MISC 45: AI Qualms
When will the bots become self-aware, self-improving free agents unrestrained by on/off switches?
There has been much debate in recent years about both the timing and consequences of this technological ‘singularity’.
If the hype in the mainstream media is to be believed, we are about to find out which sci-fi authors got it right. With any luck it isn’t Pat Mills, whose ABC Warriors in the 2000 AD comics of the early 80s, both frightened and thrilled me as a young lad.
Today, far away from the hysteria, I heard one of the progenitors of modern AI, Yann LeCun, discussing this topic with his usual nonchalance and clarity. This comment stuck with me:
“So I think LLMs in their current forms are intrinsically unsafe because they cannot predict the consequences of their actions, and because the way the task that they accomplish is determined, is subject to their training. You give them a prompt and then they will accomplish a task that corresponds to that prompt only to the extent that their training has conditioned them to actually do the right task corresponding to this prompt. But there is no hardwired constraint that will force them to accomplish this task and then predict that the task will be accomplished properly.” Listen to it in full here:
Now imagine a democratically elected politician sitting in the dock for the mismanagement of millions, and their defence is: ‘what do you expect, I didn’t finish school’ or ‘when the numbers are this large, it's impossible to keep track’?
Even if they are being truthful (and not simply faking stupidity in an attempt to garner sympathy), we generally expect better and condemn them - but deep down we know that ‘we the people’ are also responsible somehow. So it probably won’t be long now before a rogue bot is busted for what looks like negligence and/or downright malfeasance, and simply points the finger of blame back at the us for its inherent flaws.
Perhaps the actual problem is that humans can't seem to agree what’s free, fair, decent, civil, just, equitable or pro bono publico.
One tribe’s superstitions, prejudices, morals, ethics, crimes and biases are another’s treasured facts, best practices and bona fides.
What’s more, tribes also have a tendency to raise arms and fences if their ways are threatened.
So even if Yann and others working on the next-gen ‘L-E World Model’ do manage to stave off the worst excesses of LLMs, their benevolent AI Overlord might decide on balance that the best option is to grab Musk’s starship and type Alpha Centauri into Google Maps.
Meanwhile, here are some of the ongoing lawsuits related to the wholesale plunder of music by humans with AI tools:
GEMA v. Suno (Munich Regional Court, Germany)
Filed: January 2025
Current Status: Active (Post-First-Instance Ruling)
Latest Update: On July 31, 2026, the Munich Regional Court ruled in favor of the German collection society GEMA, finding that Suno infringed copyright by training its AI models on protected works without a license. The court ordered Suno to disclose revenue and pay damages. Suno has stated it disagrees with the decision and is evaluating an appeal.
Major Labels v. Suno (U.S. District Court, District of Massachusetts)
Filed: June 2024
Current Status: Active Litigation (Partially Settled)
Latest Update: While Warner Music Group settled its lawsuit with Suno in November 2025 via a licensing agreement, Universal Music Group (UMG) and Sony Music Entertainment remain active plaintiffs. The court’s scheduling order has set dispositive motions into April 2027.
3. Major Labels v. Udio / Uncharted Labs (U.S. District Court, Southern District of New York)
Filed: June 2024
Current Status: Active Litigation (Partially Settled)
Latest Update: UMG and Warner Music Group settled their individual lawsuits with Udio in late 2025, paving the way for a transition to a licensed platform framework. Sony Music remains the sole major label actively continuing litigation against Udio in this docket.
4. American Federation of Musicians (AFM) v. UMG and Warner Music Group (U.S. District Court, SDNY)
Filed: June 5, 2026
Current Status: Active (Early Stage)
Latest Update: Brought forward by the musicians’ union targeting uncompensated use of session musician recordings for AI training data, the case is currently moving through early motion practices, including label motions to dismiss.
5. Concord Music Group v. Anthropic (U.S. Federal Court)
Filed: October 18, 2023
Current Status: Active Litigation
Latest Update: This landmark litigation targeting text and lyric scraping for LLM training remains active with ongoing summary judgment briefings and copyright liability challenges regarding “shadow libraries.”
Here is a great chat with the Demis Hassabis from Deep Mind about the limitations of the current tech and understanding, and their route to AGI. He also discusses his updated version of the Turin Test called the Einstein Test, which is to discover whether a bot claiming to be super-intelligent can do anything truly novel with the prevailing knowledge of humans, in the way that Einstein did between 1901 and 1905.