Reasoning from First Principles for the AI-enabled Enterprise

Reasoning from First Principles rather than from analogy is important for new breakthroughs according to legendary businessman Elon Musk. Before we apply the concept to AI, let’s cover the basics. Most people conduct their lives by reasoning from analogy. We see someone else doing something and we copy; even if we give it a slight iteration on the theme. 

Why? Because it’s mentally easier to reason by analogy than from First Principles.

First Principles is a way of looking at the world by boiling things down to the most fundamental truths. Figure out what you know is true, or are reasonably sure is true, and then reason up from there. This takes a lot more mental energy.

Elon uses battery packs as an example. You might say battery packs are really expensive and that’s just the way it is, and that’s just the way it will always be. If you apply this analogous reasoning you’ll never be able to get to the next new thing.

Historically batteries cost about $600 per kilowatt hour. You might assume it’s not going to get much better than that. Reasoning from First Principles you’d ask what are batteries made from? What are some of the material constituents of the batteries? Cobalt, nickel, aluminum. Add some carbon and some polymers for separation, and steel. Break it down to its basic components. What would each of those things cost if you bought them on the London Metal Exchange? What is the spot market value?

You’ll discover the actual cost is $80 per kilowatt hour. Next you think of clever ways to take the materials and combine them into a battery cell. You end up with batteries that are cheaper than anyone realizes. And that is the power of reasoning from First Principles.

How do you apply First Principles to Artificial Intelligence (AI)? Much of what AI does mimics what humans do; learning (machine learning), listening/speaking (natural language processing), optimizing (prediction), moving (robotics), and seeing (computer vision). This means we need to understand how computers learn differently from humans.

Break this down to the First Principles of Data. At a basic data science level, analytics produces insights from data, Machine Learning predicts, and AI performs actions based on the percentage likelihood you’ll get the result you're looking for.

This is important for two reasons: First, there’s a lot of hype on what AI can and cannot do based on analogous reasoning. Second, if you’re basing your AI implementation solely on best practices without breaking up your problem into its component parts, you’re simply copying what someone else is doing. You might argue, Why reinvent the wheel? And you’re right, benchmarking and case studies should inform your strategy at some level. But when it comes down to being a truly innovative AI-enabled enterprise, committing the time and resources to First Principles will save you a lot of headaches in the future.




This is the place for accelerating your career insights on trends in AI and its influence on our work world. Develop a road map that gives your career a strategic advantage. Be ready for the jobs that don’t even exist yet. Sign up now. It's free.


Comments