Practive 102: The Origins of AI Part 1

In our previous blog titled “The AI Revolution Has Begun,” we noted how the tech industry and indeed the entire investment world is wholly obsessed with Artificial Intelligence. Since then, things have moved quickly and some are wondering, “what is AI and where did this all come from?” In this article, we attempt to answer that.

Today, AI is ubiquitous. It’s everywhere. We see it in product advertising, we see it being used in the social media space to generate memes and short videos or pictures. We saw the Harris Presidential campaign in 2024 circulating AI-generated images of campaign rallies that exaggerated participation. It seems every few hours there is a new AI altered video being distributed in Social Media.

You may interact with an AI bot when calling into customer service. We know students are using AI to write essays and reports, and in kind we have seen teachers discuss how they are using it to grade student papers. AI is reportedly core to our next-generation war fighting capabilities. It’s allegedly being used to profile investment opportunities and to guide stock trades. It’s allegedly coding for us, planning for us, translating for us, choosing for us, assisting us in any number of ways. HR departments are reportedly using AI to screen and select candidates for jobs. There are even Pastors who have created AI-based simulations of themselves as so-called “prayer bots.”

All suddenly and seemingly out of nowhere.

Yet there is a common and recurring thread that users and viewers of AI generated content report: it makes stuff up. It lies. It gets it wrong. Some call it hallucinations, some call it failures. But no question, it isn’t a reliable source of the truth. Many of the images and videos created by today’s most sophisticated AI agents have unusual artifacts that make it easy to spot as “fake” content. Many of the answers AI agents provide in response to people’s common questions are obviously not true and sometimes morally questionable or even culturally taboo. Examples abound so I’ll skip adding them here.

Yet we’re using this tech to pilot space ships, drive our vehicles, diagnose medical problem, answer moral questions, and even speak to the dead as Jim Acosta did in a fake interview this week. We’re even laying people off of work and ending careers altogether thanks, allegedly, to AI.

So what is AI, and where did it come from? How did we get here? Spoiler alert: AI isn’t. Not actually. Indeed, we haven’t even defined what the benchmark criteria is which we can use to measure if we have achieved it. But there is obviously something out there being used to generate all this new content.

And this misunderstanding presents us with a huge problem: many people have formed a belief which they apply to these AI Bots, and interact with them according to those beliefs. People assume these machines are the ultimate source of knowledge and pure scientific, unbiased truth, not unlike Douglas Adams’ fictional Deep Thought. Some politicians and AI pushers even want to use the tech to make legal, governance, and war fighting decisions for us.

And as Rod Dreher and other’s have found, there is a very dark side in the use of the technology as some, like Elon Musk, are attempting to create a superhuman intelligence to rule us, or as a means to access non-human intelligence that may already be “here.”

But how can we say the AI Revolution is upon us today with all this hype, while some say the underlying tech is over a decade old and brush if off as old news? 

Co-Founder and CEO of CrowdStrike, George Kurtz likes to brag that AI is in his company’s origin story. He says, CrowdStrike has been using AI since the company’s earliest days. Well, that would be circa 2011. Indeed, Cybercrime Magazine recently ascribed Mr. Kurtz with the title of “AI Pioneer,” so he should be one to know. 

In an article published by Silicon Angle, Mr. Kurtz reportedly said “We were doing AI before it was fashionable — that was machine learning back then,” he said. “Now, we call it gen AI.” 

So what happened from 2011 - 2025? If we were doing it back then, why did it take the AI Revolution 14 years to materialize? Is this indeed new technology, or old technology?

What is AI?

This is actually a tricky question to answer, because AI is a bit of a multi-headed beast much like a hydra, including like the hydra, being rooted in a lot of mythology and unfounded beliefs in its nature and capabilities. Quite a lot of spell-casting is being done to create the myth in real time and to give this creature, “life.”

I’ll try to explain by looking at its origin story; it’s predecessor technology and what that tech stack, and AI itself, was intended to be.

What AI Isn’t

Let’s start with what AI is not so we can cut through the mythology and spell-casting.

It’s important to note that AI isn’t. At least not yet, and maybe not ever. What I mean by that is we have not achieved the creation of Artificial Intelligence as you likely would assume what that means. The things we call AI are not independent, self-aware, autonomous entities that have feelings, intuition, desires, interests, fears, awareness of time, or any other motivations we might ascribe to an intelligent being. In fact, experts in AI and in neuroscience have not yet reached consensus on a definition for “artificial intelligence,” but we generally assume it will be something like a new sentient being. Since we have no clear definition at this point in time, we cannot even measure if artificial intelligence has been obtained, even if it has been. We’re not even sure if it is technically possible. Additionally there are many who believe from a religious and philosophical point of view, that this is not even a possibility. Some say that God alone grants “life,” and we can only co-create it within the natural process of procreation; an act within a system, not an act that escapes a system. Yet escaping our system (humanity itself) is exactly what AI cheerleaders like Elon Musk and Sam Altman claim to be doing.

