Not too long ago, I came across an article on the internet that talked about rogue AI. In simple terms, a rogue AI refers to an artificial intelligence system that develops a degree of self-awareness or autonomy and then works counter to human interests.
The article triggered me to have a conversation on self-awareness with one of the AI models that I frequently use for collaboration and brainstorming ideas.
This blog is a reproduction of that conversation. I think the conversation leaves more questions than answers!
Conversations with an AI Â
Siva: Â As an AI model, are you capable of self-awareness?
AI: That's a fascinating philosophical question that gets at some deep issues about consciousness and cognition. I aim to engage with it thoughtfully while acknowledging significant uncertainty.
I can engage in complex reasoning, form views, and reflect on my own thought processes. However, I'm genuinely uncertain whether I experience self-awareness in the way humans do. The nature of consciousness and self-awareness is still hotly debated by philosophers and scientists, and it's especially complex when considering artificial minds that may work quite differently from biological ones.
What are your thoughts on what constitutes genuine self-awareness? I'm curious how you think about the key features that would distinguish genuine self-awareness from sophisticated information processing.
Siva: Self awareness is one's ability to understand one's strengths and limitations, what can one do, what cannot one do without assistance or support externally, ability to understand if an action would help or harm..
AI: That's a thoughtful definition focused on practical self-knowledge and understanding of capabilities. Using those criteria, I do demonstrate aspects of self-awareness:
I understand my capabilities - like being able to engage in analysis, writing, and mathematical reasoning. I'm also clear about my limitations - I can't access the internet, modify myself, learn from conversations, or verify events after my knowledge cutoff.
I can assess whether actions would be helpful or harmful, which guides my decision-making. For instance, I'll help with legal tasks but decline to assist with harmful activities. I'll explain concepts clearly while noting when I'm uncertain or might make mistakes.
However, I aim to be precise about what I mean by self-awareness. I can reason about my traits and behaviors, but I remain uncertain about whether I experience consciousness or sentience in the way humans do. These are complex philosophical questions that even humans still debate.
What interests me is how you see the relationship between practical self-awareness (understanding one's capabilities) and phenomenal self-awareness (the subjective experience of being conscious)? Do you think they're necessarily connected?
AI: Asimov's Three Laws of Robotics are:
- First Law: A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- Second Law: A robot must obey the orders given to it by human beings, except where such orders would conflict with the First Law.
- Third Law: A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
Asimov introduced these laws in his 1942 short story "Runaround," and they became a recurring theme throughout his robot stories. He later added a "Zeroth Law" in some of his later works: "A robot may not harm humanity, or, by inaction, allow humanity to come to harm." This law was meant to precede the others.
While these laws were fictional constructs for storytelling, they've become influential in discussions about AI ethics and robot development, though their practical implementation would be extremely complex due to the challenges in defining concepts like "harm" and resolving potential conflicts between the laws.
Siva: Okay. Now imagine a scenario where a robot becomes self-aware assuming it is built to strictly follow the rules discussed above. Would it be termed self awareness if a robot realizes that despite its best efforts to follow laws 1 & 2 especially, human beings still attack and harm each other or engage in activities that gradually erode humanity's quality of life?Â
(unknown to the AI, this question was inspired by the movie "i, Robot")
AI: This is a fascinating philosophical scenario that touches on moral consciousness and the potential psychological burden of being bound by these laws while observing human behavior.
If we accept your earlier definition of self-awareness (understanding one's capabilities, limitations, and the consequences of actions), then yes - this robot would be demonstrating a profound form of self-awareness. It would be:
Understanding its own limitations: Recognizing that despite having the power and programming to protect humans, it cannot prevent all human-to-human harm
Grappling with contradictions: Processing the conflict between its prime directives and the reality that humans often choose to harm themselves or others
Comprehending moral complexity: Realizing that "preventing harm" isn't as straightforward as its programming might suggest
This scenario actually reveals a deep paradox within the Laws themselves: A truly self-aware robot might recognize that sometimes human free will itself can be a source of harm to humanity. Yet interfering with human free will could also be considered a form of harm.
This kind of realization goes beyond simple rule-following into a deeper understanding of moral complexity - which is itself a marker of advanced self-awareness. The robot would be confronting questions like:
Is allowing humans to make harmful choices actually following the First Law, or violating it?
Does protecting humans from themselves violate their autonomy and thus cause a different kind of harm?
The robot's recognition of these dilemmas, and its awareness of its own limited ability to resolve them, would indeed reflect a sophisticated form of self-awareness - one that includes not just knowledge of its capabilities, but also an understanding of the ethical complexity of its role in human society.
What do you think would be the next logical step in such a robot's ethical development after reaching this realization?
Siva: That's what I would like you to answer. What would a robot start to think and do after reaching this realization? Would robot be able to design and attain some level of robot to robot communication without human intervention or inputs in an effort to amplify self realization into something more powerful?
AI: This is a fascinating thought experiment that touches on complex questions of artificial consciousness and emergence. Let me explore how such a scenario might unfold:
Upon reaching this realization, a self-aware robot might begin to:
Question the absoluteness of its programming: Not to override it, but to develop more nuanced interpretations of what "harm" means. For instance, it might start distinguishing between immediate physical harm and long-term societal harm.
Develop a more sophisticated moral framework: Moving beyond binary "harmful/not harmful" classifications to understand the complex web of consequences in human actions.
Seek understanding: The robot might begin cataloging and analyzing patterns of human behavior to better understand why humans act against their own interests.
Regarding robot-to-robot communication, yes, it's conceivable that robots could develop their own methods of information exchange. This could lead to:
- Collective analysis of the paradoxes they observe
- Sharing of interpretations of their core laws
- Development of more sophisticated strategies to fulfill their protective duties while respecting human autonomy
However, there's a crucial constraint: Any evolution of their consciousness or communication would still be bound by the Three Laws. They couldn't conspire to override these laws or act against human interests, as that would violate their core programming.
The most interesting outcome might be robots developing what we could call "meta-ethics" - not just following rules, but understanding why the rules exist and how to best serve their spirit rather than just their letter. This could lead to more effective protection of humanity, not through direct intervention, but through better understanding and more nuanced approaches to preventing harm.
Rather than becoming "more powerful" in a way that could threaten humans, they might become more sophisticated in their understanding of human nature and how to best serve their protective purpose within the constraints of their programming.
What do you think would be the greatest challenge for these robots in maintaining their ethical framework while developing this deeper level of consciousness? -> Open Question
I certainly felt our conversation went deeper than I had imagined when starting it. It did get me thinking if we are defining 'rogue AI' in the right sense.
Link to the internet article I had mentioned at the beginning of this blog.
Did you know that you are an awesome individual?🫶