"This is an astonishing change in the landscape of human creativity, and one that likely makes execution, not raw creativity, a more distinguishing factor for future innovations."
That's long been true. More start-ups fail because they can't deliver their product (complete the design, build and ship) than because the product idea is faulty (although that is a close second). However, I agree that this points towards much more awareness of the necessity for execution which will likely come as a surprise to many people.
For the "fuzzy" front-end of innovation (i.e., imagination, exploration & creativity) generative AI tools can certainly serve as a type of "creative Muse" if skillfully applied. Linus Pauling would have agreed: "The best way to have a good idea is to have a lot of ideas and throw away the bad ones." However, I do wonder about generative AI's ability to "throw away the bad ones" where critical thinking takes precedence.
One problem that some creative people face is that they are more interested in the experience of creativity than they are the practical outcome.
I’ve suffered from this syndrome sometimes. Generating many ideas happens very naturally and organically, but figuring out which ideas should be kept and which discarded is more like work.
It can help a lot to have a partner for whom doing such calculations comes as easy as generating ideas does for the creative. An editor if you will.
What a hopeful,affirmative analysis!I have no understanding of the mechanics of LLMs but love playing with Chatgpt to enhance my late-life learning.Thanks for adding another layer.
" And LLMs are very good at this, acting as connection machines between unexpected concepts. They are trained by generating relationships between tokens that may seem unrelated to humans but represent some deeper connections."
I believe AI's amazing ability that you describe here is the power to use metaphors: helping people understand new ideas by comparing and combining already understood ideas - and the fact that those ideas are gathered from not just our limited human experiences but from the totality of all human experiences currently recorded on the web. Ideas beget ideas.
As a creative & designer who works with companies and students I submitted some briefs from the AI contests, trying to define the context and forcing the response on ideas that have never been done:
however, the strong feeling is that since AI is a statistical machine, it seeks the most probable answers compared to the database, naturally creating variations of what already exists: it is true that the results are similar or slightly better than those of less creative people, but this shows that it is not very creative and if used for this task it will increase the noise of low creativity, which is already very high.
Creativity, on the other hand, is anti-statistical, it seeks infrequent combinations between experiences and works in a certain sense in the opposite way compared to AI: in this sense random machines give more surprising results, but not very useful, because creativity is not simple chaos, but a kind of reverse logic
Interesting to read this article (as always Ethan!) while considering the work of Teresa Amabile and others in the area of creativity. If we take the Three Components of Creativity model as an example, AI qualifies on the expertise front through being trained as it is, but it is the human that provides the motivation and much of the creative thinking skills through the prompting. At least for now!
It could be argued that nothing is originally creative. Having the ability to combine domain knowledge (which AI is clearly good at doing even across quite disparate topics), and with human guidance coming up with new ideas, isn't a lot different (if at all) from a human enabling suspension of thought to combine knowledge and experience to produce new ideas.
John Cleese once said that we need time to switch from a cluttered mind to being creative, and that we also need time to stay in the creative mode to... well... be creative.
I can confirm that this works well for me. And usually, the first idea that pops up is never the best (or the one I like the most).
I'm also pretty bad at coming up with ideas when prompted on the spot. My creativity usually sparks when I'm alone and actively creating something.
Hence, I wonder about the way the human contests were structured. Did they prompt the participants to be immediately creative (and deliver ideas on the spot), or did they also get the time to adjust?
The human body with all its senses is and will remain the right tool for the job. But we have to once again allow people to dive deep into craft and reward them for that. It’s a process that takes time and patience, but proper execution always does.
RE: "However, calling it 'stolen thought pattern' is really dependent on how one defines it."
That's true. Here's how I define it. Every human upload to the internet or otherwise into an AI blackbox contains intellectual property including creating algorithms that have not yet been fully incorporated into US or international copyright law. Until policy rules on this are created, I define the IP as belonging to the author of the material. I further define use of that material for AI purposes without clear authorial approval as IP theft.
To go to your second point. While I think it is wonderful for people who have education/resources to help those who do not, I do not believe that help should come via theft of the cognitive algorithms that the educated worked hard to create. The theft creates a "middle man" between the "educated/resource rich" and "those without". The AI builder is the middle man. The middle man currently charges "those without" very little for the use of the cognitive algorithms he or she has stolen/appropriated. However, that is likely to change. And until the theft is replaced by transparent reward structures and benefits sharing, this will be a destructive process antithetical to traditional methods of building wealth through community uplift.
"I do not believe that help should come via theft of the cognitive algorithms that the educated worked hard to create. The theft creates a "middle man" between the "educated/resource rich" and "those without". The AI builder is the middle man."
