Semi Random Text Written by a Neural Network
Below is the semi-random text created by a neural network, a souped up digital version of the old Million Monkeys. It is not fully random because the neural network was programmed to create English sentences using words found in the dictionary. It also possesses rules of grammar, and started off on the topic I gave it, namely writing self referentially about random text generation. Otherwise its stream of consciousness output truly is essentially random.
Dakota: Here’s the second line, “One million monkeys banging on a million typewriters will eventually produce the world’s greatest literature.”
Jacques: The world’s greatest literature? It’s not a million monkeys, it’s a billion monkeys.
Dakota: Well, no. It’s actually ten million monkeys, but…
Jacques: Hey, that’s my point. It’s not a million monkeys banging on a million typewriters. It’s the million monkeys collectively collectively, it’s one million people banging on a million typewriters. And those people are likely to have more artistic talent than anyone who wrote anything before them.
Dakota: Well, actually, I’m not sure that’s necessarily true, though.
Monkeys may be pretty great at solving puzzles, but that doesn’t mean they’re able to write novels.
It’s because monkeys don’t care. They don’t care about the one-million monkeys writing the world’s greatest literature. They’re not going to care.
It’s like… it’s not a million monkeys banging on a million typewriters. It’s ten monkeys banging on ten typewriters, but the tenth monkey on the tenth typewriter is going to be much more creative than the ninth monkey on the ninth typewriter.
That’s a different kind of monkey-thinking.
On another note, I wanted to let you know that I didn’t just do this poem, I also designed a whole line of stamps for the Canadian postal system. You know, you can’t really talk about creative monkeys and not mention Canada.
The stamps have monkeys and monkeys on them. It’s probably more like monkeys with monkeys on them. And if that doesn’t sound creative to you, then you’re not creative at all. That is not a thing that happens, by the way, unless you’re a creative monkey, and creative monkeys really aren’t interested in stamp designs, they’re interested in animals. So I’m happy that I finally figured out what I’m supposed to do.
It’s like this: I’m an artist and I have to get paid. It’s not like I can just walk in the door and say, “I have ten thousand monkeys pounding on their typewriters right now, and I have no way of getting a paycheck.”
I had to create a stamp line. I had to make it mean something. If stamp collectors care about my work, then it will mean something to stamp collectors. And if stamp collectors care about their stamps, then it’ll mean something to stamp collectors. But I don’t really have any stamp collectors.
I’m really the kind of creative monkey who comes to work and does creative things, and then doesn’t really have much to do after that.
I think the point of all this is that things are never really random. The monkey typing away at his typewriter may like one key over another; the neural network producing random text may like Canadian stamps. The idea that a million monkeys typing randomly can produce great literature may be an interesting thought experiment, and a dig at pretentious authors, but its parameters are too restrictive to make it practical or actually achieve its stated objective.
I propose that we modify the Million Monkey Experiment by using a million random text generators, geared to producing actual English dictionary words. In fact, we can give the generators a database of the English dictionary, and let them draw words from this dictionary and arrange them at random. To increase the odds of meaningful output we can even make certain more common words such as The, And, I, He, She more common so that the sentences are more likely to make sense rather than random strings of words. And then we can use computer technology to parse the output and detect meaningful sentences.
We already have apps and programs that can analyze a text and look for grammatical errors. It would be no effort at all to change the programs so that they detect things that are grammatically correct, so we can cut through the near infinite rubbish and find the gem hidden in the verbiage.
Is this worth it, or does this move too far away from the original thought experiment? Certainly this output would not be fully random, but rather a form of constrained randomness. But even that is a form of randomness. For example, the roulette wheel only picks numbers between zero and thirty-six. In that sense, the output is not random because the majority of all possible numbers is off limits. Yet the outcome is random, and usually goes against the better.
Even the original million monkey thought experiment contains such constraints. For example it posits that the output will be limited to the keys on a keyboard, meaning that the monkeys must produce letters and numbers and not totally random symbols or mere scratches.
In the same way, I propose that we create a digital equivalent of the Million Monkey Experiment, geared towards producing legible meaningful but purely random text and see where this goes. As the neural network said, “I’m really the kind of creative monkey who comes to work and does creative things, and then doesn’t really have much to do after that.”