AI hallucinations are an alarming bug from an information point of view but may actually be a useful feature when it comes to branding and marketing, according to FiscalNote’s Tim Hwang.
AI works differently to what people expect
Hallucination is the euphemism used for when AI answers a question with stuff it has simply made up.
But, Hwang points out, many people fail to properly appreciate that generative AI platforms aren’t consulting websites to give a 100% accurate answer but are rather “just predicting language” and giving what seems an appropriate response in the context.
“LLMs [large language models] are bad at everything we expect computers to be good at,” he says. “And LLMs are good at everything we expect computers to be bad at.”
So using AI as a search tool isn’t really a great idea, but “storytelling, creativity, aesthetics – these are all things that the technology is fundamentally really, really good at”.
A different way of thinking about AI
“A better way of thinking about AI is that LLMs are concept retrieval systems, not fact retrieval systems,” he suggests.
And that notion can be applied to brands, which are, after all, essentially little more than what people think they are.
“What LLMs are doing is in some ways perfect for identifying and exploring and manipulating brands as a concept. Hallucinations are a feature, not a bug,” Hwang argues.
Building interfaces for hallucination
Hwang reports that it’s possible to ask AI to hallucinate its own interface. And “sometimes I need a sort of scaffolding for it to hallucinate on, in order for it to reveal things about the hallucination”.
An example: ask the AI to hallucinate a brand as a character in Dungeons and Dragons with various attributes – “and you can use that to basically extract more intuition”.
You can add meters – “give it any arbitrary set of objects, and tell it to do things that you wouldn’t normally be able to measure or it would be really expensive to measure through stuff like surveys”.
An example: assign objects with a certain score based on the degree to which they align with the brand.
You can apply that to consumer research – give AI a score and ask for consumers who are more likely to become lifelong consumers of the brand based on that score.
The future is sliders and knobs
Ultimately you want to move away from a basic input-output approach and introduce variables to play around with, says Hwang – and you can ask the AI to give you whatever you need.
An example: “Imagine you’re a product ideas generator that has three sliders that go from zero to ten: one is age, one is gender, one is hipness (I have no idea what that is but I think it’s just something that I want out of these products).”
“Hallucinations really are, in some ways, the foundational element of what we want out of these technologies. I think rather than rejecting them, rather than fearing them, I think it’s manipulating these hallucinations that will create the biggest benefit for people in the ad and marketing space” – Tim Hwang, founder and CEO of FiscalNote.