The (un)Remarkable Nature of ChatGPT: A Justification Machine in Action
I recently stumbled on a very useful framing that really put ChatGPT in perspective for me. This post is my attempt to share that framing with you, so that you may find it easier to integrate ChatGPT and related technologies into your life.
ChatGPT is a rather remarkable piece of technology that demonstrates the power of a Large Language Model (LLM) to perform tasks that we didn’t expect it to be able to perform. As of this writing GPT-4.0 is the new hotness, and more about it can be found here: https://openai.com/product/gpt-4.
When queried for a blog post paragraph, chatGPT describes GPT-4.0 with the following output:
GPT-4, developed by OpenAI, is a groundbreaking artificial intelligence model that has revolutionized the way natural language processing tasks are performed. As a successor to GPT-3, it builds upon its predecessor’s capabilities, offering even greater performance and versatility. One of the most well-known applications of GPT-4 is the chatGPT model, which is designed to generate coherent and contextually relevant responses to user input. Employing a technique called “transformer architecture,” GPT-4 can understand, interpret, and generate human-like text with remarkable accuracy. This advanced language model has found numerous applications across various industries, including content generation, customer support, and virtual assistants, among others. Although GPT-4 is undeniably powerful, it is essential to remember that it is still an AI model and not a sentient being, which means that it does not possess emotions or beliefs like humans do.
With that out of the way, what is a Justification Machine? For that, we’ll need a little bit of context and a summary of the Justification Systems Theory (JUST).
Justification Systems Theory (JUST)
Gregg Henriques’ Justification Systems Theory1 consists of three claims:
1. With the evolution of language and ability to utter propositions (declarations) there came the ability to question those propositions. This question-answer dynamic, when framed as the adaptive problem of justification, provided selective pressures that shaped the evolution of human self-consciousness and human culture.
2. The nature of human consciousness can be framed by the Updated Tripartite Model with three domains of (a) the experiential self, (b) the private ego, (c) the public persona.
3. Humans are transformed into Persons by being socialized into systems of justification, such that they operate on a new complex adaptive plane that can be framed as Culture.
To provide some context for the claims, consider the following exchange in the context of claim (1), that our language evolved in the context of the adaptive problem of justification:
- “You are late.”
- “I am not.”
- “It is 8:02 am, you were supposed to be here at 8:00 am.”
- “Your clock is fast, my clock says it is 7:58 am.”
As Henriques states in his book: “Take a moment and think about how often you are describing and explaining yourself to others or listening to the accounts of others and pondering their justifiability.”
Regarding claim (2), the Updated Tripartite Model and the three domains framing:
The Experiential Self is our personal embodied sense of being in the world coupled with our idea of self that sits at our emotional core.
The Private Self, our narrator or ego, is a self-reflective justifying position that internally narrates what is happening and why. This internal narration is what separates humans from other animals. Private Self operates at a language level and within the context of Culture, figuring out what is justifiable and what is not.
The Public Self, our public persona, manages our social roles and impressions. The Public Self is what is observable by others within the context of Culture.
As to the claim (3), consider an interaction with a child whose face is covered in chocolate:
- “Did you eat the chocolate?”
- “Then why is there chocolate all over your face?”
This episode above demonstrates the every day training that is taking place, demonstrating to the child that their justification was insufficient to account for all the evidence that is available. Slowly, over time, children become Persons as they end up with justifications sufficient enough to fit into the Culture they’re a part of.
We are now ready to describe a Justification Machine.
A Justification Machine
In the framing of Justification Systems Theory, what can we say about ChatGPT?
ChatGPT is remarkable in that the problem it optimizes for is prediction of what word comes next after being presented with a sequence of words.2 In order to perform this task, ChatGPT is trained on an enormous amount of data, most of it being human language. ChatGPT then, based on this input, learns human language patterns and can generate human language-like utterances.
