Beyond Substitution: When the Right Answer Is Something Completely Different
One year for Thanksgiving, my mom was craving lemon curd. Traditional lemon curd recipe is made with sugar, butter, and eggs, which is a problem when you need to make a dessert that is SCD/paleo, vegan, and diabetic friendly.
Custards and curd desserts get their thick and creamy texture as the egg proteins set with heat. I could substitute the sugar for honey and make the recipe SCD compliant but not vegan. Alternatively, I could substitute the eggs and butter with vegan substitutes, but it's unlikely that the curd would set because the vegan egg sub won't set in the way that real egg proteins would.
In the end, I pureed dates, cashew yogurt, lemon juice, and lemon zest together to create a thick, creamy, lemony dessert that satisfied my mom's craving and was good for the whole family (this recipe is actually amazing if you add a touch of orange flower water and cardamom).
What I made that day wasn't a substitution. It was a completely different recipe that arrived at the same place. As I've been building out the MVP of my diet bot, I keep running into these edge case scenarios where there's no simple one-for-one ingredient substitution. I've grouped these problems into three categories: structural, cultural, and cross-contraint conflicts.
Problem 1: Structural Conflict
In the structural category are things like puff pastry where there's no SCD compliant version because nothing will behave in the same way that puff pastry does.
Problem 2: Cultural Conflict
For the cultural conflict, you have things that a native eater simply wouldn't consider. For a hamburger, you can replace the bun with lettuce, because it is very common to eat a hamburger with lettuce. Compare this to a hot dog. You'll never see a hot dog wrapped in lettuce because no one ever eats hot dogs with lettuce.
Problem 3: Cross-Constrain Conflict
Then we have the cross constraint conflict which goes back to the lemon curd recipe that has to be both SCD and vegan and diabetic friendly. No single ingredient swap satisfies all three constraints simultaneously. The conflict between them is what forces you out of the substitution framework entirely.
Final Resolution: Convergent Solution
The ultimate solution is a whole new recipe. I'm calling this a convergent solution because you converge on the same idea, but through a completely different mechanism. Sort of like convergent evolution.
The same three problems occur in a lot of other areas outside of the kitchen. Let’s consider two more examples.
First Parallel
Let’s say a friend of yours has recently been diagnosed with cancer. There could be several problems in their treatment protocol.
Structural conflict: a diffuse systemic cancer can’t be treated with radiation.
Cultural conflict: the patient refuses traditional treatment (e.g., Steve Jobs and pancreatic cancer).
Cross-constraint conflict: the patient has a contraindication for chemo medication and cannot use radiation.
Convergent solution: abandon both chemotherapy and radiation and instead try CAR-T cell therapy. This therapy reprograms the patient's own immune cells to attack the cancer, with no drugs or radiation involved. Your friend arrives at remission through a completely different biological pathway.
Second Parallel
Let's say you're in business with a partner and you need to sue this person.
Structural conflict: statute of limitations (technical limitation)
Cultural conflict: suing your own business partner (technically viable, but socially catastrophic)
Cross-constraint conflict: you need a resolution that is both speedy and binding. The court will result in a binding resolution, but it won’t be speedy.
Convergent solution: abandon the courtroom mechanism entirely and instead rely on arbitration. You can quickly arrive at a binding resolution through a completely different process.
Conclusion
There is an example of this sort of convergence solution in every single industry. AI will make things faster, but it seems hard to believe that AI is going to come up with these sorts of convergence solutions because it requires such long-lived experience.
What in your own work isn’t a simple substitution? Where have you arrived at a completely different mechanism that gets to the same place, through accumulated experience? That’s the knowledge worth capturing and it is extremely difficult for any modern AI system to extract.