I definitely think the TV industry sees opportunities to use AI to generate the satisfictory(?) script, and then hire a writer to punch it up - much cheaper than asking a team of writers to spend weeks generating the original ideas. Netflix has built its business model on producing a quantity of satisficing programming, with occasionally great series mixed in, and lots of that wouldn't be much worse with punched-up AI. I imagine AI being most useful for writing for genre where the writing is meant to be more instrumental than artful: kids' shows, cooking shows, reality TV, home improvement, etc. - and most of all, advertising copy.
Supporting the WGA is a great way to constrict monetization!
I made a living writing for TV and print. Cliches are useful, because you can bend them. A cliche, slightly altered, can jump, leveraging the assumptions built into the reader's mind and adding to them. Nothing wrong with a cliche, as raw material. Or as a joke.
"it's best to understand Generative AI tools as cliche generators" is such an elegant thrust that it's basically got me staring , mouth agape, at an entire chunk of my reality that has been deconstructed. I'm going to have to set aside some time to re-comprehend the role and function of cliche in my own life. I still need to read the rest of your essay, but I'm so shook I had to stop and say thanks for this succinct insight.
“What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.”
― Joseph Weizenbaum (1923-2008)
"The general purpose computer is freer than the trained brain."
I would very much like this to be true. However a couple of points occured to me. First ;domains were AIs are older and more developed they have produced beyond human brilliancies. Famously move 37 in game 2 of alphago vs Lee Sedol. More recently we have seen image generation tools winning art competitions. Second; while LLMs do usually respond with bland output to intial prompts these outputs can be further crafted by skillful users. The question isn't: 'can chatGPT write a script?', it is 'what can a skilled script writer produce with chatGPT?', it seems to me that in is the latter that the promise or threat of a productivity revolution lies.
You might be right, we'll see. But, regarding Alphago, I think it's worth keeping the distinction in mind between games with an objective, fixed goal (Chess, Go, Poker) and broader human endeavors where there is no established, objective metric of quality.
May 11, 2023·edited May 11, 2023Liked by Dave Karpf
Very useful insight, again. It prompted me to create a post for my own circle advising them to read three articles, of which this is one. My post doing that: https://ea.rna.nl/2023/05/11/where-are-gpt-and-friends-going/ One thing I added there is an explanation of the biggest misunderstanding about Generative AIs, namely that they contain information. Nope. Generative AI’s are hallucinating, *even* when they are producing correct replies.
Thank you for this perspective, it was less 'sky is falling' than a lot of thinkpieces I've seen lately.
The focus of your post seemed to be more on the written AI, but what are your thoughts on AI imagery? There's a lot of fears in the photography community about being replaced, when someone can upload a few selfies and get pro-style headshots for the corporate website, or an algorithm can generate images that look like photos of products being sold. Of course, that's the more commercial side. On the creative side, an AI generated image already won the Sony Photo Awards last month, which was controversial. There seem to be some similar anxieties in other visual arts mediums. Why paint when you can tell DALL-E what you want the canvas to look like? There's still some limitations to how real or accurate-seeming these images are now, but they're developing at breakneck pace and, for a lot of people, they'll probably be "satisificing" enough. Or at least that's the current anxiety. I brought all this up because I was curious about where you stand on the visual side of AI platforms.
1. I suspect the professions hardest hit by AI will be the ones that have the weakest mechanisms for collective power. Lawyers and doctors aren't going to be replaced, because even if an AI can diagnose you, it can't order an X-ray or write a scrip that will be honored by hospitals/pharmacists/insurers. The flip side is that freelancer-heavy creative professions are going to be decimated. (I made this argument in a piece I wrote in December. It's unfortunately already coming true https://www.theatlantic.com/technology/archive/2022/12/chatgpt-ai-chatbots-openai-cost-regulations/672539/ )
2. Yeah, relatedly, the trouble is that many of the scenarios where a visual artist would be commissioned to work are also situations where the client may be fine "satisficing" for cheap. Right now venues like WIRED magazine are holding firm that they'll keep paying for visuals, and won't use AI generated images. But what happens if/when they enter a budget crisis and have to decide which staff roles are mission-critical and which can be cut?
3. This again brings me back the Ted Chiang article I mentioned in the post (it's just so good...). I don't think people will stop painting or taking photographs because DALL-E can do it for them. But I suspect the already-rough road to being a full-time artist who can eat and pay rent is going to get much worse over the next few years. So that means art becomes the purview of the wealthy (even moreso than it already is). AI doesn't HAVE to push us in that direction, but it will likely do so given our existing power structure and levels of inequality.