Agreed. However, approaching this problem in a binary manner isn't helpful either. A couple of assumptions that we've made here is that the AI builder is the only middlemen benefiting. To that extent, I believe that it's multi factorial. The middle men are also the governments, corporations and other systems already set in place for uplifting disadvantaged communities. The inadequacies in the traditional methods of wealth creation is what is creating opportunities for the "AI builder middle men" to create systems based on IP theft. To be sure, I'm not defending the thievery and I completely support your argument towards transparent reward structures and benefits sharing but I also want to add that it's a collective responsibility, not just limited to the tech companies.
Great comment. Moreover, Ideas cannot get stolen because nobody owned them in the first place. Our brain composes "ideas" from past experiences. And those experiences are collections of events with other people other impressions, and so forth. If you would be born and kept in a black box with no light, no images, and no sound, your ideas would gravitate around the given experiences from your DNA and come from your ancestors. It's a great practice in innovation teams to say goodbye to "MY IDEA" and say hello to "OUR IDEAS".
The greatest ideas ever created were by amazing teams - not by solists.
"There is more underlying similarity in the ideas that the current generation of AIs produce than among ideas generated by a large number of humans"
Boy is there ever. I wonder how much time will pass until this is resolved, or if somehow won't ever be resolved and we'll just sort of reconfigure our notions about what creativity is.
Currently, all the LLMs are basically trained on more or less the same corpus, easily accessible internet text. So it makes sense that they would all be fairly generic. If you asked a bunch of college students to come up with ideas, in some ways that idea set would also be similar because all the people generating it are undergoing the same experience and often have similar backgrounds. I bet that as AIs evolve, we'll figure out how to craft or gather more interesting data to feed them and get more diverse ideas (or at least the capability of coming up with more diverse ideas).
Re: "Where previously, there were only a few people who had the ability to come up with good ideas, now there are many. This is an astonishing change in the landscape of human creativity, and one that likely makes execution, not raw creativity, a more distinguishing factor for future innovations."
Would you clarify? From what I've read in your paper, we still have the same number of people able to come up with creative ideas.
ChatGPT contains millions of people whose cognitive capacity has been stolen. If incapable people use that tool, they haven't come up with any creative ideas. The millions of people whose ideas are trapped in the AI software have been used without their consent to bolster the incapable ones. The fact that all of those people are trapped in a black box makes it impossible to judge whether any of them was creative before their cognitive patterns were scraped. So I don't understand how you're coming up with your numbers.
This is such a meta observation. "The millions of people whose ideas are trapped in the AI software have been used without their consent to bolster the incapable ones." - I see where you are going with this but this statement is based on an assumption that it's either/ or and not a complementary relationship between LLMs or any other AI system and humans. As far as scraping cognitive patterns goes, its a totally different beast to tackle. Something, which would still indirectly address such concerns is companies balancing transparency around model training & frameworks while ensuring that there isn't any potential misuse.
Re: " I see where you are going with this but this statement is based on an assumption that it's either/ or and not a complementary relationship between LLMs or any other AI system and humans."
I see there is a possible complementary relationship, but that doesn't go to Ethan's point that the software creates more creative people. A human who uses the stolen thought patterns of 1 million people to generate a list of possible companies is not being creative. He or she is just engaging in a kind of theft. Or appropriation. Maybe that's creative in the way thief or slaver creative, but that's not how creativity is typically understood.
100% transparent and consensual model training where people had to opt in to have their work scraped would make a big difference here.
I completely agree with the fact that transparency is important. However, calling it 'stolen thought pattern' is really dependent on how one defines it. I do understand the point that you are trying to make here and understand how it'd put some people at disadvantage to say the least. The other way, to look at it however would be to imagine someone who never had proper education/ resources to ideate & execute and now has disposal to a LLM. Wouldn't it be great if they could generate and idea and execute it at least locally with the least possible resources to help their community and pave the way for monetization and wealth creation.
"This is an astonishing change in the landscape of human creativity, and one that likely makes execution, not raw creativity, a more distinguishing factor for future innovations."
That's long been true. More start-ups fail because they can't deliver their product (complete the design, build and ship) than because the product idea is faulty (although that is a close second). However, I agree that this points towards much more awareness of the necessity for execution which will likely come as a surprise to many people.
