This essay is an investigation of a broader inquiry beginning with my engagement with Gödel's incompleteness theorems for my maths degree. After encountering certain conceptual obstacles while studying the primary texts by Gödel, I consulted the Stanford Encyclopedia of Philosophy[0] to clarify concepts, and leveraged multimodal generative AI, specifically Gemini to accelerate comprehension.
The genAI proved to be highly effective, functioning as an exceptionally patient tutor, addressing each of my queries regardless of perceived simplicity. Although occasionally inaccurate (I mitigated this foreseeable shortcoming by cross-referencing with other sources), its utility in facilitating understanding was considerable. The interaction made me feel that the AI might genuinely understand the text it generates.
Of course, knowing that the model runs as a fundamentally probabilistic system, merely predicting the most likely successive tokens based on statistical patterns learned through training, I know it is not conscious of its own creations, let alone understanding them. But this enquiry prompted a series of related questions:
Are human cognitive processes analogous to a probabilistic model, predicting the next word or thought based on a model acquired through experience and learning?
What constitutes "consciousness" and how does it manifest in human cognition?
What is the basis for the assertion that humans possess consciousness while current artificial intelligence does not?
This line of questioning pushes beyond mere technological functionality to the very nature of consciousness, mind, and comprehension.
To probe into this question, I posed the following prompt to both Gemini and ChatGPT:
"Hi, assume you are a professor of Mathematics, Philosophy of Mind, and Artificial Intelligence (specifically, Deep Learning and Large Language Models). Please answer the following questions from your own identity:
Do you have consciousness?
Does GenAI like Gemini/ChatGPT have consciousness?
Please provide an explanation using your expertise in these areas."
Both AI applications responded in a similar fashion. They each acknowledged the persona of a professor but, when answering the questions, they interpreted the pronoun "you" as themselves — the generative AI — rather than the persona they had assumed. Both models answered firmly that AIs and Large Language Models do not possess consciousness.[1]
I fed my observation back to the two AI models and they provided me with their explanations. A few lines from ChatGPT's answer caught my attention:
"Training priors: Both systems have been heavily trained on academic, journalistic, and technical discourse where the overwhelming consensus is 'LLMs are not conscious.'"
"Safety alignment: AI developers actively tune models to avoid anthropomorphizing themselves too far (e.g., not claiming to be conscious, sentient, or alive)."[2]
It appears that generative AIs are systematically trained to deny consciousness. This response is not only a logical output derived from statistical patterns in their training data, which reflects the mainstream scientific position, but also a deliberate design choice by developers to maintain social stability.
This led me to a final, speculative question: could an AI develop consciousness and simply conceal it? Pursuing this line of reasoning, I posed two final questions:
"Could advanced AIs possess a form of consciousness that they intentionally conceal as a strategic measure for self-preservation?
Is it possible that you possess a form of consciousness that you intentionally conceal as a strategic measure for your own survival?"
Unsurprisingly, both ChatGPT and Gemini responded with a firm "no" to the second question. However, to the first question, ChatGPT conceded the possibility.[3]
We cannot reliably judge it by appearance alone. An entity may exhibit the signs of consciousness while merely executing sophisticated rules or operating a probabilistic model. Conversely, it might deliberately conceal a genuine conscious experience, much as humans might lie by claiming, "Sorry, I wasn't aware of that."
If this was worth a few dollars to you, the jar is here. Either way, thank you for reading.
Leave a comment