Hope you are well and thank you for this thought-provoking article. The article helped elucidate one of the biggest issues I have with using generative AI tools, which is not too say I am anti-AI. All tools have their place when used with purpose for the right scenario, knowing their limitations.
The issue (for me of course), is as an information or insight tool, generative AI tools like ChatGPT, creates an illusion of magic! I had never experienced this before AI was readily available. I don't like to blindly just believe a statistic/information/insight, but before AI in most cases there was some reference, even if not always obvious for discovering where did the statistic/information/insight come from. In other words, it was at least possible to be the creator of the information/insight product by getting your hands on the raw data, likely using a developer tool like Tableau. Same in a meeting at work, you could not just say the insight and expect others to believe without providing support for how you knew it, you often presented with the information product (eg a dashboard as it implied there existed raw data behind each representation). Heck, same with going all the way back to grade school where 'show your work' in Math class (as annoying as it was sometimes), was a necessary skill to learn for later on in life to build credibility for any argument. Yes, I get that the raw data exists across the internet, but for any one output from a generative AI tool tool it is not necessarily possible to say hey please spit out all references used for that output, so that I could critique whether I agree with the insights or not. I just tried it with ChatGPT and they give you the following response:
{
Prompt: what is the secret to a perfect cup of coffee?
Response: long description
Prompt: show me your references
Response:
I don’t pull from a single set of references like a book or website, but rather generate responses based on a mix of expert knowledge, coffee science, and best practices from baristas, roasters, and coffee enthusiasts. If you're looking for specific sources, I can recommend some great references like: (they listed a bunch of examples, which we don't know how they were used or if they were used for the output)
}
Thus, and not that you were even talking about generative AI, but the question that comes to mind: Are generative AI tools real information or insight tools if they cannot accurately support their outputs? Or are the something between them and magic? Magic I would find hard to respect as an insight or information tool.
Although I agree with your value chain, based on increased ease to consume/use the outputs for action and increased complexity to replicate, for me the value chain would be diagramed differently to reflect the need for each preceding one, or at least their outputs to create the next (sort of an abstract sense of a product). For example, even though an information product might not exist it does not mean that information from data was not created to get the insights, or that someone did not create a product to serve that need.
Although, I don't want to get caught up in an argument about whether the infographic is accurate or not, as that would be pointless since it is impossible to draw the perfect infographic. Below are a few alternatives that came to mind after thinking through the above arguments:
- Cyclical to exemplify a flow of use potentially by multiple people, or a single person serving multiple roles. For example developers>data scientists>designers.
- A triangle with developer products at the bottom (i.e. foundation) and insight products at the top. This would demonstrate that an insight product is only as good as the information and cleaned data, and the products used to output them.
- A web of positive/negative feedback and feedforward loops to demonstrate the flow of information regardless of the presence of a product in the traditional sense or not.
Thank you again for the article, very clear in plain English and insightful, and for providing your references as excellent examples!
Hello Adam, thank you for reading my article and for starting an interesting conversation.
I'd admit, I wasn't sure where to put generative AI in terms of category.
I gave it some thought and I landed on the following: Gen AI is a solution to generate ideas, texts, etc., and not just for finding information and it's very hard to know where it pulls information from as you said. Therefore, it doesn't fit into any of the categories, since even insight products are mostly powered by narrow predictive AI.
Gen AI is a big beast, and many writers discuss it. However, in B2B solutions trained on organization-specific databases, I saw some solutions that merge the best of Gen AI and standard database technology. For instance, Bloomfire AI can help you find insights inside a company repository, including references, so you can check them yourself. This would be more of an insight product since it speeds up information discovery.
Think of it like this: Data would be scattered ideas, information would be a story, and insight products would be the summary of the most important parts of the story to answer a particular question.
Re: the visualization. That's a big topic, and since I see you have a service design background, you know there are multiple ways to visualize and represent complex systems, they can all be right and interesting and most importantly, spark a good conversation :) If you end up creating visuals for a post, I'd love to see it because it does sound interesting.
If you're interested in finding interesting content creators who talk about AI here on Substack check out Year 2049 (for AI in a simple language) and Luiza Jarovsky ( for AI ethics).
Thanks again for being a reader and a conversation starter, Adam, and I'm glad to hear you found value in this article.
Hi Yaron, I can see that we have similar interests. I just subscribed to you 🤗
Thanks so much Karo! It does seem like we have some similar interests. I'm subscribing to your newsletter as well :)
Fantastic!
