Op Ed: There’s Something About Artificial Intelligence… Maybe We Let the Robots Inside
Old Man Yells at Cloud
That’s me in the meme up there, or rather, It was me. A few months ago I made a blog post that was critical of AI usage. That was the old me. That was a me who was frustrated by successful Vibe Coders and AI art. I fell into the algorithms that displayed content from newbies looking successful while solely using AI agents. That, to me, seems like a cop-out. I’ve grinded. I’ve shed blood, sweat, and tears throughout my career, and these newcomers with their fancy tools are tackling tasks in timeframes that I can only dream of.
(not being dramatic. I took a server to the knee in my early years. Still have the scar to prove it!)
But, I’ve changed my tune. AI is powerful. AI is useful. AI is #1.
So, What Changed?
Two major things have changed my mind.
- I’ve acknowledged the changing times, started studying for AI certifications, and have begun learning the ins-and-outs of how AI actually works. All of the buzzwords involving AI are being ingrained in my brain; from LLMs, AI agents, generative AI to ANI (artificial narrow intelligence), you name it, I’m brushing up on it.
- I recently purchased a new electric vehicle that has supervised self driving capabilities. Put your pitchforks down! This isn’t the place to discuss politics. I’m more interested in the technology and cost savings. This beast runs on two AMD processors running a neural network. Its self-driving capabilities are amazing, and its a way better driver than I am.
Explain Yourself
While studying the ethics and depths of AI, I’ve come to the personal conclusion that AI should be used to advance us further. It yields faster results and derives information from a much larger collection of information that I can even fathom. It provides mostly accurate solutions.
I’m a big automation guy, if you haven’t noticed, and the more workflows I can automate the better. My automations free me up to write this blog after all.
Now, I’m not talking about replacing human input. We still need data scientists and AI engineers to do their due diligence, manage infrastructure, and train our AI. Additionally, since we are “automating the boring stuff”, let’s improve overall efficiency in the workplace and at home by leveraging AI tools.
Let’s Get Educational
So, AI isn’t this big scary terminator villain that alarmists make it out to be. In simple terms, AI takes input (A) runs it through a neural network and outputs (B). This A to B workflow is kind of basic in most cases, but what the hell is a neural network? It’s a collection of calculations, called neurons, that are given weight. In a simplistic example, think about what happens in your mind when you are making a decision.
“Should I go to bed now?” (Input/A)
Break it down, now.
My mind is calculating the pros and cons of going to bed. I can stay up and finish this blog, or I can pause it and go to bed. Each train of thought is calculated and given a weighted value.
Ok, what’s the pros and cons of finishing this blog? Again, calculated and given weight.
This happens over and over against the collected data until the neurons in our neural network decide it’s best to stay up and write some more, or more technically; they weigh the calculated values and the winning value is passed onward to the output.
Now here’s the thing, our minds are pumped full of information that can be recalled instantaneously. AI needs that information pumped in, and information is just data given context.
So in our neural network example, I have all of the information needed to make an accurate decision. My mind has calculated each outcome, considered how sleep deprived and grumpy I’ll be tomorrow, weighed all options, and the final neuron says:
“Stay up and keep writing” (Output/B)
But How Does It Understand Me?
The most common example to use is an AI agent like Microsoft’s Copilot or OpenAI’s ChatGPT. When you enter in a prompt, each word is given a token, or a numerical value.
Think about our previous prompt “Should I go to bed?” There’s not much weight behind those words, so lets engineer a better prompt:
“Should I go to bed, or should I finish this blog?”
Each word in this prompt is given a token number.
Some things to note first off:
- Repetitive words are given the same token — i.e. “I” and “should”
- Words that are used in common english language to flesh out a sentence, like “to” and “this” will be ignored. This is called stop word removal, and filters out words that don’t need calculated into the equation.
- Punctuation is also filtered out.
So our prompt is transformed into: “Should go bed or finish blog”.
Now, let’s discuss vectorization with these words. What the hell is vectorization?
Vectorization, or vectoring, is the process of classifying and organizing non-numerical data such as text or images. Each word in our example prompt is assigned a three-dimensional numerical value, called a vector. These values relate to values of other dictionary words, and in turn strengthen the AI’s confidence in the context of the prompt.
RAG, or “retrieval augmented generation” is the meat-and-potatoes of AI. This workflow is what queries the language model and retrieves data in relation to our prompt. So in our example, maybe our AI agent reads the word “bed”, relates that to “sleep” then queries the local timezone to see that it’s 2:00 AM, then it queries what day of the week it is and discovers that it’s a work night, then it queries what’s the recommended amount of sleep required for a human and what are the side effects of sleep loss. Then it will query “finish”, categorize that as accomplishment, retrieves data regarding the dopamine rush that is fulfilled in the human brain during task accomplishment. Cutting to the chase, our AI agent compares the calculated values between a full night’s rest verses dopamine rush via task completion.
AI isn’t some magical thing, it’s math, and boy am I bad at math. All AI does is assign values to the input (user’s prompt), reads more numerical values (structured data), then outputs information to the user (system’s prompt).
Still Not On Board?
There’s many types of AI, but let’s focus on two: generative AI such as AI agents like Copilot or ChatGPT and artificial narrow intelligence (ANI) like the neural network running in my vehicle.
Use Cases
Generative AI: A corporate graphic designer needs stock photos for a slide deck. That graphic designer can engineer a prompt in the AI agent of their choosing and get the desired images output to them in a timely manner. No longer does that graphic designer need to burn their irises manually scrolling through photographs only to have to fork up some dough for the usage license. This frees up time for the graphic designer for the next project.
Generative AI: An IT engineer has no idea how to build an Always-On VPN. The engineer can ask an AI agent for materials to review for understanding the technological components better and for building their network expansion instead of scrolling through dozens of unanswered StackOverflow and Reddit posts.
ANI: A radiologist reviews x-rays in-between speaking with patients. By using an ANI system that reviews x-rays for fractures or infections, the radiologist has more time to consult with patients.
ANI: A self-driving vehicle which uses image detection to determine placement of the vehicle on the road, distance between the other vehicles and obstacles, and analysis of traffic signals. A complex concept such as driving can easily be replicated by ANI with enough data training. Because who really enjoys a 30+ minute commute to the office?
But There’s a Catch!
Notice how not one of my examples express the replacement of human input or supervision. AI, much like human input isn’t perfect. There’s still a need for review. The self-driving vehicle, when left unsupervised, can put human lives at risk. The radiologist still needs to double check the AI’s output to ensure a proper diagnosis. The IT engineer needs to review the study material to ensure it is relevant, or else they’ll waste time and company resources by reviewing unrelated documentation. The graphic designer can risk the company’s brand and profits by using a graphic that’s a poor representation of the company’s image.
The famous saying goes “garbage in, garbage out” meaning ambiguous or faulty data from the input leads to a skewed or corrupted output. There’s still a need for human supervision for either cleaning up the data to be ingested by the AI system or for performing quality control on the output.
Wrap It Up Already
AI is a powerful tool. Many analysts think it’s just a fad, but the benefits to this new type of automation are lucrative. While the market may be bloated right now, and corporate buzzwords keep popping up out of nowhere, AI is here to stay. Gone will be the days of this blog showing you how to automate a task in PowerShell because AI can automate the automation.
It’s time to adapt. The future is here.