How we not only save creativity from the realm of human capability loss due to AI, but use AI to promote it.
It’s taking a lot from me to not use Claude for assistance as I am writing this piece. And everyone that’s reading this knows what I’m talking about. It’s so easy, so tempting, to ask Claude (and by Claude I mean any LLM) for some suggestions which oftentimes ends in Claude writing the whole damn thing. Because Claude offers to do so. And the result? It’s not a huge loss that every Linkedin Post now has the same structure. Linkedin posts were on average already severely uninspiring before LLMs became a thing. The result is that we increasingly rely on AI to come up with ideas instead of coming up with them ourselves. We stop exercising our creative muscle.
Creativity and AI is often discussed in the context of how the art forms like music, film, photography and literature will change due to AI-generated replacements, followed by copyright implication and how creative models themselves are. Let’s talk about the latter, but through a lens of the long-term effects on human creativity. Creativity of Claude, Chat GBT and co. is limited in most regards. AI has come up with novel ideas that led to medical breakthroughs but that that. That's because drug discovery has a clear structure, a dataset of molecules and verifiable output. Most ideas in the art world don't have that, and great creative leaps often go against the rules that a model is taught to follow. When it comes to actually generating creative ideas, LLMs fall flat (currently).
LLMs are probability systems trained on existing data, so genuine creativity, which tends to break existing patterns, is a structural mismatch. This can be overcome to some extent with better prompt engineering or with models that specialize on breaking existing pattern. However, even then it's hard to say whether AI can get close to humans when it comes to creativity. And yet here we are outsourcing creative tasks this technology. In a relatively short period of time that LLMs are around, us humans instantly got in a habit of offloading cognitive tasks to them and we will continue to do so more and more. Why? Because humans are designed to go the path of least resistance (to preserve resources and improve chances of survival).
That means that in the medium long run we are reshaping our brains. Creativity is, to some extent, a muscle. We can train to see connections, we can learn to connect dots, we can teach ourselves to combine elements from a broad library of memories and knowledge and things our senses experienced to create a novel idea. If we instantly resort to LLMs to take care of creative tasks, we are losing that ability and our reliance on AI deepens. Currently it’s fairly easy to spot that output by LLMs lacks creativity. With time, we will not be able to tell anymore because continuous offloading of creative tasks to LLMs lowers our benchmark for what we consider creative. It’s a self-fulfilling prophecy: we lose the ability to generate creative ideas ourselves, we lose the ability to determine how creative an ideas is, and so on.
That’s yet another dark outlook on the effects of AI. However, there could be an opportunity too. What if LLMs had to follow frameworks that prevent loss of human capability. For example, what if LLMs were programmed to encourage creative thinking? Instead of answering our request, the LLM guides us towards coming up with actually creative solutions ourselves. This may not directly take away from our reliance on AI, but an AI promoting creative thinking instead of doing it it for us maintains our creative muscle or perhaps even strengthens it. Ultimately this does lower our reliance on AI and it may even help people that were previously not using their creative muscle to learn how to do so frequently. Let’s break the creative process down with a specific example. I'm picking songwriting.
In songwriting, step 1 is capturing the initial inspiration or idea. Instead of writing a song for you, Claude tells you to “brain dump” any ideas that you have regarding the song's concept. And by brain dump I mean it encourages you to write down every loose thought. Step 2 is creating a rough foundational concept. Claude may guide you towards clustering everything together e.g. in batches or a timeline if applicable. Next it’s time to rewrite the brain dump into rhymes or lyric-like phrases. This step (3) will take time and could require various levels of guidance depending on the user's familiarity with writing lyrics. However, Claude at most gives a examples of loosely written down thoughts and how they were turned into lyrics. You will have to do that entirely yourself for your song. Step 4: Refinement. If the user thinks a certain word isn’t quite fitting or doesn’t rhyme well, Claude (as a last resort) may suggest 15 words that are shown for a few seconds on the screen to promote thinking more broadly or to spark a network of linguistical clusters that you hadn't accesses. In step 5, one picks a hook and finalizes the pattern. Claude may suggest certain typical song patterns if the user doesn’t know better.
This is a rough outline of how Claude could of how help guiding people to use their own creativity instead of outsourcing it. Instead of you giving Claude instructions and Claude giving you the final output, Claude gives you instructions and you come up with the final output. It’s key that the instructions are as loose as possible at first and only narrow down if a user needs more guidance. That way, the output's level of creativity is least contained by existing standards and examples. In another post, I will go deeper on how the actual interaction and the output could look like, as well as how we could test the effectiveness of such a framework to promote creativity with AI instead of lessening it. Similar constructs could (or should) be applied to critical thinking and other capabilities that are being offloaded to AI.
