The “Silicon vs. Soul” Experiment
As New Zealand Music Month dominates the cultural calendar in Wellington, Parvis School of Economics and Music celebrated in characteristic fashion: by turning a concert into a rigorous economic experiment.
Last Friday night, the main recital hall at Inverlochy Place hosted the “Silicon vs. Soul” gala. The Parvis String Quartet performed six new chamber works. The twist? Three were composed by final-year Composition students, and three were generated by a Long Short-Term Memory (LSTM) neural network trained by the Computer Science department.
The audience, comprised of local critics, investors from the CBD, and faculty, was kept in the dark. They were not told which piece was human and which was algorithmic. Instead of applause, they were asked to submit a sealed bid—a monetary value representing their “Willingness to Pay” (WTP) for the rights to perform the score.
The Economic Upset: Efficiency Over Emotion?
The results, tabulated by the Economics Department over the weekend, were uncomfortable for the purists.
The highest average bid of the night went to Opus 1010, a piece characterised by rigid, mathematical counterpoint and rapid tempo changes. The audience valued it at an average of NZ$450 for performance rights.
When the envelopes were opened, it was revealed that Opus 1010 was the work of the AI.
“It suggests a shift in the perceived value of labour,” noted Professor Julian Sterling, Chair of Computational Logic. “The audience financially rewarded ‘complexity’ and ‘structural perfection’—traits where the AI excels. They undervalued the human pieces, which were more texturally experimental but structurally looser. Economically speaking, the market voted for efficiency over soul.”
The “Overfitting” Scandal
However, the victory for the machine was short-lived. Following the concert, the winning score was subjected to a routine intellectual property audit using Shazam-style fingerprinting algorithms—a standard procedure for the school’s commercial releases.
The audit flagged a severe legal breach in the AI’s masterpiece. In the third movement, the neural network had “hallucinated” a sequence of 12 bars that were statistically identical to a copyrighted jingle from a 1990s television commercial, which happened to be present in its training dataset.
“It is a classic case of overfitting,” explained a sheepish Dr. Sarah O’Connell from the CS faculty. “The model didn’t ‘compose’ a new melody; it memorised a pattern it thought was highly optimal because it had appeared frequently in the dataset. It effectively committed plagiarism because it lacks the concept of ownership.”
The Value of Human Liability
This “glitch” provided the Economics faculty with the most valuable insight of the night. While the AI produced the product with the highest initial market value, its retained value dropped to zero the moment the copyright infringement was discovered.
“This is the ‘Liability Discount’,” Dr. Percival Thorne told the gathered students on Monday morning. “You can hire a human composer, and if they plagiarise, you can sue them. You cannot sue an algorithm. Therefore, the economic premium we pay for human art is not just for ‘soul’—it is an insurance premium against legal liability.”
Student Resilience
Despite being outsold by a bot, the human students took the results in stride.
“It was actually a relief,” said Tama Ropata, a Year 13 composition scholar whose piece came second in the bidding war. “The AI wrote a perfect, boring fugue that stole a TV jingle. I wrote a piece about the wind on the Wellington south coast. I think my job is safe for a few more years.”
The recordings from the concert—minus the infringing bars of the AI piece—will be available on the Parvis intranet for analysis. The school plans to repeat the experiment next May, with a “cleaner” dataset and a more skeptical audience.
Leave a Reply