Today’s post continues the Mistakes of Mainstream Management (MMM) series and explores how lack of understanding of Artificial Intelligence (AI) can lead Tech. executives and entrepreneurs to waste money and resources on misguided projects and investments.
A simple shift in perspective can many times lead to profound insights. Turing’s test is a test given to an “intelligent" machine to check if its intelligence is indistinguishable from that of a human. Let me being today’s post with a pertinent quote by Heinz Von Foerster:
The way I see it, the potential intelligence of a machine is not being tested. In actual fact, the scholars are testing themselves (when they give the Turing test). Yes, they are testing themselves to determine whether or not they can tell a human being from a machine. And if they don’t manage to do this, they will have failed. The way I see it, the examiners are examining themselves, not the entity that is meekly sitting behind the curtain and providing answers for their questions. As I said, “Tests test tests.”
Table of Contents
Knowledge without Understanding
Knowledge without understanding is a misguided missile.
At some point all of these abstractions about knowledge and understanding might break down, but the point I’m trying to make can be explained with humanity’s battles with scurvy.
Scurvy, a devastating disease has plagued sailors and those with limited access to fresh food. For centuries, various cultures stumbled upon effective remedies for scurvy, ranging from evergreen bark to citrus fruits, and leafy greens. Sailors and explorers observed that these foods could prevent or even reverse the gruesome symptoms of the disease. However, this knowledge remained largely empirical and was not underpinned by a good explanation. This meant that while cures existed, their application was inconsistent, often doubted, and easily undermined by flawed theories about the nature of the illness.
The absence of understanding, specifically the recognition of vitamin C deficiency, led to tragic consequences even though we had the cure on hand. Physicians and ships' surgeons often unintentionally eliminated all the vitamin C from known vegetable cures by attempting to incorporate them into more complex medicines, such as by boiling herbs or crushing them for their juices. Vitamin C is delicate and easily oxidized. By frequently destroying the active component of effective remedies, medical practitioners and mariners sometimes cast doubt on genuine cures.
If you are curious to learn more, I’ll point you to this terrific post by Anton Howes where he explains this in detail. I’ll end this section with a pertinent quote from the post that drives home the point:
It is possible to discover that a thing works, and even to use it for hundreds upon hundreds of years. But without knowing why it works, its potential will never be realized. In fact, the lack of understanding can make a perfectly useful method of solving a problem extremely vulnerable to being discredited and lost. It can lead to all sorts of fads for solving the problem that will come and go — some actually effective, but often not. It’s something to ponder for those trying to improve things at the very edges of our understanding today.
Big Data vs. Pandemic
AI and Large Language Models (LLMs) in specific are trained with “Big Data”. But as Rory Sutherland regularly reminds us that all Big Data comes from only one place - the past!
The only thing you can reliable tell about the future by carefully studying the past is that it is unpredictable. A terrific example of this pitfall came during the pandemic. Many airlines had to shutdown their pricing algorithms and switch to manual pricing. Here’s why…
The airlines had originally trained their algorithms in a time when if you drop the price of a flight, more people wanted to fly. But, suddenly in the pandemic they found themselves in a situation where that rule no longer applied - 2% of people who wanted to get home would pay almost anything to get on an airplane and 98% of people wouldn't board a plane, even if you put a gun to their head.
So, pricing flights in 2020 as though it were 2018 led to complete disaster because the algorithms just kept dropping the price of flights and nobody wanted to get on board. Watch Rory explain this elegantly in this video:
A Popperian Lens on AI
Epistemology is just a fancy word for ‘the theory of knowledge’ and dives deep into to its methods, validity, and scope. It is a branch of Philosophy with various schools of thought - as much as I'd like to discuss about them today, it will be major digression to the topic on hand. For example, Michael Polanyi put forward the notion that all knowledge is personal. From this perspective, one could argue that “artificial intelligence” itself is an oxymoron. But, for today’s post I’ll take the lens of Karl Popper’s epistemology.
Karl Popper, one of the 20th century's most influential philosophers of science, challenged the traditional view of knowledge acquisition. He rejected the idea that we build knowledge through induction – accumulating observations and then generalizing from them.
Instead, Popper proposed that knowledge progresses through conjectures (constructed by our mind) and refutations. Here's the core idea:
Conjectures: We create theories (guesses, hypotheses) about how the world works. These aren't derived from data; they are imaginative leaps.
Refutations: We then test these theories rigorously. The goal isn't to prove them right (which is impossible), but to falsify them. A theory that survives repeated attempts at falsification is considered corroborated – the best explanation we have so far, but always open to future revision.
Good Explanations: David Deutsch, the Father of Quantum Computing, building on Popper, emphasizes that good explanations are "hard to vary." They aren't just any guess; they tightly constrain reality, making specific, risky predictions. A good explanation doesn't just fit the data; it explains why the data is the way it is, and could not easily be tweaked to fit any other data.
In essence, knowledge isn't passively received; it's actively constructed and then rigorously tested. It's a process of creative problem-solving, not data processing.
Popper pointed out how relying on inductivism and empiricism can be dangerous. Many people in Tech. continue to work as if knowledge is “induced” from observation of past data (e.g. Sun rises in the morning).
They proceed first to collect lots of data, induce a general “theory” from it and hold the position that more data further proves their theory. But, this is mistaken. Karl Popper elegantly made the following argument:
...no matter how many instances of white swans we may have observed, this does not justify the conclusion that all swans are white.
This underpins Popper’s work on falsification. AI models are just induction machines. An AI model, does not generate new knowledge as people do.
Subscribe to "I'm Serious" to read the rest.
Multidisciplinary Insights the improve the effectiveness of Tech. Practitioners, Executives and Entrepreneurs!
Let's go!A subscription gets you:
- ✅ 𝐀𝐜𝐭𝐢𝐨𝐧𝐚𝐛𝐥𝐞 𝐢𝐧𝐬𝐢𝐠𝐡𝐭𝐬, real-world examples, reusable templates and more!
- 👩💻 Online access to the premium content archive!
- 🤩 Unlock ability to interact with Comments, Surveys, etc.
- 💡 Multidisciplinary insights for passionate human-centric 𝗲𝗻𝘁𝗿𝗲𝗽𝗿𝗲𝗻𝗲𝘂𝗿𝘀!
- 💸 Survive-and-thrive guidance for post-ZIRP era 𝗺𝗮𝗻𝗮𝗴𝗲𝗿𝘀!
- 🎉 A new way to think and lead organizations for "systems" aware 𝐞𝐱𝐞𝐜𝐮𝐭𝐢𝐯𝐞𝐬!