I am reminded of what Alan Kay, inventor of the DynaBook and Object Oriented Programming used to say: any powerful tool takes time to learn: language, mathematics, playing an instrument — all take time to learn. He thinks the idea that you can learn to use a powerful tool without effort is absurd. I agree and one of the reasons so many firms use powerful tools in a commodity fashion is because they believe in the fallacy that easy to use tools are all we should invest in….
one analogy that we have found very helpful is swimming (similar to your riding a bike example). Both can't be learned by watching a YouTube video or reading an article, both need an initial guide or teacher (parent or swim coach), both need repeated practice to improve muscle-memory, both are intimidating at for the novice, both have novices who often have STRONG opinions with very little data, both have an initial learning curve for basic competency but also have path for world-class performance (Olympian swimmer or bikecylist), both need the learner to "want" to learn and get the benefits from the pain of learning, both don't have a high cash cost to slow adoption, both need practice, both have varied learning on-ramps that are effective (some people just jump in with a bike and swimming, others needs training class, etc.), both take a while before it becomes second nature but it does, etc.
Karim, I'll propose a different analogy—that the emergence of LLMs / GenAI is actually more like the invention of the internal combustion engine. What we needed for large-scale adoption of the engine was a usable (dare I say user-friendly [1]) automobile that people can actually drive. Although the internal combustion engine was invented in the 1870s, it wasn't until 1905 that automobiles started seeing mass adoption—with horses for transportation all but eliminated by 1915, ten years later. In other words, it took about three decades for the engine to be adopted in an everyday product, but that product quickly resulted in a massive, global sea change.
I believe a similar story will play out with GenAI, although the three decades from invention to mass adoption via a killer product might be sped up to 2-3 years. Over here at Parsnip (a company you might be indirectly invested in), we have a good idea for a killer car to build around that engine, with nice cushy seats instead of wooden boards and a steering wheel instead of a tiller. Only this car isn't for transportation — it's a machine for supercharging how quickly you can suck knowledge into your brain. Here's our thesis:
And... with regard to products powered by AI, this quote attributed to Henry Ford likely applies too: “If I had asked people what they wanted, they would have said faster horses.” I don't think many of the killer apps are obvious yet.
Thank you prof Karim, great article and references to follow. Yes, I encountered challenges when searching for application examples. Having a prompt is one thing, but comprehending their relevance in the context of business use cases is significant. That's how I learn—by grasping concepts through concise and straightforward examples that I can immediately apply, making them stick in my mind.
I think another underappreciated issue is that using gen AI is often pretty effortful. It requires coming up with a reasonable prompt, more or less critically reading the output, and iterating as needed. Including all the needed context in the prompt can be somewhat challenging. So the learning curve isn't the only limiting factor.
I am reminded of what Alan Kay, inventor of the DynaBook and Object Oriented Programming used to say: any powerful tool takes time to learn: language, mathematics, playing an instrument — all take time to learn. He thinks the idea that you can learn to use a powerful tool without effort is absurd. I agree and one of the reasons so many firms use powerful tools in a commodity fashion is because they believe in the fallacy that easy to use tools are all we should invest in….
This is really awesome John - I really love it.
Karim, great article.
one analogy that we have found very helpful is swimming (similar to your riding a bike example). Both can't be learned by watching a YouTube video or reading an article, both need an initial guide or teacher (parent or swim coach), both need repeated practice to improve muscle-memory, both are intimidating at for the novice, both have novices who often have STRONG opinions with very little data, both have an initial learning curve for basic competency but also have path for world-class performance (Olympian swimmer or bikecylist), both need the learner to "want" to learn and get the benefits from the pain of learning, both don't have a high cash cost to slow adoption, both need practice, both have varied learning on-ramps that are effective (some people just jump in with a bike and swimming, others needs training class, etc.), both take a while before it becomes second nature but it does, etc.
Thanks Paul. Suggestions for people to follow. What to watch?
FYI you may enjoy this.. backward bicycle… I use it to teach how learning works thru desirable difficulties/productive struggle… https://www.youtube.com/watch?v=MFzDaBzBlL0&vl=en
Karim, I'll propose a different analogy—that the emergence of LLMs / GenAI is actually more like the invention of the internal combustion engine. What we needed for large-scale adoption of the engine was a usable (dare I say user-friendly [1]) automobile that people can actually drive. Although the internal combustion engine was invented in the 1870s, it wasn't until 1905 that automobiles started seeing mass adoption—with horses for transportation all but eliminated by 1915, ten years later. In other words, it took about three decades for the engine to be adopted in an everyday product, but that product quickly resulted in a massive, global sea change.
I believe a similar story will play out with GenAI, although the three decades from invention to mass adoption via a killer product might be sped up to 2-3 years. Over here at Parsnip (a company you might be indirectly invested in), we have a good idea for a killer car to build around that engine, with nice cushy seats instead of wooden boards and a steering wheel instead of a tiller. Only this car isn't for transportation — it's a machine for supercharging how quickly you can suck knowledge into your brain. Here's our thesis:
https://parsnip.substack.com/p/vision-part-two
I'd love to get your dissection on this with your legendary combination of business acumen and technical expertise.
[1] maybe not with hand cranks, huh? 😂
And... with regard to products powered by AI, this quote attributed to Henry Ford likely applies too: “If I had asked people what they wanted, they would have said faster horses.” I don't think many of the killer apps are obvious yet.
Thank you prof Karim, great article and references to follow. Yes, I encountered challenges when searching for application examples. Having a prompt is one thing, but comprehending their relevance in the context of business use cases is significant. That's how I learn—by grasping concepts through concise and straightforward examples that I can immediately apply, making them stick in my mind.
Many thanks Professor Lakhani. It was amazing to be at your presentation at the Fall HBS Reunion. Please keep us enlightened.
I think another underappreciated issue is that using gen AI is often pretty effortful. It requires coming up with a reasonable prompt, more or less critically reading the output, and iterating as needed. Including all the needed context in the prompt can be somewhat challenging. So the learning curve isn't the only limiting factor.
Amazing article!