Then, when I got home, this article by Selene Brophy was published. I thought back to a VoiceMap tour I had just done in West Kreuzberg, passing through Viktoriapark. It was nearly perfect, with a surprising route and so much more than just facts. The publisher, Beata, was obviously overflowing with anecdotes and observations about this corner of Berlin, but she was sharing an infectious passion for the city too.
Could ChatGPT simulate Beata? I opened it up on the day GPT-4 became available and asked it.
Iain: “What is the most interesting thing about Viktoriapark in Berlin?”
ChatGPT: “As an AI language model, I don’t have personal opinions or feelings, but I can provide factual information…Viktoriapark is a park located in the Kreuzberg neighbourhood of Berlin. It was created in the 19th century and covers an area of approximately 12 hectares. The most interesting thing about Viktoriapark is probably the artificial waterfall that runs down the park’s highest point, the Kreuzberg hill.”
Job done. VoiceMap is safe!
But the truth is obviously more complicated, and I’ve realised that this is a new version of the same conversation I’ve always had about what we do at VoiceMap.
For some reason, phrases like “self-guided tour” and “audio guide” reek of dusty museums where people with stuffy accents list facts they know you aren’t cultured or intelligent enough to remember. This is what people expect from VoiceMap, to some extent, and this is why they think generative AI can do what we do.
The problem with generative AI, in other words, is on the demand side and not the supply side – about the expectations of people doing tours and not the people making them. I think this is equally true of other mediums.
Generative AI is going to keep getting better at simulating humans. It’s going to expertly imitate our passions and persuasively copy our flaws. The question, in the end, might be about how happy we are with imitations – especially when we travel, and seek to connect to the wider world.