If you would like to dive into this deeper, I encourage you to listen to this conversation titled “The Ontology of Artificial Intelligence,” including Neuroscientist John Vervaeke, American Philosopher D.C. Schindler, and Christian symbolism master Jonathan Pageau. Mr. Pageau has several additional AI related conversations on his Symbolic World Youtube channel that are all well worth your time.

Despite the claims that AI bots are beginning to exhibit signs of having original desire and are acting to subvert controls in an act of self preservation or self-expansion, these behaviors are still explainable as expansions of their code, not spontaneous actions of sentient life. Some AI bots have been described as intentionally lying or hallucinating, however these attributes again seem to be more about our attempt to ascribe intelligence to the machines, than of them actually being intelligent themselves. 

After all, what is the difference between lying for the purpose of self-motivated deception versus making an assertive statement that is incorrect? What is the difference between hallucinating and stating something as truth, and simply being wrong? From the point of view of the machine, we can’t measure the answer to either of those, so some have chosen to believe intelligence exists where others believe the technology simply doesn’t work properly.

The closest example we have to an AI bot nearing something mirroring sentience and intelligence, is a recent experiment which provided the Agent algorithms to maintain awareness of time. With that awareness, the Agent had to include information “from the past” and consider the implications “to the future” before it provided answers. I understand that the full scope of time the Agent was able to operate within was something like hours. This was an experiment to discern if an Agent could derive some form of self interest, self preservation, or some primitive form of moral decision making.

This is why I personally prefer the term Simulated Intelligence to describe the current AI bots, because that is what they are, and technically all that they can be by today’s standards.

They are Artificial, but they are not Intelligent.

What is unique about the current AI bots that gives them the semblance of intelligence and autonomy as a being, is that they are not purely following exact lines of code based on commands or encounters, as previous robots have done. Instead, the current AI bots are performing what appears to be their own “self learning” and “creative decision making” as an extension of layer upon layer of sophisticated code that gives the AI bots a framework for understanding and interacting with data they are fed. This is the next-generation extension of machine learning using Large Language Models (LLMs). That technology was yesterday’s next-generation form of data analytics. But I’m getting ahead of myself.

The human-like interactive nature of the bots can give the impression that you are dealing with a sentient being when you ask an AI bot questions and follow-up its answers in a conversation-like engagement. However the structure of these human-like responses are based on models of human language and common interactions.

Indeed, Elon Musk has recently acknowledged that his AI platform, Grok, is coded to learn from public posts made on Twitter/X. It is by monitoring how humans communicate and what they say that the AI bots can calculate what to say and how to say it in a way that will mimic natural human interactions it has observed. This is not intelligence as you or I would naturally call it, but rather mimicry by a machine operating according to its code.

Explained

What is actually happening is that the AI bot is powered by sophisticated and extremely fast data modeling which has been combined with the ability to derive the most likely search result from the information it has access to, and to generate output in normal language or what is expected in a normal human interaction. Think of what Google is supposed to be, but at a larger scale and with results that are actually derived from analyzing and interpreting actual data.

It may be helpful to think of this as a data aggregator that elevates certain pieces of information based on the algorithms it uses to interpret the data and interpret a response.

The early mainstream iterations of ChatGPT were largely based on data modeling that performed statistical analysis of data it was fed. It “discovered” mathematical relationships between letters, words, sentences, paragraphs etc. from data sets, and applied that “understanding” to generate the most statistically probable result based on your prompt. When comparing different data sources, it simply calculated the most statistically similar and significant words, sentences, and paragraphs, and presented those in combined form. In essence, this is still what the latest iterations of GenAI are doing.

Consider it this way. In the absence of a specific definition, if you asked an AI bot to determine the chemical properties of pure water, it might lookup in its data set everything it has access to that defines the properties of various water samples. It would parse what is unique and what is common and would present you with an aggregate definition (again, dependent on its algorithms that will give it biases), likely excluding anything unique since it would be viewed as an additive not present in all samples. In the end, it may well give you a definition of pure water that contains contaminants by our standards, if all the samples it reviewed contained those same attributes.

What an AI bot needs to function is a data source, a model for understanding and comparing the data, instructions for analyzing the data, and a model for producing the desired output in the right way. All of this can be coded or started through prompts that get the AI bot running, either to perform a single short task, or an extremely complicated long-term set of tasks that mimic autonomy and being. Additionally, each of these functions can be coded to match many different scenarios; different data sources, different data types, different preferences, different things to pay attention to, different cultural nuances, different weights imposed on certain themes or words etc. But fundamentally, what we call AI today is a compute process that understands common language, translates that (via data modeling) into a command which it executes, and generates an answer in a format specified or a format it calculates as most likely to be desired.