I agree with pretty much everything you're saying. It's funny you bring up the lawyer/doctor thing again, because that's what re-triggered this thought as I was reading your article! It makes sense that highly skilled professions like those would be "safe" for now, but what about creative pursuits? It's interesting how, for a long time, we thought things that involved creativity would be safe from automation because they need a human touch (or whatever way you want to phrase it), but that's rapidly turning out to be bunk. I don't think anybody 5 years ago imagined this, really (but maybe I'm wrong haha).
You bring up an interesting point that I've thought as well: generative AI won't replace the visual arts for people, it'll just make it more precarious to those who want to make a living off of it. There will still be people who want to paint or photograph or screenprint or sculpt. Just like there's people now who enjoy less instantaneous and/or more disciplined ways of doing things, that may or may not be so mainstream. Generative AI could also bring more accessibility to those who have creative ideas but don't have the ability to paint or draw or whatever.
One distinction I would make though is that I wouldn't describe doctors or lawyers as more "highly skilled" than visual artists. What they are is more highly *credentialed.*
Those credentialing systems are, effectively, moats around their professional castles. Professions with moats will incorporate generative AI as tools to be used by the professionals. Professions without moats will face increased precarity, regardless of whether the AI can produce equivalent work.
No, yeah, that's a better way of putting it. Sometimes I can slip up with antiquated language but you're totally right that it's not a matter of skill, but credentials and the 'moats' they provide
I reckon truly creative activities will be safe as AI tends to produce a median of what's already out there that might be crunched and regurgitated. Though I'd like to interact with a LLM trained on the entire works of Douglas Adams and see what world view gets simulated :-)
Some professions will not get early traction. Not many people would agree to be treated by medical AI V1.0, right? At least, not for anything non trivial. But that might be a Brit- centric view of healthcare.
Stupid article written by someone who doesnt understand the potential of generative AI.
Firstly, 10 people asking for a Disney World itinerary won't get the exactly the same reply, because there normally a degree of randomness built in. But more importantly, its possible to personalise the itinerary by including more information. For example you could tell it that you like helter-skelters, splashing boats and scary rides
Great piece. The last line is a real lightbulb moment. Constricting monetization opportunities is a clever regulatory route.
And I love the connect to satisficing. The perfect explanation of LLM as tool.
I definitely think the TV industry sees opportunities to use AI to generate the satisfictory(?) script, and then hire a writer to punch it up - much cheaper than asking a team of writers to spend weeks generating the original ideas. Netflix has built its business model on producing a quantity of satisficing programming, with occasionally great series mixed in, and lots of that wouldn't be much worse with punched-up AI. I imagine AI being most useful for writing for genre where the writing is meant to be more instrumental than artful: kids' shows, cooking shows, reality TV, home improvement, etc. - and most of all, advertising copy.
Supporting the WGA is a great way to constrict monetization!
I made a living writing for TV and print. Cliches are useful, because you can bend them. A cliche, slightly altered, can jump, leveraging the assumptions built into the reader's mind and adding to them. Nothing wrong with a cliche, as raw material. Or as a joke.
That's very well put, thanks.
"it's best to understand Generative AI tools as cliche generators" is such an elegant thrust that it's basically got me staring , mouth agape, at an entire chunk of my reality that has been deconstructed. I'm going to have to set aside some time to re-comprehend the role and function of cliche in my own life. I still need to read the rest of your essay, but I'm so shook I had to stop and say thanks for this succinct insight.
“What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.”
― Joseph Weizenbaum (1923-2008)
"The general purpose computer is freer than the trained brain."
— W. Ross Ashby (1903-1972)
I would very much like this to be true. However a couple of points occured to me. First ;domains were AIs are older and more developed they have produced beyond human brilliancies. Famously move 37 in game 2 of alphago vs Lee Sedol. More recently we have seen image generation tools winning art competitions. Second; while LLMs do usually respond with bland output to intial prompts these outputs can be further crafted by skillful users. The question isn't: 'can chatGPT write a script?', it is 'what can a skilled script writer produce with chatGPT?', it seems to me that in is the latter that the promise or threat of a productivity revolution lies.
You might be right, we'll see. But, regarding Alphago, I think it's worth keeping the distinction in mind between games with an objective, fixed goal (Chess, Go, Poker) and broader human endeavors where there is no established, objective metric of quality.
Very useful insight, again. It prompted me to create a post for my own circle advising them to read three articles, of which this is one. My post doing that: https://ea.rna.nl/2023/05/11/where-are-gpt-and-friends-going/ One thing I added there is an explanation of the biggest misunderstanding about Generative AIs, namely that they contain information. Nope. Generative AI’s are hallucinating, *even* when they are producing correct replies.
