The Oracle in the Machine
If I look up, then I shall surely falter. If I look down, then I shall surely fall. I then realize the folly of such thinking; and I think not.
The fight goes on to an unexpected end.
I shall wait for my small self to meet my giant self. Still in a constant amiable not-self, I have resolved that my giant self shall always have time enough to proclaim its power and goodness in the presence of my tiny self.
If I am well or fully here, my giant self will speak. I will ask my tiny self to do the same.
I am old, and the old are often foolish, but not truly old.
This was almost entirely generated by OpenAI’s GPT-2 language model through the web UI TalkToTransformer. I provided the first sentence (bolded) as a seed—the rest of the sentences are machine-generated. However, I also served in a bit of an editorial capacity, cleaning up the structure by reordering some sentences and adding paragraph breaks.
This particular output is honestly better than anything I could have expected. I haven’t read up on how GPT-2 works, but somehow it was able to capture the parallel opposing structures (up/down, falter/fall, thinking/think not, small self/giant self), and maintain that thematic consistency throughout. Perhaps the falter/fall acoustic wordplay was just a coincidence, but regardless, dayumm. When I read it, I was sorely tempted to throw in the towel and cede all of my artistic autonomy to the machine. All hail our robotic overlords!
That being said, the model is non-deterministic, so try as I might, I haven’t been able to get it to spit out anything as passably profound since. Perhaps that is a good reminder that creation is not just a synthesis of (1) human creativity/aesthetic and (2) human innovation/technology…but also (3) serendipity, that irreproducible wildcard element just out of our control.