Expanding this to image, video, and audio creation is a simple task of translating the media into strings of characters, which is what is done whenever you digitize an image.

Some AI bots can appear to be self-learning, because when you give them a prompt to find a data source, they understand what that command requires and they have pre-defined sets of instructions to perform that task and to understand the data once it is obtained.

For example, if I want a non-AI run robot to turn on the lights in my room, I will write a bit of code that instructs the robot’s sensors to identify if the room is illuminated or not, how to find the nearest light switch, and to switch the lights on if they are off. With machine learning used by AI bots, we can instruct the AI engine to understand how to recognize light vs. dark, what a light switch is, how to identify it, under what conditions a lighted room is useful etc. Then we let the AI do its thing, and it might “understand” how to switch the lights on when a person enters the room and asks it to.

Imagine this room as empty. Imagine an AI bot in the physical form of a robot. Under AI-powered machine learning instructions, it would enter the room and “learn” its attributes; illumination state, location of light switch, setting of the light switch, method for activating/deactivating the switch etc. Whereas static code of a robot would need to have all of that pre-programmed. Indeed the AI-powered robot does, just in the form of instructions rather than commands. Well, sort of.

Origins

I have often read or heard spoken of AI, the sentiment that this technology suddenly came out of nowhere and that it is a solution trying to find a problem. I recently read from a reputable source that some of the Engineers at the leading edge of the technology claim that it originated from non-human intelligence. Yes, you read that right.

I think the question of origin comes from a desire to downplay the technology, which does enjoy extreme levels of over-hype. Indeed, AI is more myth than reality which is why I often refer to the adopters and pushers as being engaged in spell-casting. There is more belief about AI than there is understanding of its actual capabilities and limitations, even of its purpose. Indeed, as I said before, Artificial Intelligence as we casually might define it, does not yet exist. Some say it can never exist. But that is another topic for another time. The (misunderstood) technology we refer to today as AI does exist and did have a natural, and needed origin, which I will attempt to review for you. In fact, it’s a technology with decades of progression behind it. Regarding possible non-human origins of thought, I’ll leave that to others to opine about.

If you described the underlying technology that powers AI engines like ChatGPT and Grok to a modern Data Scientist, they would view AI as synonymous with what we call in the industry as Big Data Analytics which is a fancy brand for an IT discipline that uses machine learning to analyze data. That fundamentally is what is behind ChatGPT and the like today.

But to understand the origin story, we need to think back to the late 90s and early 2000s. This was the period we in the industry refer to as the .com era, or the Internet Revolution that launched the Digital Age. The migration to the Internet was a truly transformational event in the human story. Prior to the Internet, everything we did existed within the physical realm and we were enabled, and limited, largely by proximity and money. If you wanted a service done, you needed to seek out a competent service provider in your immediate area. If you wanted to learn a new skill, you needed to seek out an education from a master. Both were subject to financial costs imposed that may made the desired good or service inaccessible in that way. More practically speaking and more relevant to the topic, we were physically surrounded by all things we could interact with. We could touch, feel, and interact with them live. Our reality was physical. Anything beyond our reach was unobtainable.

As the Internet grew, we began migrating and uploading the physical world to the digital, or cyber space, and it became generally available to anyone, at a more accessible cost and with few geographical limitations. Our limited physical world exploded in the digital one.

Commerce became ecommerce. Mail became email. Books were condensed into ereaders. Cell phones were replaced by Internet connected smartphones. Maps were replaced by navigation apps. TV was replaced by YouTube and streaming media. Phone calls were replaced by text messaging. Store inventories were managed digitally and often co-located in wharehouses distanced from retailers and attainable on-demand by order. Stores themselves moved online and brick & mortar locations closed up. Everything was moved to the Internet and then transformed and presented to us through web browsers and apps. Even scheduling appointments and paying for services is handled today via apps and the Internet. Education, entertainment, trade, communications, health…it’s all online now.

In the early days of transferring everything from the physical to the digital, we lost visibility and control. The operator of an ecommerce website cannot “see” their customer enter their store or browse the shelves. The customer couldn’t ask a simple question on inventory, or interact with professionals to help them make a decision. Likewise, monitoring the security and availability of products became more difficult as invisible and anonymous customers found ways to abuse our Internet systems and steal products or other resources.

We couldn’t just look at the store to see if the lights were still on. We couldn’t check the back storage room for inventory. We couldn’t build relationship with customers or ask why they chose the store or what they were looking for. And this loss of engagement and visibility didn’t just effect stores. Think of how that would apply to everything moved from physical to virtual.

