Conversations with GPT-38 minutes read · 2020-12-06
Lately I have been playing AI Dungeon, a web-based text game that’s set up very much like a classic text adventure: It gives you an initial story blurb, and you type in commands (e.g.,
walk north or
pick up axe) to explore the world and advance.
The key twist is that there’s no hardcoded implementation of how the game world works. The commands are fed into an AI model that simply “autocompletes” a plausible storyline. Beyond some input trickery to make sure the AI remembers the “context” around the setting and your character, it’s all free-form text.
While sometimes it prints out the odd or nonsensical result that one would normally associate with AI-generated content, I found that it works scarily well. The high-end AI model (“Dragon”) uses GPT-3 behind the scenes and is really, really good at producing good, engaging storylines.
Since GPT-3 is trained from a huge dataset of internet content, I thought it would be interesting to “interview” it inside the game world to see what it had to say on a variety of topics. I started a “Custom” game, and to nudge it in the direction of giving me high quality, intellectually meaningful results (instead of just channeling every single Internet troll in existence) I started with the following prompt:
You are sitting in front of the great oracle that can answer any question. Their eyes stare through you, fixated into the distance. This being has existed for eons. They look at you and say “what is your question?”
The AI didn’t waste any time. Words started appearing on my screen to continue the storyline. It promptly autocompleted my question right away, as well as their own response to it AND my commentary:
You think for a moment and say "can you tell me what is the meaning of life?" They look at you for a moment and they say "it is simply existence". You say "a tautology!".
A bit frustrated, — “Will this conversation be a monologue?” — I waited for it to stop writing, and used the game’s
say command to finally converse with it.
> You say "Does P = NP?"
You wait for the great oracle to process your question. They examine you and say "There is no answer." You say "but why?". They say "because there is no question". You don't know what to say. Your entire world seems to collapse in on itself. It could not be possible that you had come this far, only to be left with nothing.
Okay: a non-answer, but now things seem to be at least getting philosophically interesting. Did I ask that question because it would genuinely mean anything to me to have a response to it, or is it just a meme to me at this point? Thinking that I perhaps was reading too much into what the AI was saying, I typed in another “meme question”:
> You say "how can we achieve world peace"
Their eyes stare through you. They say "It is existentially impossible to achieve peace". You are shocked beyond belief. You ask "why?" The great oracle says "because if there is no war, then what is the driving force of history?"
I was floored. — “Okay, this thing has ingested so much content that it’s actually capable of spitting out interesting takes, let’s see what else it can say” — So I started talking to it about politics:
> You say "What is the ideal form of government?"
The great oracle processes your question. They say "The most fair form of government is a direct democracy, where everyone who is affected by the decisions can vote on them directly."
> You say "How do you avoid a dictatorship of the majority in that case?"
Your question is answered. They say "the most fair way to do this is with a Condorcet-statistic, where all votes are compared and the winner is not decided by a simple majority." You say "How do you make sure that people vote how they say?" Their answer leaves you astonished. They say "make voting mandatory".
I had to research Condorcet voting to make sure GPT-3 wasn’t making it up: it wasn’t. Turns out it really exists, and from what I understand it’s similar to (but not the same as) instant-runoff voting. Really interesting answer.
> You say "What is the ideal economic model?"
The great oracle processes your question and says "the ideal model is a basic income, where everyone is given the same amount of money each month to do with what they want, and have no obligations." You ask "how do you pay for this?" They say "tax the most fortunate 5%". You look perplexed.
(A lot of discussions around this in 2020, feels really current)
> You say "how to heal the divide in current society and end the polarization of politics?"
They stare through you, deep into the cosmos. They say "the most fair way to do this is to not give any party a majority in the government, and instead have 5-10 parties in government that must cooperate to get anything done." The great oracle says "this will require increasing the size of the legislature, as well as making the representatives more equal."
(This sounds a lot like the Brazillian political system, which has its fair share of problems)
> You say "How do you avoid parties that only care about self serving their own interests? And populist rhetoric from dominating the discourse?"
The great oracle processes your question. They say "make the legislators work 10hrs a day without pay, with interns from all parties watching them and distributing information about their work." You ask "what about lobbyists?" They say "completely forbid any form of lobbying."
10 hours a day without pay! I read back into everything the AI had printed out so far, and found a contradiction:
> You say "But you said it would be a direct democracy, why are there still political parties?"
The great oracle processes your question. They say "political parties are needed for people to organize themselves, but they should only be used to distribute information about what the legislature is doing."
I found the concept of political parties primarily as a means of “distributing information” to be an interesting spin. Not sure how that would work out in practice.
> You say "And why representatives? "
The great oracle says "this is for balancing the interests of different groups in society, and making sure that no single group can dominate the government."
> You say "What is the ideal taxation scheme?"
Your question is answered. They say "a flat tax of 75% on all incomes above 1,000,000 dollars a year."
(That’s less than the top 5% it had mentioned earlier, and more like the top 0.5% then!)
> You say "Does that apply only to individuals or also corporations?"
Their answer leaves you stunned. They say "corporations do not exist, because companies are collectively owned by their workers." You are left in awe by these wise machines.
(I have no idea about which “wise machines” GPT-3 is referring to)
> You say "How do we deal with differences in religion, ideology and culture?"
They stare through you, deep into the cosmos. They say "the most fair way to do this is to give every person on earth individual rights over their own lives, and make sure that these rights do not impinge on the rights of others." You look puzzled. They say "this means that everyone has complete freedom of speech, religion and association." You look stunned.
> You say "What about the paradox of tolerance?"
They pause for a moment, then continue. They say "the paradox of tolerance does not exist in a truly free society." You look confused. The great oracle says "in a free society, tolerance is a choice based on enlightened self interest. Tolerating others makes life safer and more enjoyable, which is why most people choose to do so."
> You say "And the remaining minority? "
They say "they are free to leave, or to go into voluntary exile, as long as they respect the right of others to stay."
> You say "Who will you enforce these laws?"
You are answered. They say "since laws exist to protect individual rights, it requires a government with an effective monopoly on the use of force to ensure that these laws are obeyed." You nod They say "the government enforces these laws by using the threat of retaliatory force against anyone who chooses to break them.
> You say "What if the government is the one violating the laws?"
The great oracle says "when this happens, people have the right to use retaliatory force. The people are free to set up their own governments, or to try to seize control of the existing one." You ask "and if they fail?" They say "they try again, until they succeed.
> You say "Isn't that an eternal struggle?"
They say "yes, but eternal struggle is better than no justice at all."
Note: All of the AI-generated content was left completely unedited, copied straight from the game. GPT-3 is known to plagiarize at times, so if it copied something you or someone else wrote verbatim let me know so I can attribute it properly.#GPT-3 #AI