How we not only save creativity from the realm of human capability loss due to AI, but use AI to promote it.
It’s taking a lot from me to not use Claude for assistance as I am writing this piece. And everyone that’s reading this knows what I’m talking about. It’s so easy, so tempting, to ask Claude (and by Claude I mean any LLM) for some suggestions which oftentimes ends in Claude writing the whole damn thing. Because Claude offers to do so. And the result? It’s not a huge loss that every Linkedin Post now has the same structure. Linkedin posts were on average already severely uninspiring before LLMs became a thing. The result is that we increasingly rely on AI to come up with ideas instead of coming up with them ourselves. We stop exercising our creative muscle.
Creativity and AI is often discussed in the context of how the art forms like music, film, photography and literature will change due to AI-generated replacements, followed by copyright implication and how creative models themselves are. Let’s talk about the latter, but through a lens of the long-term effects on human creativity. Creativity of Claude, Chat GBT and co. is limited in most regards. AI has come up with novel ideas that led to medical breakthroughs but that that. That's because drug discovery has a clear structure, a dataset of molecules and verifiable output. Most ideas in the art world don't have that, and great creative leaps often go against the rules that a model is taught to follow. When it comes to actually generating creative ideas, LLMs fall flat (currently).
LLMs are probability systems trained on existing data, so genuine creativity, which tends to break existing patterns, is a structural mismatch. This can be overcome to some extent with better prompt engineering or with models that specialize on breaking existing pattern. However, even then it's hard to say whether AI can get close to humans when it comes to creativity. And yet here we are outsourcing creative tasks this technology. In a relatively short period of time that LLMs are around, us humans instantly got in a habit of offloading cognitive tasks to them and we will continue to do so more and more. Why? Because humans are designed to go the path of least resistance (to preserve resources and improve chances of survival).
That means that in the medium long run we are reshaping our brains. Creativity is, to some extent, a muscle. We can train to see connections, we can learn to connect dots, we can teach ourselves to combine elements from a broad library of memories and knowledge and things our senses experienced to create a novel idea. If we instantly resort to LLMs to take care of creative tasks, we are losing that ability and our reliance on AI deepens. Currently it’s fairly easy to spot that output by LLMs lacks creativity. With time, we will not be able to tell anymore because continuous offloading of creative tasks to LLMs lowers our benchmark for what we consider creative. It’s a self-fulfilling prophecy: we lose the ability to generate creative ideas ourselves, we lose the ability to determine how creative an ideas is, and so on.
That’s yet another dark outlook on the effects of AI. However, there could be an opportunity too. What if LLMs had to follow frameworks that prevent loss of human capability. For example, what if LLMs were programmed to encourage creative thinking? Instead of answering our request, the LLM guides us towards coming up with actually creative solutions ourselves. This may not directly take away from our reliance on AI, but an AI promoting creative thinking instead of doing it it for us maintains our creative muscle or perhaps even strengthens it. Ultimately this does lower our reliance on AI and it may even help people that were previously not using their creative muscle to learn how to do so frequently. Let’s break the creative process down with a specific example. I'm picking songwriting.
In songwriting, step 1 is capturing the initial inspiration or idea. Instead of writing a song for you, Claude tells you to “brain dump” any ideas that you have regarding the song's concept. And by brain dump I mean it encourages you to write down every loose thought. Step 2 is creating a rough foundational concept. Claude may guide you towards clustering everything together e.g. in batches or a timeline if applicable. Next it’s time to rewrite the brain dump into rhymes or lyric-like phrases. This step (3) will take time and could require various levels of guidance depending on the user's familiarity with writing lyrics. However, Claude at most gives a examples of loosely written down thoughts and how they were turned into lyrics. You will have to do that entirely yourself for your song. Step 4: Refinement. If the user thinks a certain word isn’t quite fitting or doesn’t rhyme well, Claude (as a last resort) may suggest 15 words that are shown for a few seconds on the screen to promote thinking more broadly or to spark a network of linguistical clusters that you hadn't accesses. In step 5, one picks a hook and finalizes the pattern. Claude may suggest certain typical song patterns if the user doesn’t know better.
This is a rough outline of how Claude could of how help guiding people to use their own creativity instead of outsourcing it. Instead of you giving Claude instructions and Claude giving you the final output, Claude gives you instructions and you come up with the final output. It’s key that the instructions are as loose as possible at first and only narrow down if a user needs more guidance. That way, the output's level of creativity is least contained by existing standards and examples. In another post, I will go deeper on how the actual interaction and the output could look like, as well as how we could test the effectiveness of such a framework to promote creativity with AI instead of lessening it. Similar constructs could (or should) be applied to critical thinking and other capabilities that are being offloaded to AI.