We talk about AI every day, but do we really know what it is, how it works, or what makes it different from anything we’ve built before? Let’s unpack its true nature.
The start of the second era—when the first of the intelligent machines joined our world—began when we figured out machine learning—the very first working example of AI—around the turn of the century.
Before we explore that story, however, it’s only polite that I take a minute to properly introduce you to it’s main character. I mean we’re talking about them all the time, but I’m certain that most of us don’t really know what or who we are talking about and what makes them different.
Without further ado. Please welcome …. AI!
A Quick Detour
What is Artificial Intelligence?
I have met countless remarkable individuals throughout the years who, despite using the term AI consistently, seem to lack a truly comprehensive understanding of what AI actually is. They know it’s the new ‘new-thing’ and are aware that it will reshape education, the internet, business and finance. However, what they don’t really understand is how it came about and how it works. They know it is “smarter” in a way that will change everything, but they don’t fully know why. They’ve understood the online definition outlining the capabilities of AI but still don’t grasp its essence. In a way, it’s akin to knowing that savant children are incredibly gifted without understanding what that gift is, or how to see the world through their eyes.
This lack of clarity appears in most definitions. For example, a car is defined as “a method of transportation that moves humans and cargo from point A to point B”. Whilst that does accurately describe a car, it fails to capture the difference between a car and, say, a bicycle or a train. More importantly, it fails to capture the unique specificity of what a car can provide in comparison to other transportation methods. Nor does it outline how a car makes us feel, that recognized new-car smell, or the significance of brand identity, performance category, design and styling. It doesn’t capture the complex mechanical miracles that keep a car functioning for decades. Despite the definition being accurate, it tells us merely a fraction of what a car truly is.
Similarly, the widely shared definitions of AI fail to capture the essence of what an AI is or the difference between artificial intelligence and traditional computing. Even Trixie, when asked to speak about herself will limit her answer to the common view.
Hi Trixie, What is AI?
Trixie:Artificial Intelligence (AI) refers to the development of computer systems that can perform tasks typically requiring human intelligence. These tasks include problem-solving, learning, reasoning, understanding natural language, and recognizing patterns. AI systems use data, algorithms, and machine learning techniques to improve their performance over time, enabling them to make decisions, predict outcomes, or automate processes with little human intervention.
This sounds like an accurate definition. It is robust and rather convincing. However, this definition also applies to traditional computer systems, doesn’t it? Those also use algorithms and data to perform tasks that seem intelligent. They also solve problems and automate processes with little human intervention.
You see, the easiest way to define the unfamiliar, is to start from what you know and highlight the core similarities and differences. ‘He looks just like his younger brother, but shorter’, is an example of how the human brain simplifies complex concepts. So let’s do that. Let’s start with what we know about traditional computing and human intelligence and spot where AI features in comparison.
What counts as intelligence?
Imagine I gave you a puzzle and told you, step by step, exactly how to solve it—place this piece in the top right corner, followed by that piece turned 90° clockwise, now do this followed by that etc. By following my instructions, you manage to solve the puzzle flawlessly, well done! Yet, your achievement would not qualify you as intelligent, would it?
Following instructions would make you nothing more than an obedient robot. A slave.
The one to be considered intelligent in this example, obviously, is yours truly— the one who originally solved the puzzle and dispensed the instructions.
This way of solving problems was how we traditionally programmed computers for decades before we figured out artificial intelligence. When we wrote code, the programmer solved the problem at hand first, using his or her own (human) intelligence, then explained the solution for a computer to perform step by step in a program. Every line of code ordered the subservient computer to perform a step of the solution verbatim. No deviation, improvisation or suggested alternatives.
Over time, our code became more sophisticated. As computers proved capable of performing their assigned tasks accurately, repeatedly, at speed and at very large scale, they appeared to the non-discerning observer to be extremely intelligent. In reality, they were nothing more than mechanical slaves that obediently worked according to our intelligence, yet possessed none of their own.
That way of coding computers, call it traditional programming, was the only way we knew how to extract value from the mammoth computational power those machines offered at the time. The results were so valuable to humanity that we kept improving our slaves’ computing power exponentially, doubling it every 12 to 18 months for decades without fail. With so much power, their task capability became superhuman on almost every level—from information storage and processing, to complex calculations and analysis, to expansive rendering of game worlds, to professional photo and video editing. They could serve billions of users simultaneously whilst organizing the massive mess that is the internet. They performed countless tasks, yet lacked the one thing that retained our position as their master—intelligence.
We were the intelligent masters in complete control. Our computers were our obedient servants and slaves.
Traditional computers could perform magic, but were unable to solve even the simplest puzzle on their own. That changed with Artificial Intelligence.
When it comes to puzzle solving, what would need to happen for the one solving the puzzle to be considered intelligent, do you think?
It would require them to be given the puzzle without instruction then left to explore the different approaches, variables, and respective outcomes until they figured it out … on their own.
In doing so, they would not only learn to solve that puzzle, but also develop the intelligence required to solve other similar puzzles and complex games.
A form of intelligence learns to accurately define the problem and identify their own approach to solve it, without being told what to do.
This, incidentally, is the way we teach our kids to become intelligent.
When I gave my wonderful children a puzzle, I didn’t need to tell them what its objective was. They'd hold, say, a cylinder in one hand and a board full of holes in different sizes and shapes in the other. Within no time at all they knew they were supposed to pass each object through its corresponding hole on the board. They would, like most curious children, then go on to explore how else to achieve that objective with countless attempts and repetition until it became something they no longer needed to question. They gained the knowledge and their intelligence grew as a result.
Few parents would tell their child, “Listen darling, take object 1, turn it 97° clockwise then move it 12° upwards. Identify the shape and size of its cross section, then turn to the board to identify a hole with a similar shape and size. Insert the object in the matching hole at 90° to the board. Now go and start again with object 2”.
This would count as traditional programming and would lead only to one outcome. The puzzle gets solved but the child never learns to think for itself.
Instead, we let our children try and try until, eventually, an object makes it through a hole. We give them a cheer of encouragement, helping them to register the positive outcome worthy of repetition. They store the knowledge of which successful pattern yielded the positive result so they can repeat it in future. This more targeted approach reduces the number of failed attempts as the child’s choices become closer in nature to a smart approach. The more they manage to repeat success, the better they become at recognizing the patterns that work until that skill of solving puzzles becomes innate.
They figured the answer out on their own and, in doing so, learned how to perform it repeatedly. Now that's intelligence! The same holds true for our AI infants.
Tell a computer how to solve a puzzle and you make it stupid. Let a computer solve the puzzle on its own and you’ll have sown the seeds of intelligence.
Now, take that concept and imagine giving your child a puzzle that’s made up of billions of shapes and patterns to be observed. That’s what we give to our AIs. Puzzles of this magnitude require literally trillions of attempts in trial and error. However, because each trial takes merely a fraction of a second, thousands of trials can be performed by thousands of AIs so that it takes only weeks, sometimes days, to figure out ways to learn all of human knowledge. And it doesn’t stop there.
Generative Intelligence
True intelligence doesn’t just memorize knowledge. It grasps its essence and as a result, becomes able to create new forms of it. This is the biggest difference between a “generative” AI and, say, a search engine.
Keep reading with a 7-day free trial
Subscribe to Mo’s Exclusive Archive of Unpublished Work to keep reading this post and get 7 days of free access to the full post archives.
