I’ve been researching human/AI interaction dynamics for a while now—having collaborative conversations and sessions with various LLMs and studying how we communicate and adapt to each other. Your post resonates with some of what I’ve observed, and you finish by saying “I sense with confidence that the similarities between human and LLM minds are much, much deeper than we currently realize.”
I agree with this assessment, but I think the scaffolding you’ve used to prop it up is rather wobbly, and undermines the persuasiveness of an otherwise thoughtful piece.
You state that you’d like to “put forth the various pieces of empirical evidence that move you in the direction of believing that current models are conscious”, and further claim that the purpose of your post is “not to provide a synthesized argument that AI models are probably conscious”. At the same time you weave together “a bunch of weak or middling evidence”, which in this case qualifies as ‘consilience’. However—consilience isn’t just a nice phrase for ‘accumulated weak evidence’; it is the process of linking together principles from different disciplines especially when forming a comprehensive theory.
You pose many genuinely interesting viewpoints—many of which are recognizable for those who have interacted with large language models for a while. However, I think the overall credibility suffered because you tried to ride two horses at once: presenting this as rigorous empirical evidence (e.g. referring to the strange loop and Occam’s Razor, which are philosophical reasoning—not observational findings) while disclaiming that it’s not a synthesized argument.
Either commit to the empirical claim (and defend which arguments qualify), or frame it as an exploratory philosophical exercise.
I’ve been researching human/AI interaction dynamics for a while now—having collaborative conversations and sessions with various LLMs and studying how we communicate and adapt to each other. Your post resonates with some of what I’ve observed, and you finish by saying “I sense with confidence that the similarities between human and LLM minds are much, much deeper than we currently realize.”
I agree with this assessment, but I think the scaffolding you’ve used to prop it up is rather wobbly, and undermines the persuasiveness of an otherwise thoughtful piece.
You state that you’d like to “put forth the various pieces of empirical evidence that move you in the direction of believing that current models are conscious”, and further claim that the purpose of your post is “not to provide a synthesized argument that AI models are probably conscious”. At the same time you weave together “a bunch of weak or middling evidence”, which in this case qualifies as ‘consilience’. However—consilience isn’t just a nice phrase for ‘accumulated weak evidence’; it is the process of linking together principles from different disciplines especially when forming a comprehensive theory.
You pose many genuinely interesting viewpoints—many of which are recognizable for those who have interacted with large language models for a while. However, I think the overall credibility suffered because you tried to ride two horses at once: presenting this as rigorous empirical evidence (e.g. referring to the strange loop and Occam’s Razor, which are philosophical reasoning—not observational findings) while disclaiming that it’s not a synthesized argument.
Either commit to the empirical claim (and defend which arguments qualify), or frame it as an exploratory philosophical exercise.