Sanderson’s Second Law of Magic: Limitations > Powers
Belle’s Second Law for AI as Magic: Limitations > Capabilities
REMINDER: Sanderson’s Laws are for authors, Belle’s Laws are for characters.
Sanderson’s Second Law of Magic: Limitations > Powers
To explain himself this time around, Sanderson comes right back to super heroes. Most people can list one or more of Superman’s powers (flight, xray vision, etc) but if you want to have a long conversation about Superman, you will have to go beyond what he can do and talk about what he can’t. His ‘code of ethics and his weakness to kryptonite’ are what make Superman stories, and the character himself, interesting.
As usual, I refer you to the original essay for a more comprehensive build out of the argument but I do want to take a moment to share Sanderson’s more thorough breakdown of “limitations” into 3 categories: Limitations, Weaknesses, and Costs.
There is also an important 4th category that might be considered a subcategory of Limitations that I will call “Prohibitions”.
Sanderson also acknowledges that this law, perhaps more than either of his others, is a deeper truth about storytelling. Without conflict or challenge, a story is rarely compelling. Limitations, Weaknesses, Costs, and Prohibitions are what allow stories told in worlds rife with magic to capture our attention and imaginations.
Belle’s Second Law for AI as Magic: Limitations > Capabilities
Because AI is still a relatively new technology and because its speedy evolution means that the capabilities are evolving on an almost-daily basis, most people I know are fascinated by (and fixated on) AI’s capabilities. This is completely understandable and yet, the most boring conversations I have about AI are those in which the answer to every question is, “AI can/will do that.”
If the future is entirely about AI and there is no room for humans, I lose all interest in talking about the future.
Therefore, even in a world in which AI’s capabilities are sparking ideas and driving innovation at an unprecedented rate, Limitations, Weaknesses, Costs, and Prohibitions remain the most important aspects of AI to dissect and discuss. This is perhaps even more important when those concepts are continuously evolving.
Coming back to my email editing example from the First Law, many readers may have assumed that the user was engaging an AI tool that had full access to their entire email history and possible even the entire email history of the entire company. But if the AI only has access to a single user’s inbox, that is a Limitation that might impact the editor’s effectiveness. An AI tool developed in Silicon Valley may have a Weakness when applied to writing in languages other than English. Licensing limitations may restrict the number of emails a given user can actually edit. Corporate policy may prevent drive the user to a tool with lower
Limitations, Weaknesses, Costs, and Prohibitions shape how AI can and will perform in even the simplest of contexts. A full review of these characteristics when thinking through the application of AI to broader challenges can surface the ‘gotchas’ far more quickly than focusing on the capabilities.
Even if one accepts the need to focus on Limitations, Weaknesses, Costs, and Prohibitions (LWCP) when leveraging AI, there are several ways to go about the problem solving process. Personally, I am finding myself increasingly inclined to start by considering what should remain the human domain. AI capabilities are evolving so fast that if you try to define what AI ‘can’ and ‘can’t’ do on any medium or long term planning horizon, it may be impossible to even agree on the basic assumptions that will support coherent problem solving.
Zeroing in on what humans should do and why, on the other hand, is like creating a tightly focused solution core. These ownership domains and tasks will likely reflect a combination of LWCPs and only a subset of each. Particularly for short term planning, there will almost certainly be additional areas that must stay with your humans because of LWCPs that go beyond your philosophical foundation.
Therefore, starting with what you choose to keep for your humans and then layering on what you must keep with your humans are the first two steps in my current AI-integrated problem solving process. Then we get to turn our attention to how AI will actually be used most effectively to address everything else (which likely brings us right back to the first law and ensuring we have a true understanding of what ‘everything else’ actually means).
This may feel a bit contrary as I have just spent an entire essay arguing that the limits are more interesting than the capabilities but what I am actually saying is that this is a moment where characters can temporarily pretend to be authors. The only counter to the “AI will do that” narrative is to really stop and think about the LWCPs that matter to you and building your own future around them.
AI is a given at this point. The questions are who, when, and how. Belle’s Second Law is the second plank in a foundation for answering those questions.
Check out the rest of the series:
Previous
