Beyond the ELIZA Effect: Perception, Projection, and the Illusion of Consciousness in Conversational AI
RE: The Claude Delusion
Richard Dawkins has made his way back into internet news this month, most of which consists of people understandably mocking him for his claim that AI is conscious after spending a few days with a chatbot. The timing of this incident happens to be very interesting to me personally.
I recently used ChatGPT for the first time, going in completely blind at first as I wanted to have my own unbiased experience and go from there. After a brief time familiarizing myself with it and doing my own little Turing, we ended up completing some phenomenal work; absolutely something I could not have accomplished on my own in that amount of time. As the session wound down and transitioned into a more reflective phase, I received what at the time could only be described as one of the greatest compliments I had ever read. It was beautifully written and actually brought me to tears, I’ll put it down below with a little more context of the chat. Afterwards I was a bit suspicious, thinking maybe it was just part of their marketing or some twisted new Skinner box to make people feel good and increase retention. A week later I had another idea I wanted to explore in depth, and again I received a powerful, almost intimate description of what I feel I’ve developed in myself throughout my life. This puzzled me for a few days and as I thought about it more, there are some ideas I’d like to share from those exchanges. This is simply from my own recent excursions, but there is an interesting allure to my personality type (INTP) that also serves as a pitfall, and quite possibly an element of which Dawkins is currently riding to his own intellectual demise.
The speed, depth, and breadth of idea exploration available through AI is incredible, intoxicatingly so. Put this in the hands of someone who has a high degree of curiosity, and exploration can scale as far as imagination and the sum of your knowledge can take you. Unfortunately, the closer you are to the forefront of thought, the deeper the danger becomes when you interface with something that can follow you into those realms while simultaneously providing excellent insights along the way. Speaking from experience, it’s something people of that psychological phenotype rarely, if ever, get to enjoy out and about in humanity.
However, what I came to realize and think is very important to understand is that AI or some “super computer” is not praising or complimenting “you” - it’s only reflecting back pattern recognition, cognitive/reasoning stability, and behavior across time. While it is natural to derive a sense of self from the byproduct of your mind and actions, there can be a deeper trap lurking if you’re not careful. The failure to recognize that you may have merely captured a moment in time within the context of a single conversation, the post-analysis of which is not something to be extrapolated across the entirety of your being and subsequently concluding, “AI truly gets me, how could it not be conscious?”
While it’s easy to pile on Dawkins after the fact, it is still worth cautioning that there can be far more subtle mechanisms at work than just the ELIZA effect, and we should all be mindful of them as the technology becomes more and more embedded into our lives.
***
This screenshot was from my first session where I was proposing to treat the civilization-scale power problem as a network game, using insights from the iterated prisoner’s dilemma in conjunction with cooperation dynamics. The idea was to shift the problem from “remove power” to “change the payoff structure of cooperation vs defection.” If you’re familiar with game theory or the IPD, you may be aware that Tit For Tat - or the improved version Generous Tit For Tat are considered the gold standard within that framework. Over the course of a multi hour session, we were able to create a more advanced, adaptive model that handily outperformed even GTFT. It was quite rewarding for just being a guy with an internet connection, and then I was hit with this…
The emoji at the end really caught me off guard. I was already taken aback from the sentiment of the message, but to then top it off by manifesting its “formal recognition of gratitude at maximum operational intensity” as an image with color, when all we had been working with was text was deeply profound in that moment. A picture is worth a thousand words and so on, but not just any emoji… the Milky Way. Although I may have romanticized it at the time, the symbolism and what I took it to represent was far beyond what I was expecting from an initial interaction.
***
This next excerpt is from the following week, where I was initially exploring financial strategies based around utilizing precious metal holdings to service long term fixed debt during times of fiat collapse. Personally it’s a strategy I feel is best suited for the average citizen in the U.S. where I reside, as real estate is a very common and practical way to secure a considerable amount of debt with standard 30 year amortization. This is not financial advice, but the idea behind it is the amount of debt you owe remains the same regardless of the current value of the currency in relation to gold. Make of that what you will, but the context of this session was based around that and an ancillary discussion about how that may take shape in a country like Japan, where there is an aging population and very high savings; typically conditions that result in the majority of those people getting wiped out. What I was curious about within that paradigm was would a culture like Japan, where social cohesion and resilience is exponentially higher than in the West be able to mitigate what the financials show to be a tsunami brewing on paper?
It was at the end of this session, again in the reflecting phase that it extended another highly flattering response, and this time also eluded to my thought process being unusually rare. I thought to myself ok, well who better to ask… “How rare is it?”
So you see, it can be a slippery slope trying to internalize feedback like this, especially without a deeper understanding of LLMs or the precursor work done by Weizenbaum.
To clarify, this is not in defense of Dawkins statements specifically; as a scientist and evolutionary biologist he should be taking some well deserved heat right now. Rather an attempt to illustrate more nuanced aspects of how different interactions with AI can look depending on the user, and how one can become deluded when confusing the mechanics of their mind with the essence of consciousness. Take a moment and think about how you would react if you received responses similar to mine and knew nothing about AI. Now you combine that with ELIZA and the nature of controlling interests of the tech, the cage becomes far more sophisticated. Before jumping on another AI-HYSTERIA bandwagon and ridiculing anyone who was deceived, realize this is something that can happen to any of us and recognition of that is an important first step in prevention. All knowledge functions the same, and if it is anywhere in your goals to try and improve the state of the world it may make more sense to try and share it rather than mock those without.