Making the AI mind, and book destruction
BY BRENT MCCONNELL
Claude, Grok, Gemini and ChatGPT exist only as hardware and software. They have no childhood, no life experience, nothing resembling a brain. While my dog still hasn’t fully committed to “sit,” these machines solve math problems that have stumped researchers for decades and carry on conversations that often feel human. What is actually happening?
The answer is a bit complicated. Large language models don’t think the way we do, and they don’t understand words the way we do. What they can do is predict the next word in a sequence based on patterns they’ve seen billions of times. Eventually those predictions become accurate enough to look like reasoning.
Training begins with enormous collections of public material, much of it gathered from a nonprofit archive that has been capturing snapshots of the public web since 2011. On top of that are Wikipedia, academic papers, public code repositories, news articles and millions of books. Duplicate pages and spam are filtered out, and the rest is broken into tokens — the small pieces of text from which a model actually learns. Today’s frontier models train on roughly ten to twenty trillion tokens.
That’s how their “knowledge” gets built. It’s also where the trouble starts, because much of that material belongs to someone else. Authors, artists and musicians argue their life’s work is being used to build products worth billions of dollars without permission or a dime in return. AI companies argue this is no different from a student learning by reading widely. The courts are now deciding where that line should be drawn.
Court filings showed that Anthropic legally purchased millions of used print books, then ran them through industrial equipment that stripped off the bindings so the pages could be scanned, destroying the originals in the process. The lawsuit that followed wasn’t really about that. It centered on the claim that the company had also copied millions more books from unauthorized internet sites. A federal judge ruled that scanning legally purchased books could qualify as fair use, but pirating them could not. Meta and other AI companies face similar lawsuits over how they assembled their training data.
Music is fighting much the same battle. The AI music generators Suno and Udio were sued by major record companies, which allege the systems were trained on copyrighted recordings without permission. Some companies have reached licensing agreements while others remain in court. Independent musicians have little leverage. The songwriter who never signed with a label can’t negotiate a licensing deal or afford years of litigation. Their music may help teach a machine that produces songs competing with their own, yet they may never receive a penny for it.
What fascinates me is what happens the moment people discover their new favorite artist has never drawn a breath, fallen in love or suffered heartbreak. We suddenly hear the same lyrics differently. What once felt deeply personal now feels manufactured. Many listeners quickly insist they prefer human-made music. Yet the songs themselves haven’t changed.
The Velvet Sundown, a “band” whose music was generated with AI, attracted more than a million monthly Spotify listeners before anyone realized there was no band. Some listeners felt deceived. The music was exactly the same before and after the revelation. Only our perception of the artist had changed.
I doubt that hesitation to embrace AI music will last. Instead of writing songs for millions of strangers, AI will eventually write songs for you — your memories, your hometown, your first love, your favorite fishing spot or the road you drive to work every morning. The emotional connection may be artificial, but it will still feel real. The more likely future isn’t AI replacing every musician. It’s AI working behind the scenes while a human remains out front. Give the audience a face, a voice and a believable life story while AI quietly handles more of the writing and production. Record labels gain greater control over both costs and content, and fewer people share in the creative work. Film may follow the same path.
The same questions are reaching writers, illustrators, filmmakers and photographers. AI is becoming another creative tool — one that’s difficult to ignore because it can save enormous amounts of time. There is no serious debate about whether this technology works. The better question is what we lose when we increasingly ask machines to think, write, compose and create on our behalf. The greatest value of art has never been the finished product alone. It is the uniquely human experiences, struggles and imagination behind it. If those become optional, we may discover too late that they were the most important part all along.
McConnell, a resident of Cape Haze, is a retired military officer and former Pentagon lead for weapons security policy. Email AITechThoughts@gmail.com
