Don’t shoot yourself in the foot with AI
There is a lot of discourse on how using AI can make you “dumber” somehow. At the same time using AI can make you - if used right - obviously more productive. A lot.
So how to master both: Staying “smart” (we will later make this more concrete) and being more productive by using AI? In the last few days, I’ve been thinking (wow I can still do that although I am using AI for over 2 years heavily) about that and I have found a framework which - at least for me - explains how to stay “in charge” of cognitive work, to keep the most relevant cognitive capabilities strong, while getting the performance boost which AI offers.
It also explains how you can shoot yourself in the foot and weaken these capabilities. I have no doubt that by using AI incorrectly you can really hurt your ability to solve cognitive tasks yourself efficiently and effectively, which is bad for many reasons. Even if you only care about productivity (and not about health) that is a bad outcome which will make you less valuable as a colleague or employee.
The nature of most cognitive tasks in work contexts
When you are doing cognitive tasks in a work context what is it you do in the most generalized way? You are searching for a solution s which solves your problem p in a good way. To evaluate how good the solve is you have to apply some evaluation function v giving you a value between 0 (no solve at all) and 1 (perfect solve).
Example 1: You can solve problem p, slow client onboarding, with solution s1: hiring a partner success manager, or s2: automating platform configuration. Your estimated evaluations of these solution options could be v(s1)=0.8 and v(s2)=0.5.
Example 2: The problem p of an uncomfortable office kitchen can be solved with the solution s1: buying a couch, or s2: removing the soda fridge. Your estimated evaluations of these solution options could be v(s1)=0.3 and v(s2)=0.9.
By searching for possible solutions in the solution space and estimating the value of these with respect to solving your problem at hand you are optimizing solution value, thus solving your problem in the best possible way.
This is what you do, all the time, probably without being aware of it. That is how the mind works, if you talk about it like this or not.
The value of understanding the three pillars really well
If you want / have to solve a problem you have to understand these three pillars very well:
The problem itself. Its nature, its properties and the the reasons why it is a problem and for whom.
The space of possible solutions (aka solution space). This includes really everything that can be a partial or complete solution for the given problem.
The evaluation function. You need to have some notion of how to evaluate how well a given solution solves the problem. Most of the time this involves multiple factors you have to consider, like multiple stakeholders that have to be happy, legal, functional, social or technical requirements that need to be fulfilled, feasibility, viability etc.
Based on that solid understanding you can look at different options s in solution space and evaluate them, finally picking the one with the maximum value v(s).
One very crucial part of solving problems is the execution, which comes after picking the winning solution s. In order to execute efficiently and effectively the understanding of the three pillars is crucial. If your understanding of the problem, the solution approach or the evaluation function which was used to pick the solution is shaky, the execution will very likely be ineffective in solving the problem. You will be an underperforming colleague or employee, or president to use a vivid example. Keep this in mind.
How to shoot yourself in the foot with AI (boom 💥)
Given the framework defined above we can identify at least two ways how using AI can hurt you:
By using AI to understand the problem, the solution space and the evaluation function for you there is a real chance that you lose the ability to construct such understanding yourself. This ability is a bit like a muscle. If you use it, you get better at it. If you do not use it, you will get worse at it. This makes you useless when AI is not available and given that AI is not as good as humans (yet or maybe forever) in certain high level creative thinking, this will make you worse at your job even if AI is around.
If you do not make sure that while working with AI you stay on top of understanding of the three pillars - you are in charge cognitively - there is a high risk of ineffective solution execution. A great solution approach will be butchered by dumb, unaware execution. This will also make you worse at your job.
How to not shoot yourself in the foot while being more productive
Based on the above derivation on how to shoot yourself in the foot we can say: do not do the above and thou shall be fine. But what does that mean concretely when working with AI? Let’s give it a try with an example recipe, or an exercise. Following this strictly in daily work is not practical but it is a good exercise to improve mastering the “three pillar framework”. Once you have it internalized the fundamental success factors will naturally be part of all your work with AI.
Exercise
You first build yourself a really good understanding of the problem, the solution space and the evaluation function yourself. You can use web search, books, talking to people, drawing, also talking to AI while validating its output (AI boost 🚀) for individual aspects of any of the three pillars in an isolated way - the usual first research phase. You get to a state where you are very confident that you have a solid understanding so that you could describe it to another person really well.
Now you provide a detailed, holistic description of the three pillars to an AI and ask for potential improvements on them. This can make your problem solving framework better (AI boost 🚀). Maybe there are blind spots or errors in your problem understanding, the solution space mapping or the evaluation function. Iterate on this, validate it (super important!), think about it, extend your understanding. The final state should be that you again can explain this all to another person really well, because you really understand it all.
Now that the framework - the three pillars - are as good as they can be, it is the time to ask an AI to identify possible solutions and rate them using your evaluation function (AI boost 🚀). Use this as input into your own solution search process, challenge and validate the suggested solution options and evaluations. Iterate on this, ping pong with AI, colleagues and in the end make the call yourself on which solution to pick. Reach a state in which you can explain the whole thing, the three pillars, all relevant solution options and their evaluations to another person really well. You should ideally do that for a final QA step anyway.
Execute with the joy of knowing you have a really good shot at solving the problem well and being able to turn it into reality effectively. If this is about coding software, or writing text, or other things AI agents can produce, you can use such AI agents in execution as well obviously (AI boost 🚀).
Aha!
I hope this post has a positive effect on your work with AI. Be in charge. Train your cognitive muscles.
