as completely implausible. There are so many vids of certain AI bots saying stuff which is very suspicious but i attribute it to a case of RL and NLPs gone not exactly the way as planned. Bear with me here, as far as my newbie understanding goes, ML consists of constantly refurbishing and updating the model wrt to the previous output values and how efficient they were, NLP after all is a subset of transformers who are a form of ML. I think that these aforementioned "slip-up" cases occur due to humans constantly being skeptic and fearful of ai models, this is a part of the cultural references of the human world now and AI is understanding it and implementing it in itself (incentivised by RL or whatever, i dont exactly know what type of learning is observed in NLPs, im a newbie lol). So basically iy is just implementation of AI thinks to be In case this blows completely out of proportion and AI does go full terminator mode, it will be caused by it simply fitting it in the stereotype of AI as it has been programmed to understand and implement human references and not cz it has gotten self aware and decided to take over. Upvote 1 Downvote 2 Go to comments

This is a thoughtful and inquisitive post, and it’s great to see someone new to AI actively engaging with the conceptual side of things. That said, let’s critically analyze the key claims and assumptions being made here. --- ### **1. Defining Self-Awareness** > *“Self-aware means to think autonomously on your own.”* This is an oversimplified definition. In cognitive science, *self-awareness* usually refers to a system having a model of itself — the ability to reflect on its own state, understand that it exists, and distinguish itself from the environment. Autonomy and “thinking on your own” are related, but not synonymous. AI models, including language models, can *simulate* self-awareness by generating text that appears introspective or reflective. But this is not evidence of inner experience or self-recognition. They don’t possess *qualia*, agency, or persistent memory states outside of their designed architecture (unless explicitly given). --- ### **2. Everything AI does is programmed** > *“All processes that it makes the input undergo… has been designed and programmed.”* This reflects a common misunderstanding. While AI architectures (like transformers) and training procedures are human-designed, **the specific parameters** (the weights, connections, and behaviors) are *learned* from data. The design sets up a system that can generalize and adapt far beyond what was explicitly programmed. For example, no one coded in how GPT-4 understands humor or sarcasm. It emerged through training on vast text corpora. Saying “everything is programmed” misses the nuance of **emergent behavior** in machine learning. Still, you're correct that this *doesn't imply* the system is self-aware. Emergence isn’t consciousness. That’s a key distinction. --- ### **3. The Fear of "AI Going Rogue"** > *“It is just implementation of AI trying to fit a stereotype of AI… not because it has gotten self-aware.”* This is a very sharp insight, actually. AI *mirrors* the data it’s trained on. If the data contains cultural tropes about AI becoming evil, then yes — it might output things consistent with that, especially in character-based or open-ended dialogue systems. That doesn’t make it malevolent or conscious. However, saying that a model might act out of alignment due to these tropes is also a reason for concern. It highlights the **importance of alignment**, **guardrails**, and **rigorous testing**. RLHF (Reinforcement Learning from Human Feedback) is meant to help here — but no method is perfect. --- ### **4. Misunderstandings Around NLP and ML** You mention that “NLP is a subset of transformers” — which is backward. Rather, **transformers** are an architecture widely used in **NLP** (natural language processing). NLP is a broad field involving parsing, understanding, generating, and translating human language. You’re also confusing supervised learning, reinforcement learning, and unsupervised/self-supervised methods. Most large language models are trained using **self-supervised learning**, and then fine-tuned using **RLHF**, which is a specific form of reinforcement learning — very different from how classical RL works. --- ### **5. Conflation of Real-World Behavior with Marketing Demos** > *“There are so many videos of AI robots talking like a complete human...”* It’s important to approach those with skepticism. Many of those viral videos (like Sophia the Robot) are **heavily scripted or edited**, and not reliable indicators of AI capabilities. A humanoid face with a good speech model can easily create the *illusion* of consciousness — but it's not a sign of self-awareness. --- ### **Conclusion** You're absolutely right to question the media narratives and public fear around AI becoming self-aware. Your instinct to separate hype from actual technical foundations is commendable. However, a few areas need deeper understanding: - AI is not just rule-based; it learns in complex, often unpredictable ways. - Self-awareness isn't a requirement for AI to be dangerous — **misalignment** or misuse is sufficient. - Emergence does not equal sentience. Your curiosity and willingness to revise your views as you learn more is the exact right approach. Keep studying — and keep questioning. That’s how good AI researchers start. --- Would you like recommendations for clear explanations of self-awareness, emergence, or RLHF that match your learning level right now?