Generative AI can Ideate Harder

How to use LLMs to solve "really hard problems"? After all, an LLM doesn’t know or understand the world: "it only knows “how the world talks about the world”... [while] Truly novel ideas are at least partly the result of understanding how the world works, and the application and porting of those conceptual models to new situations."

Consider: "the best way to address elevator’s passenger dissatisfaction at the duration of an elevator ride is not, beyond a point, to spend more on engineering; it is to put mirrors on its walls... a solution that LLMs find very hard to come up with by themselves, out of the box, especially when prompted in a very narrow way".

So instead "tell the AI to

• think about radically different ideas that combine solutions from different spaces
• take an expansive perspective of the user (or at least abstract away from the problem)...
• apply “lenses” to the problem. Lenses like “think of a solution as if you were the Dalai Lama”."

Do this "recursively in a chain... with human interventions in the right places."

Frameworks

While AIs can only infer how the world works through the words we've given them, many of those words reflect theories and frameworks which embed complex reasoning and an understanding of the world. These understandings are symbolic frameworks, "the results of lengthy research ... by gifted individuals ... that weeded out connections between things that didn’t work... [resulting in] ideas, crystallized in carefully crafted text, and propagated by thousands of examples of their use".

These form part of the LLMs' input data, and it's why prompts like “paint this like Andy Warhol” work - for Midjourney et al, "Andy Warhol is a framework... built with his brain and all the stimuli he processed ... Andy Warhol’s style is an embedding of his symbolic representation of the world", something you can ask an LLM to use.

"Frameworks are creative constraints ... to explore problems and solutions through a lens. That “passing through the narrows of the constraint” is often the spark for true innovation".

AIdea Colliders

The "Deliberate Framing, Recombination, and filtering (DFRF)" process creates an AIdea Collider, which:

• explodes and explores problems,
• "with LLMs outputs constrained through theories and frameworks ...
• before the deduced AIdea outputs are recombined and then filtered, at scale"

Example: MIT's Ideator doesn't have a chat interface - we "force the machine to apply a series of lenses to the problem statements, then humans recombine the pieces... to help solve problems through a specific design framework called Supermind Design. It has a strong emphasis on exploration of the problem space, and on organizational designs that leverage collective intelligence...

theoretically generate millions of idea fragments — and recombine them, geometrically expanding the output. Some triaging can be done by machines" - eg different models critiquing each other via fact-checking, applying ethical standards, etc.

But humans are probably also needed in this phase. Here useful systems would include distributing the work to large networks, providing AI summaries and "mapping the exploration space more visually".

This "combination of machines and people ... is called a supermind... Colliders could be a new type of [OpenAI] plug-in. Individual organizations could build their own" and apply it to their "knowledge graphs, hence adding a layer of signals that maps connections between topics, between people, and between people and topics — ultimately yielding additional ways of exploring the solution space."