Very helpful review of literature. I made similar points and proposed an approach to institutionalize AI assisted creativity here Generative AI can Ideate Harder - Medium https://medium.com/@giannigiacomelli69/generative-ai-can-ideate-harder-bdd9e37a01d8
For the "fuzzy" front-end of innovation (i.e., imagination, exploration & creativity) generative AI tools can certainly serve as a type of "creative Muse" if skillfully applied. Linus Pauling would have agreed: "The best way to have a good idea is to have a lot of ideas and throw away the bad ones." However, I do wonder about generative AI's ability to "throw away the bad ones" where critical thinking takes precedence.
That’s a useful comment, thanks.
One problem that some creative people face is that they are more interested in the experience of creativity than they are the practical outcome.
I’ve suffered from this syndrome sometimes. Generating many ideas happens very naturally and organically, but figuring out which ideas should be kept and which discarded is more like work.
It can help a lot to have a partner for whom doing such calculations comes as easy as generating ideas does for the creative. An editor if you will.
What a hopeful,affirmative analysis!I have no understanding of the mechanics of LLMs but love playing with Chatgpt to enhance my late-life learning.Thanks for adding another layer.
" And LLMs are very good at this, acting as connection machines between unexpected concepts. They are trained by generating relationships between tokens that may seem unrelated to humans but represent some deeper connections."
I believe AI's amazing ability that you describe here is the power to use metaphors: helping people understand new ideas by comparing and combining already understood ideas - and the fact that those ideas are gathered from not just our limited human experiences but from the totality of all human experiences currently recorded on the web. Ideas beget ideas.
A prompt that works for me is to use multiple personas to generate varied perspectives. Full prompt here https://www.linkedin.com/posts/tombarrett1_aiforeducation-promptcraft-ai-activity-7094436062348906496-zPMa
As a creative & designer who works with companies and students I submitted some briefs from the AI contests, trying to define the context and forcing the response on ideas that have never been done:
however, the strong feeling is that since AI is a statistical machine, it seeks the most probable answers compared to the database, naturally creating variations of what already exists: it is true that the results are similar or slightly better than those of less creative people, but this shows that it is not very creative and if used for this task it will increase the noise of low creativity, which is already very high.
Creativity, on the other hand, is anti-statistical, it seeks infrequent combinations between experiences and works in a certain sense in the opposite way compared to AI: in this sense random machines give more surprising results, but not very useful, because creativity is not simple chaos, but a kind of reverse logic
Interesting to read this article (as always Ethan!) while considering the work of Teresa Amabile and others in the area of creativity. If we take the Three Components of Creativity model as an example, AI qualifies on the expertise front through being trained as it is, but it is the human that provides the motivation and much of the creative thinking skills through the prompting. At least for now!
It could be argued that nothing is originally creative. Having the ability to combine domain knowledge (which AI is clearly good at doing even across quite disparate topics), and with human guidance coming up with new ideas, isn't a lot different (if at all) from a human enabling suspension of thought to combine knowledge and experience to produce new ideas.
John Cleese once said that we need time to switch from a cluttered mind to being creative, and that we also need time to stay in the creative mode to... well... be creative.
I can confirm that this works well for me. And usually, the first idea that pops up is never the best (or the one I like the most).
I'm also pretty bad at coming up with ideas when prompted on the spot. My creativity usually sparks when I'm alone and actively creating something.
Hence, I wonder about the way the human contests were structured. Did they prompt the participants to be immediately creative (and deliver ideas on the spot), or did they also get the time to adjust?
Very insightful read. This makes me think of a great article on craft that I read recently: https://buildinghope.substack.com/p/the-right-tool-for-the-job
The human body with all its senses is and will remain the right tool for the job. But we have to once again allow people to dive deep into craft and reward them for that. It’s a process that takes time and patience, but proper execution always does.
Thanks Nimit.
RE: "However, calling it 'stolen thought pattern' is really dependent on how one defines it."
That's true. Here's how I define it. Every human upload to the internet or otherwise into an AI blackbox contains intellectual property including creating algorithms that have not yet been fully incorporated into US or international copyright law. Until policy rules on this are created, I define the IP as belonging to the author of the material. I further define use of that material for AI purposes without clear authorial approval as IP theft.
To go to your second point. While I think it is wonderful for people who have education/resources to help those who do not, I do not believe that help should come via theft of the cognitive algorithms that the educated worked hard to create. The theft creates a "middle man" between the "educated/resource rich" and "those without". The AI builder is the middle man. The middle man currently charges "those without" very little for the use of the cognitive algorithms he or she has stolen/appropriated. However, that is likely to change. And until the theft is replaced by transparent reward structures and benefits sharing, this will be a destructive process antithetical to traditional methods of building wealth through community uplift.
"I do not believe that help should come via theft of the cognitive algorithms that the educated worked hard to create. The theft creates a "middle man" between the "educated/resource rich" and "those without". The AI builder is the middle man."