Thank you for this perspective, it was less 'sky is falling' than a lot of thinkpieces I've seen lately.
The focus of your post seemed to be more on the written AI, but what are your thoughts on AI imagery? There's a lot of fears in the photography community about being replaced, when someone can upload a few selfies and get pro-style headshots for the corporate website, or an algorithm can generate images that look like photos of products being sold. Of course, that's the more commercial side. On the creative side, an AI generated image already won the Sony Photo Awards last month, which was controversial. There seem to be some similar anxieties in other visual arts mediums. Why paint when you can tell DALL-E what you want the canvas to look like? There's still some limitations to how real or accurate-seeming these images are now, but they're developing at breakneck pace and, for a lot of people, they'll probably be "satisificing" enough. Or at least that's the current anxiety. I brought all this up because I was curious about where you stand on the visual side of AI platforms.
Three relevant thoughts about AI imagery:
1. I suspect the professions hardest hit by AI will be the ones that have the weakest mechanisms for collective power. Lawyers and doctors aren't going to be replaced, because even if an AI can diagnose you, it can't order an X-ray or write a scrip that will be honored by hospitals/pharmacists/insurers. The flip side is that freelancer-heavy creative professions are going to be decimated. (I made this argument in a piece I wrote in December. It's unfortunately already coming true https://www.theatlantic.com/technology/archive/2022/12/chatgpt-ai-chatbots-openai-cost-regulations/672539/ )
2. Yeah, relatedly, the trouble is that many of the scenarios where a visual artist would be commissioned to work are also situations where the client may be fine "satisficing" for cheap. Right now venues like WIRED magazine are holding firm that they'll keep paying for visuals, and won't use AI generated images. But what happens if/when they enter a budget crisis and have to decide which staff roles are mission-critical and which can be cut?
3. This again brings me back the Ted Chiang article I mentioned in the post (it's just so good...). I don't think people will stop painting or taking photographs because DALL-E can do it for them. But I suspect the already-rough road to being a full-time artist who can eat and pay rent is going to get much worse over the next few years. So that means art becomes the purview of the wealthy (even moreso than it already is). AI doesn't HAVE to push us in that direction, but it will likely do so given our existing power structure and levels of inequality.
Thanks for getting back to me!
I agree with pretty much everything you're saying. It's funny you bring up the lawyer/doctor thing again, because that's what re-triggered this thought as I was reading your article! It makes sense that highly skilled professions like those would be "safe" for now, but what about creative pursuits? It's interesting how, for a long time, we thought things that involved creativity would be safe from automation because they need a human touch (or whatever way you want to phrase it), but that's rapidly turning out to be bunk. I don't think anybody 5 years ago imagined this, really (but maybe I'm wrong haha).
You bring up an interesting point that I've thought as well: generative AI won't replace the visual arts for people, it'll just make it more precarious to those who want to make a living off of it. There will still be people who want to paint or photograph or screenprint or sculpt. Just like there's people now who enjoy less instantaneous and/or more disciplined ways of doing things, that may or may not be so mainstream. Generative AI could also bring more accessibility to those who have creative ideas but don't have the ability to paint or draw or whatever.
I'll check out that Ted article, thanks again!
Thanks, likewise.
One distinction I would make though is that I wouldn't describe doctors or lawyers as more "highly skilled" than visual artists. What they are is more highly *credentialed.*
Those credentialing systems are, effectively, moats around their professional castles. Professions with moats will incorporate generative AI as tools to be used by the professionals. Professions without moats will face increased precarity, regardless of whether the AI can produce equivalent work.
It's all about precarity!
No, yeah, that's a better way of putting it. Sometimes I can slip up with antiquated language but you're totally right that it's not a matter of skill, but credentials and the 'moats' they provide
I reckon truly creative activities will be safe as AI tends to produce a median of what's already out there that might be crunched and regurgitated. Though I'd like to interact with a LLM trained on the entire works of Douglas Adams and see what world view gets simulated :-)
Some professions will not get early traction. Not many people would agree to be treated by medical AI V1.0, right? At least, not for anything non trivial. But that might be a Brit- centric view of healthcare.
Stupid article written by someone who doesnt understand the potential of generative AI.
Firstly, 10 people asking for a Disney World itinerary won't get the exactly the same reply, because there normally a degree of randomness built in. But more importantly, its possible to personalise the itinerary by including more information. For example you could tell it that you like helter-skelters, splashing boats and scary rides