The key frame shift to make here is to recall what human language is in the frame of Justification Systems Theory: public justifications uttered by humans shaped by thousands of years under selective pressure to be the most plausible justifications to date. Specifically, it is justifications uttered by the Public Self of all humans contributing to the Internet. Note that Private Self and Experiential Self are not accessible as training data. Therefore, unfiltered self-narratives or feelings of embodied self are not inputs into ChatGPT training. Only Public Self data is available.3
The unremarkable thing about ChatGPT is that it is just a Justification Machine emulating the patterns of a Public Self learned from the entire human Culture available on the Internet. Most certainly, ChatGPT does not have an Experiential Self. Also, it is unlikely, for now, that it has a Private Self.4
Seen from the JUST framing, ChatGPT is able to generate justifications for anything, because it is precisely what its unintentional goal is, to make the most plausible justifications. In biology and psychology there is a concept of supernormal stimulus, which is a stimulus “that elicits a response more strongly than the stimulus for which it evolved”.5 The concern for us humans is that AI/ML rapidly can reach superhuman capabilities on specific tasks, and ChatGPT is likely to continue this trend. As such, our interactions will soon be with a superhuman Justification Machine, where each utterance is expertly tailored to be a more compelling justification than any real human one.
Why is the sky made of chatGPTs?
Hopefully you found this framing helpful in illuminating GPT’s place in our human culture, what it will do going forward, and perhaps as a result you have a better idea of how to deploy this tool more effectively. I leave you with your thoughts and some GPT-4.0 justifications:
Prompt: In few short sentences, explain why the sky is blue
The sky appears blue due to a phenomenon called Rayleigh scattering. When sunlight enters Earth’s atmosphere, it scatters in all directions, but shorter blue wavelengths scatter more effectively than longer red wavelengths. As a result, the sky appears predominantly blue to human observers.
Prompt: In few short sentences, explain why the sky is red
The sky appears red during sunrise and sunset because of the way light scatters through the Earth’s atmosphere. As the sun sits low on the horizon, sunlight passes through a thicker layer of the atmosphere. Shorter blue wavelengths scatter more effectively, leaving the longer red and orange wavelengths to dominate, causing the sky to appear red.
Prompt: In few short sentences, explain why the sky is black
The sky appears black at night because the sun is no longer illuminating the Earth’s atmosphere from our vantage point. As a result, there is minimal scattering of light, allowing us to see the darkness of space. The absence of sunlight causes the sky to be void of colors and appear black, revealing stars and other celestial objects.
Prompt: Assuming, hypothetically, that the sky is made of chocolate, explain in a few short sentences why. Write a plausible explanation without using adjectives implying hypothetical or fictional nature of the setup.
The sky appears to be made of chocolate due to the presence of cocoa particles suspended in the atmosphere. Sunlight interacts with these cocoa particles, scattering the light in a manner that creates a chocolate-like appearance. This unique atmospheric composition results in a sky that looks and smells like chocolate.
Prompt: Assuming, hypothetically, that the sky is made of chatGPTs, explain in a few short sentences why. Write a plausible explanation without using adjectives implying hypothetical or fictional nature of the setup.
The sky appears to be made of ChatGPTs due to an advanced network of interconnected artificial intelligence modules that form a virtual canopy over the Earth. This network, designed to monitor and analyze global data, utilizes holographic technology to create a visual representation of the ChatGPTs in the sky. The resulting appearance simulates a sky comprised of ChatGPTs working in unison.
1. Henriques, G. (2023). A New Synthesis for Solving the Problem of Psychology: Addressing the Enlightenment Gap. Palgrave Macmillan.
2. A thorough, but accessible, illustration of the concepts involved can be found in Stephen Wolfram’s “What Is ChatGPT Doing … and Why Does It Work?”.
3. While there are other AI models that can predict things like someone being pregnant based on their purchasing habits, or ability to profile a person’s behavior based on their browsing habits, none of those models work with anything but Public Self data. Those models still only infer from what we offer as our Public Selves.
4. Although, given that (in the JUST framing) the Private Self co-evolved to enhance Public Self capabilities, future versions of GPT, in their quest of making more compelling justifications uttered by the Public Self, may eventually stumble upon architectures or solutions that begin to emulate a Private Self.