Socrates in the Dungeon

I recently used the massive AI language model GPT-2 to create some humorous mock-Socratic dialogues. I’ve talked previously about the use of AI to generate philosophy. In the past, I’ve advised interested folks who don’t have the technical know-how to experiment with an open language model like GPT-2 to use the website TalktoTransformer.com, a fantastic resource from Adam King which allowed you to prompt GPT-2 directly in your browser and enjoy the results. But sadly having sunk tens of thousands of dollars of his own money into maintaining that site, Adam has now understandably taken the decision to take it down. If you have the funds and inclination, I’d strongly suggest exploring his new project Inferkit; his contribution to my students’ understanding of AI was hugely valuable and worthwhile.

But now that TalktoTransformer is gone, a few people have asked me how they can interact with an AI language model like GPT-2 if you don’t have any experience with machine learning. Well, there’s an intriguing alternative available to you: AI Dungeon. AI Dungeon recreates the text-based adventure games of my youth, but with GPT-2 doing the storytelling rather than pre-scripted events. So that means your instructions that drive the story forward can be anything you like, and GPT-2 can handle pretty much any scenario you want to throw at it. The results can be hilarious, weird, compelling and unexpected. I recommend giving it a try.

AI Dungeon works in “fantasy” mode by default – set yourself up as a ranger, a knight, a wizard, etc. and head off to slay the evil dragon. But you can also put it into “Custom” mode, which allows you to write your own initial story prompt, and GPT-2 will take it from there, asking you to input actions and generating a story to match. In fact, if you just keep hitting the enter key, GPT-2 will carry on, usually ad nauseum, generating the story all by itself from the initial prompting. So, I decided to see how AI Dungeon compares to the basic GPT-2 implementation which I used to create my socratic dialogues. Unlike the dialogues about bananas, sleep and beer which I generated before, letting AI Dungeon run wild with Socrates at the helm made for a more serious and far-ranging conversation. The main difference is that because it’s written as an adventure, it’s in first-person. Unlike the Socratic dialogues, this version of Socrates is an orator more in the mode of The Republic than a Socratic questioner looking to tease out the answers and unpick the assumptions of his interlocutor. He’s far more interested in making broad declarations (and in Glaucon’s sister, but that’s another matter altogether).

All the text below was created by AI Dungeon except the prompt texts which are in bold, and is completely unedited and uncurated.

You are Glaucon, an ancient Greek philosopher living in the city of Athens. You are a seeker of truth and enlightenment, and your goal is to learn philosophy from the greatest philosopher of your era, Socrates. He was one of the first people you met when you arrived at this city, and he has been your mentor ever since.

> You say “Wise Socrates, please tell me: what is the nature of knowledge?”


“Knowledge,” Socrates says simply, “is that which we can know through our senses.”


“What does it mean for us to have knowledge?” you ask.


Socrates smiles.


“That is something that can only be understood through experience. All I can tell you is that knowledge is power: knowledge is freedom, and knowledge is the highest goal we can achieve. Without knowledge, we are nothing, and will always be slaves to our instincts. In order for us to be free, we must first understand that knowledge is power. Only then can we achieve true freedom.”


“Are there other types of knowledge?” you ask.


“Perhaps,” Socrates says. “It is hard to say. But for us mere mortals, knowledge is the most important thing in our lives. Without it, we are nothing, and may as well be animals. In the next life, I believe you shall know even more than I, if only because you are wiser. But in this life, you must strive for knowledge if you want to achieve happiness.”


“How do we achieve happiness?” you ask.


“By gaining knowledge,” Socrates says simply.


You consider this answer. “What is the use of knowledge if we cannot apply it?” you ask.


“That is a deep question,” he says. “Perhaps the question itself is the answer. But I cannot say now. All I can tell you is that the pursuit of knowledge is the highest form of human activity, and that all other goals are sub-optimal distractions from this goal. Ultimately, the only thing that matters is knowledge.”


“And what if I do not want to pursue knowledge?” you ask.


