Over the years I’ve followed Charlie Munger. He’s considered to be one of the greatest investors and thinkers on decision making in the past century, and spent six decades as Warren Buffett’s business partner. He often quoted a 19th century German mathematician called Carl Jacobi, whose advice for solving difficult problems was “invert, always invert”. In other words, don’t ask how to succeed. Ask how to fail, then avoid doing those things. Munger applied this logic to investing with consistently successful results. He summarised his philosophy as: “It is remarkable how much long-term advantage people like us have gained by trying to be consistently not stupid, instead of trying to be very intelligent.”
I’ve been wondering how this would apply in the context of AI. A lot of the questions ask how you can get more out of it, automate more and do more with less. Munger advises you to invert. “Don’t ask how AI can help you think better, ask how AI can make you worse at thinking, then watch out for those things,” he says.
The answers are rather specific. A paper published in January this year by researchers Steven Shaw and Gideon Nave at the Wharton School called it ‘cognitive surrender’. They ran a series of experiments asking participants to solve logic and reasoning problems, with optional access to a chatbot. More than half consulted the AI tool. The interesting part was what happened when the chatbot was wrong. Participants accepted incorrect AI answers around 80% of the time, and rated their own confidence in those answers nearly 12% higher than people who had reasoned through the problems themselves. The AI made them wrong, but certain that they were right.
Shaw and Nave distinguish this from ordinary ‘cognitive offloading’, which we do when we use a calculator or set a reminder. Offloading delegates a task but leaves the reasoning intact. Cognitive surrender is when AI stops assisting your thinking and replaces it without you noticing.
This is the paradox of AI. The more capable these tools become, the more valuable it is to use them; and the more we use them, the more our own independent reasoning can weaken. When AI takes over your thinking, it just gives you a confident-sounding answer and raises your satisfaction slightly without telling you why.
Engineers are already familiar with Munger’s concept. Rather than asking what could go right, a well-designed HAZOP systematically asks what could go wrong. Fault tree analysis, pre-mortems and failure mode studies are all structured inversions. Good engineering practice uses inversion in other ways too. Instead of asking how to keep a plant running perfectly, it is more useful to ask what shuts it down completely and then systematically eliminate every answer. These could be uncalibrated sensors, ignored wear indicators, skipped maintenance cycles or unauthorised code changes. If you list the failure modes and guard against them, reliable operation will follow almost by default. The discipline is to think backwards from disaster before the disaster gives you no choice. AI will suppress this process if it is used carelessly.
In South Africa we have a skills shortage and a young engineering workforce that is growing up with AI already embedded in everyday tools. For example, Microsoft’s AI Skills Program provides free training and certification to help anyone learn how to use AI. Over 1,4 million people have been trained since 2025 and 500 000 have been accredited. This number looks encouraging until you ask Munger’s inverted question: “How many of these people are being trained to interrogate it, rather than simply use it?” The Institute of Mining & Metallurgy warns that the country’s AI skills gap is not a coding gap, it’s a critical-thinking gap. Because AI programs can write basic scripts for you, the real value comes from your human judgment, problem solving and critical thinking.
Munger suggests a way forward that allows you to use AI in a smarter way rather than use it less. Before you accept an answer, invert. What would AI have to get wrong for this to be a bad decision? What failure mode am I not checking? What is the AI tool confident about that I haven’t verified myself? These questions don’t slow you down much, but they keep systematic thinking in the process, and result in a better outcome.
For engineers and anyone else working with AI tools, rather than asking: How do I get the most from AI, think what you would have to do to make sure AI gradually chips away at your cognitive ability or judgement? Write a list and make sure you don’t do those things.
| Tel: | +27 11 543 5800 |
| Email: | [email protected] |
| www: | www.technews.co.za |
| Articles: | More information and articles about Technews Publishing |
© Technews Publishing (Pty) Ltd | All Rights Reserved
printer friendly version