You Think AI is Intelligent, Actually It's Just Guessing

The AI marketing machine doesn't want you to know something. It's a secret that's been around since WWII, maybe even earlier.

Up until the past 10 years, most computer software you dealt with always behaved the same way. If you gave it a list of employees and asked it to do payroll for those employees, it would do it. The same way. Every time.

The main place you would find computers doing things unpredictably back then was in video games. The clever games would cause the enemy characters to sometimes behave in random, unpredictable ways to keep the game fun.

But then randomness started to sneak in on all the big websites. You started to see "Recommended Productions" on Amazon, and these products were different for you than for everyone else. And they rotated in a way that you couldn't exactly explain. Then recently, when you chatted with ChatGPT you asked it a question that you had asked before but it answered differently.

Since the beginning of computers there has been a divide between processes that are the same every time as long as the input is the same and processes that are fundamentally based on randomness. While the processes based on randomness have existed from the beginning, they were not at the heart of most computer applications you used.

Let's put some specific language to these terms so we can talk about them more accurately. The processes like payroll, which happen the same every time given the same input, are called deterministic processes. The processes which are based on a random roll of the dice, like video game enemy behavior, are called stochastic processes.

Deterministic Processes

At the beginning of the computing revolution, most of the tasks computers were replacing had to do with stable processes and calculations that needed to come out the same every time. They were deterministic. The initial ENIAC computer, for example, was developed to calculate firing tables for American WWII artillery at different ranges.

As the years went on and the Internet arrived, we continued to focus on making sure the output from computers was consistent. Websites aimed at presenting information. When you logged in, usually the system just saved your information, possibly calculated some reports, and then perhaps helped you to place an order or send an email. Not much of what you did required randomness.

Stochastic Processes

Except for computer games. They were so boring if the enemies did the exact same thing every time. Imagine if the space invaders always came from exactly the same part of the screen. To keep things interesting, game designers had to constantly apply randomness in careful ways that would keep the game consistent but fun.

As the internet matured, companies wanted to give their customers better service. So they started using customized processes on their websites such as "Recommended Products" sections, which were not standard but were based on a guess of which products the customer might like based on something known about the customer. Making this guess in a computer is harder than you might expect. There are many ways to do it but they all come down to rolling some virtual dice to pick the most likely products you will be interested in based on some information known about you.

Fast forward to today. LLMs like ChatGPT are entirely built around picking specific words in a response based on a huge set of probabilities. You can ask it a question like "Plan me a trip to Naples, Italy" and it will be able to give you a response. It can do this because it's simply rolling the dice for each word in the response. The plan may not be amazing, but at least it can give you an answer.

The Fundamental Difference Between AI and the Software You Grew Up With

Most people think the fundamental different between AI and traditional software is that AI is intelligent. This is not entirely correct.

The fundamental difference between AI and traditional software is that AI is stochastic. This means that it's able to solve a whole different set of problems that traditional "payroll system" software simply cannot.

The Shocker

But this also means something AI companies don't talk about much. In most of life, we don't like randomness. In fact, most people attribute the success of McDonald's and Starbucks to their ability to pull off consistency in their products across geographic locations and even other countries.

This means that in software automation cases where you want consistent results, choosing to give a process to AI is stupid. It's stupid because you want a consistent result and AI can't give you that.

This means that...

So when you apply AI to a particular step in your business, you should ask yourself, "Do I want a deterministic output or a stochastic output?"

If you are doing something that needs consistency, such as generating a quote or payroll, this should most likely be a deterministic process that uses traditional software and not AI.

If you are doing something that needs some flexibility, such as summarizing a quote for an email to a customer, then AI is your choice.

AI is not going to replace the traditional software and automations you already have in place. What it does is add an ability to handle new situations that the traditional automations couldn't.