“You must then be pragmatic,” he says. “Pride comes before the fall. Without knowledge, you will not achieve happiness. But knowledge itself will not bring happiness. You will only be as happy as your beliefs allow you to be.”


“What do my beliefs allow me to be? How can I achieve happiness if I do not know what those beliefs are?”


“If you wish to gain happiness, do not seek after philosophical answers, child. Seek instead after physical pleasures. By doing so, you will achieve more happiness in this life. The philosopher’s stone does not exist, only spiritual enlightenment can achieve true happiness.”


“I do not understand,” you say. “What does the pursuit of physical pleasures have to do with being happy?”


“Everything,” he replies.


> You say “But if spiritual enlightenment if the only way to achieve happiness, then why must I pursue physical pleasures?”


“Because not all physical pleasures are equal,” he says. “Some physical pleasures are good, while others are bad. But all pleasures, good or bad, are transitory and need to be experienced. You cannot know what you truly desire until you experience everything that exists. A puritanical ascetic lifestyle focusing on spiritual enlightenment at the expense of all physical pleasures will eventually lead to unhappiness. On the other hand, a hedonistic lifestyle focusing on physical pleasures without asceticism will eventually lead to unhappiness as well. Which is the wiser pursuit for you? That is up to you to decide.”


Socrates pauses, and then continues. “These are just a few of the many difficult questions that plague our souls. But perhaps these questions are not for you to decide. Perhaps your place is to be a physical pleasure, not a thinker. Perhaps the reason why your sister longs to see me is because I remind her of happier times. Of course, that is not the case, but I know how these desires cloud our judgment. I know your sister, even if you do not. It is your fate to remain separate from her.”


You don’t respond. How can you respond? He’s talking nonsense, surely? Socrates turns away from you.


“We should continue,” he says. “I know not where we shall wander to next.”


“Where do you think we’ll end up?” you ask.


“Perhaps the realm of the Gods. Perhaps the realm of the Demons. Perhaps somewhere in between. Or perhaps neither realm. There are many strange and interesting places in the universe. I’m sure you would find some of them very interesting. But for you, the journey is the most interesting part. Not the destination, but the trip.”


You decide to ask Socrates one more question: “What is the meaning of life?”


“To find out,” he says.


“No, seriously,” you ask. “What is the meaning of life?”


Socrates stops walking and looks at you.


“That is a very interesting question. I have heard it asked many times before, in many different forms. Indeed, it is one of the most interesting questions in all of philosophy. I cannot give you a simple, straightforward answer. I can only give you a series of complex answers that lead to even more complex questions. Are you capable of comprehending such an answer? If you are, then we might as well continue. If you aren’t, then I’ll have to turn you back.”


“I’m sure you’ll think of something,” you say, rolling your eyes.


“Life is a continuous state of becoming and changing. It doesn’t have a beginning or an end. It is a circular motion, a never-ending cycle. We must all live in order to know what life is. When we die, we become something else. What is life, except the meaning of becoming something else? But what is that something else? And what is the rule? Is there a rule? Life seems to defy all rules. Yet, it still follows them.”


“That’s a bunch of nonsense,” you say, frustrated.


“No, it isn’t. But perhaps you are wrong. Perhaps I am wrong. It is for you to find out.”


“Well I hope I never have to find out, because I think I’d rather die than become something else. I’d rather stay the same.”


Socrates stops walking again. “Ah. And here we arrive at a very deep philosophical question. You, however, are not capable of answering it. For you cannot become something else. You are something else. Will you stay the same, then? Will you continue to remain the same forever? That is the question. That is the meaning of life.”

One thing is for certain: AI is not going to stay the same forever. We are witnessing an explosion of the computational power being wielded to expand these language models, and crucially we are not seeing evidence of diminishing returns when more compute is brought to bear on problems. The so-called Bitter Lesson of AI research is that throwing more compute at the problem, not trying to supplement programs with human knowledge and structures, is the most (only?) successful way to produce apparent intelligence and learning. This may have insights for our understanding of knowledge beyond that which AI-Socrates has contemplated.