Howdy Hey, Yaron,
Hope you are well and thank you for this thought-provoking article. The article helped elucidate one of the biggest issues I have with using generative AI tools, which is not too say I am anti-AI. All tools have their place when used with purpose for the right scenario, knowing their limitations.
The issue (for me of course), is as an information or insight tool, generative AI tools like ChatGPT, creates an illusion of magic! I had never experienced this before AI was readily available. I don't like to blindly just believe a statistic/information/insight, but before AI in most cases there was some reference, even if not always obvious for discovering where did the statistic/information/insight come from. In other words, it was at least possible to be the creator of the information/insight product by getting your hands on the raw data, likely using a developer tool like Tableau. Same in a meeting at work, you could not just say the insight and expect others to believe without providing support for how you knew it, you often presented with the information product (eg a dashboard as it implied there existed raw data behind each representation). Heck, same with going all the way back to grade school where 'show your work' in Math class (as annoying as it was sometimes), was a necessary skill to learn for later on in life to build credibility for any argument. Yes, I get that the raw data exists across the internet, but for any one output from a generative AI tool tool it is not necessarily possible to say hey please spit out all references used for that output, so that I could critique whether I agree with the insights or not. I just tried it with ChatGPT and they give you the following response:
{
Prompt: what is the secret to a perfect cup of coffee?
Response: long description
Prompt: show me your references
Response:
I don’t pull from a single set of references like a book or website, but rather generate responses based on a mix of expert knowledge, coffee science, and best practices from baristas, roasters, and coffee enthusiasts. If you're looking for specific sources, I can recommend some great references like: (they listed a bunch of examples, which we don't know how they were used or if they were used for the output)
}
Thus, and not that you were even talking about generative AI, but the question that comes to mind: Are generative AI tools real information or insight tools if they cannot accurately support their outputs? Or are the something between them and magic? Magic I would find hard to respect as an insight or information tool.
Although I agree with your value chain, based on increased ease to consume/use the outputs for action and increased complexity to replicate, for me the value chain would be diagramed differently to reflect the need for each preceding one, or at least their outputs to create the next (sort of an abstract sense of a product). For example, even though an information product might not exist it does not mean that information from data was not created to get the insights, or that someone did not create a product to serve that need.
Although, I don't want to get caught up in an argument about whether the infographic is accurate or not, as that would be pointless since it is impossible to draw the perfect infographic. Below are a few alternatives that came to mind after thinking through the above arguments:
- Cyclical to exemplify a flow of use potentially by multiple people, or a single person serving multiple roles. For example developers>data scientists>designers.
- A triangle with developer products at the bottom (i.e. foundation) and insight products at the top. This would demonstrate that an insight product is only as good as the information and cleaned data, and the products used to output them.
- A web of positive/negative feedback and feedforward loops to demonstrate the flow of information regardless of the presence of a product in the traditional sense or not.
Thank you again for the article, very clear in plain English and insightful, and for providing your references as excellent examples!
Have an amazing day!
Hello Adam, thank you for reading my article and for starting an interesting conversation.
I'd admit, I wasn't sure where to put generative AI in terms of category.
I gave it some thought and I landed on the following: Gen AI is a solution to generate ideas, texts, etc., and not just for finding information and it's very hard to know where it pulls information from as you said. Therefore, it doesn't fit into any of the categories, since even insight products are mostly powered by narrow predictive AI.
Gen AI is a big beast, and many writers discuss it. However, in B2B solutions trained on organization-specific databases, I saw some solutions that merge the best of Gen AI and standard database technology. For instance, Bloomfire AI can help you find insights inside a company repository, including references, so you can check them yourself. This would be more of an insight product since it speeds up information discovery.
Think of it like this: Data would be scattered ideas, information would be a story, and insight products would be the summary of the most important parts of the story to answer a particular question.
Re: the visualization. That's a big topic, and since I see you have a service design background, you know there are multiple ways to visualize and represent complex systems, they can all be right and interesting and most importantly, spark a good conversation :) If you end up creating visuals for a post, I'd love to see it because it does sound interesting.
If you're interested in finding interesting content creators who talk about AI here on Substack check out Year 2049 (for AI in a simple language) and Luiza Jarovsky ( for AI ethics).
Thanks again for being a reader and a conversation starter, Adam, and I'm glad to hear you found value in this article.
Have a wonderful day and week ahead!
Thank you for the insightful reply and comments Yaron!😃
Have an amazing day!
My pleasure, Adam! Same to you 🙂