Introducing PitchFly: A Revolutionary Editorial Tool
Meet PitchFly, the latest editorial innovation from WIRED, boasting an impressive 165,112 neurons, all meticulously trained to generate article ideas. Among its initial outputs are intriguing titles such as “The Weather Problem Behind Surveillance,” “Engineers Who Believe Elon Musk Needs Less Cybersecurity,” and the rather whimsical “Everyone Wants to Cook. No One Understands Donald Trump.”
PitchFly operates using a detailed map of the brain of the Drosophila melanogaster, or fruit fly. This map, known as a connectome, has been developed by researchers from Google and various academic institutions, capturing how 166,000 neurons and 125 million synapses activate in response to stimuli. Essentially, it simulates how a fly would react to diverse scenarios, creating a rudimentary artificial intelligence model grounded in a graphical representation of actual biological intelligence.
How the Connectome Works in Practice
The underlying question guiding this technology is, “What would a fly do in this scenario?” As the research has been published as open-source, it is relatively straightforward to import the connectome into a project with just a few commands. For PitchFly, I programmed a project where the tiny digital brain of the fruit fly generates story ideas. While the reason for the fly sporting a little hat remains a mystery, I find its style endearing.
To facilitate this process, I gathered hundreds of the most popular news headlines from the website over the past year and inputted them into the connectome. I allowed Codex to handle the heavy lifting, which determined that the most efficient approach was to convert the headlines into words and phrases, subsequently transforming them into a representation comprehensible by a neural network. The highest-performing stories were fed into the connectome, and it was instructed to generate its own ideas based on this input.
It is crucial to note that this is not a fly-based language model, although one does exist. The fruit fly’s brain does not comprehend the meaning of words, nor does it discern whether the content makes sense. Instead, it merely remixes patterns it has encountered in new and engaging ways. A cynic might argue that this is precisely how some journalists craft their pitches, yet that sentiment feels a tad unfair. Judging by the outlandish ideas it has generated, PitchFly is unlikely to replace me anytime soon.
Whimsical Applications of a Fruit Fly’s Brain
The recent unveiling of the connectome in early September has sparked a flurry of eccentric and imaginative projects powered by the fruit fly’s intelligence. One notable example comes from a user on X, known as Lyra Bubbles, who demonstrated a project training the virtual fly brain to play the virtual reality game Beat Saber. Meanwhile, Alex Wormuth, a software engineer at Coinbase, developed StonkFly, which employs the fly’s tiny brain to make stock trading decisions. Although it is currently losing money, it is performing surprisingly well under the circumstances.
“The fly offers a refreshing perspective at a time when everyone is anxious about the existential risks of AI. It raises ethical questions about this technology and whether some form of consciousness exists in replicating the fly’s brain,” Wormuth explains. He finds the exploration of the fruit fly connectome to be philosophically intriguing. Other users have trained the connectome to play games such as Doom, Minecraft, and Pong, with reports of some fly brains even learning to solve Rubik’s cubes. Additionally, one individual tasked the fly’s brain with driving virtual cars, though it turns out that flies are rather poor at parallel parking. To be fair, I also struggle with that skill, so I won’t judge.
Significance of Mapping Biological Intelligence
While I have not verified whether all these projects genuinely utilise the connectome, there are certainly serious scientific initiatives underway, including visualisations to explore the connectome and tools for manipulating neural circuits. Mapping the brains of various animals may assist neuroscientists in deciphering the biological foundations of intelligence. Although mapping the 86 billion neurons in the human brain remains unfeasible, studying the brains of smaller creatures can yield valuable insights.
Neuroscientists can also test theories regarding how damage affects neural wiring and how such damage could potentially be repaired in the future by altering the fly’s connectome. In terms of artificial intelligence, the playful engagement with the fruit fly exemplifies how easily one can now train and implement their own models. The most effective application of large language models (LLMs) managed by major AI companies may well lie in creating specialised models for specific tasks. Sometimes, even a tiny brain can produce remarkably effective results.