Agreed. However, approaching this problem in a binary manner isn't helpful either. A couple of assumptions that we've made here is that the AI builder is the only middlemen benefiting. To that extent, I believe that it's multi factorial. The middle men are also the governments, corporations and other systems already set in place for uplifting disadvantaged communities. The inadequacies in the traditional methods of wealth creation is what is creating opportunities for the "AI builder middle men" to create systems based on IP theft. To be sure, I'm not defending the thievery and I completely support your argument towards transparent reward structures and benefits sharing but I also want to add that it's a collective responsibility, not just limited to the tech companies.
My twopenny worth about "stolen" ideas.The other side of this maybe making people's underrated or forgotten ideas..in my case novels...known again.
Great comment. Moreover, Ideas cannot get stolen because nobody owned them in the first place. Our brain composes "ideas" from past experiences. And those experiences are collections of events with other people other impressions, and so forth. If you would be born and kept in a black box with no light, no images, and no sound, your ideas would gravitate around the given experiences from your DNA and come from your ancestors. It's a great practice in innovation teams to say goodbye to "MY IDEA" and say hello to "OUR IDEAS".
The greatest ideas ever created were by amazing teams - not by solists.
Casting my vote for this:
“Ideas cannot get stolen because nobody owned them in the first place.”
Idea ownership is mostly a self induced fantasy fueled by ego and business agendas.
"There is more underlying similarity in the ideas that the current generation of AIs produce than among ideas generated by a large number of humans"
Boy is there ever. I wonder how much time will pass until this is resolved, or if somehow won't ever be resolved and we'll just sort of reconfigure our notions about what creativity is.
Currently, all the LLMs are basically trained on more or less the same corpus, easily accessible internet text. So it makes sense that they would all be fairly generic. If you asked a bunch of college students to come up with ideas, in some ways that idea set would also be similar because all the people generating it are undergoing the same experience and often have similar backgrounds. I bet that as AIs evolve, we'll figure out how to craft or gather more interesting data to feed them and get more diverse ideas (or at least the capability of coming up with more diverse ideas).
I think that's right, Greg. I really look forward to watching this unfold! And, of course, continuing to pay some attention so we can benefit.
Thanks for another interesting article.
Re: "Where previously, there were only a few people who had the ability to come up with good ideas, now there are many. This is an astonishing change in the landscape of human creativity, and one that likely makes execution, not raw creativity, a more distinguishing factor for future innovations."
Would you clarify? From what I've read in your paper, we still have the same number of people able to come up with creative ideas.
ChatGPT contains millions of people whose cognitive capacity has been stolen. If incapable people use that tool, they haven't come up with any creative ideas. The millions of people whose ideas are trapped in the AI software have been used without their consent to bolster the incapable ones. The fact that all of those people are trapped in a black box makes it impossible to judge whether any of them was creative before their cognitive patterns were scraped. So I don't understand how you're coming up with your numbers.
This is such a meta observation. "The millions of people whose ideas are trapped in the AI software have been used without their consent to bolster the incapable ones." - I see where you are going with this but this statement is based on an assumption that it's either/ or and not a complementary relationship between LLMs or any other AI system and humans. As far as scraping cognitive patterns goes, its a totally different beast to tackle. Something, which would still indirectly address such concerns is companies balancing transparency around model training & frameworks while ensuring that there isn't any potential misuse.
Thanks Nimit.
Re: " I see where you are going with this but this statement is based on an assumption that it's either/ or and not a complementary relationship between LLMs or any other AI system and humans."
I see there is a possible complementary relationship, but that doesn't go to Ethan's point that the software creates more creative people. A human who uses the stolen thought patterns of 1 million people to generate a list of possible companies is not being creative. He or she is just engaging in a kind of theft. Or appropriation. Maybe that's creative in the way thief or slaver creative, but that's not how creativity is typically understood.
100% transparent and consensual model training where people had to opt in to have their work scraped would make a big difference here.
+1 - Having someone's work scraped without their consent is definitely a problem that needs to be solved.
I completely agree with the fact that transparency is important. However, calling it 'stolen thought pattern' is really dependent on how one defines it. I do understand the point that you are trying to make here and understand how it'd put some people at disadvantage to say the least. The other way, to look at it however would be to imagine someone who never had proper education/ resources to ideate & execute and now has disposal to a LLM. Wouldn't it be great if they could generate and idea and execute it at least locally with the least possible resources to help their community and pave the way for monetization and wealth creation.
Time to update this excellent article.
Really liked reading your piece Ethan. Adding constraints is a great tip while generating ideas. Always work better.