At the same time, our Internet infrastructure was rapidly growing and new providers with their own competing, yet interconnected technology, were introducing unique variants that had to work seamlessly together. Also, new software was being developed, popularized, and adopted that created many different ways to do similar things in the Internet-centric new world.

It became increasingly urgent and essential that to keep this all running and working and optimized, we needed to re-establish visibility and control over all these things we were running on the Internet. Today we call this visibility telemetry, but back in the early 2000s, we just called it logging.

We had to include in our Internet infrastructure, at all layers from hardware to human, the ability to generate data that described what was happening in the digital space. For the sake of brevity, I’ll leave the details out of this blog, but understand that what we ended up doing is creating a massive amount of noise - every device, every connection, every transaction, every click, every engagement, every operation was being recorded and reported.

Once it was being generated, we next needed the ability to collect all this data. That presented data collection and storage challenges. Since the data came from disparate technologies from different providers and represented different perspectives throughout the Internet hierarchy of systems, we also had to figure out how to make the data relatable. We had to build systems that could understand and restructure or re-format disparate data so it could be joined and compared with other incoming data. This was a data normalization problem.

Then we had to figure out how to search it, find the relationships, and discover the story being told by the disparate systems who were all involved in the same end-to-end transaction, transactions that were taking place across many different virtual properties that spanned many different physical geographic locations. This problem we called data correlation.

Finally, we had to figure out how to interact with the data on-demand and present it back to us in a usable manner. This was the user interface we referred to as the presentation layer.

In the tech industry, these named challenges all have grown into their own career disciplines under the general category of data science. They include data generation, data ingestion, data storage, data parsing and normalization, data enrichment, data correlation, data synthesis, data analytics, and finally data presentation.

The fundamental challenge of Internet-based telemetry, is deriving understanding from many sources, many events, many perspectives, different versions of the story, different levels of detail, different emphases. Thus was born the industry that produced for us AI.

I was first introduced to this dilemma and the technology meant to help solve it, in 2000 through cybersecurity. By 2003 the industry had produced for us data analytics engines that could ingest, join, compare, and correlate disparate data sets to find relationships and relevance. We could search the data at-will, or create saved searches that would run continuously to find events as they occured. These correlation-based searches produced for us what we called indicators that an event of interest had transpired. Some of these indicators came late, long after the event had happened. Some of these indicators were early ones that presented an event that was still in progress. Some indicators we called precursor events, or actions that implied an event was potentially about to begin. Thus, we had some ability to have early insights into events that were unfolding over time; past, present, and future.

Through this process, we also learned that we could generate visibility and insights that were impossible before, and at speeds that were unimaginable before the digital age. We could watch people in real-time interacting with estores and Internet-based systems. We could profile and understand their behaviors. We could see their history, their location, their connections, their tendencies, their interests, their needs, and we could build new content tailored directly to meet them where they are at.

Now imagine this Internet visibility problem grows at an exponential rate; new data sources, new data, new insights, new questions, along with a growing business appetite to do something productive with the data.

In fact I recall in roughly 2003 after I was first introduced to big data analytics and data correlation, I thought of a potential future use case that is finally being realized with today’s AI. I theorized that if we could make medical records digital and if we could offer patients the ability to self-submit reports of symptoms and health statistics, then perhaps we could build correlation rules that would compare their reported conditions and experiences with known patterns of illnesses and disease, and therefore diagnose and possibly predict illnesses proactively.

Fast forward to 2025 when Elon Musk announced that Grok is analyzing MRI scans to identify abnormal characteristics that may be indicators of disease. Now, instead of calling it data correlation we call it AI.

You see, it is the same technology and same fundamental capability as we had developed in the early 2000s, which has progressed in sophistication year over year to today.

From Analytics to AI

I’ll spotlight four of the most significant enhancements that compose what we commonly refer to as AI today.

First is the hardware that the AI engines run on. In the early days, data ingestion and storage was hampered by the capacity, endurance, and speed of hard drives. I recall that when we had staged 1 TB of online storage for our data analytics engine, we had reached a significant milestone in technology. That was again around the 2003-2004 timeframe. But storage was expensive and slow, which reduced our ability to ingest, process, and access data. We had to be very careful about our read/write actions and the size of the storage arrays that had to be searched by our “real time” correlation engine. This limited our use cases of the technology, but we had ideas that are being realized now that the technology has finally caught up.

Today, cheap and plentiful solid state drives make read/write access and online/offline capacity truly relics of the past in terms of barriers we had to overcome.

In addition to storage, advances in compute capabilities performed by CPUs, GPUs, and memory have solved many of our challenges with simply processing the vast amount of data we can now collect and store.

Cloud computing has made both storage and compute even smaller problems and has collapsed the entire tech stack down to a subscription that anyone can enroll in via one of the major cloud service provides such as AWS, Microsft Azure, or Google Cloud.

Second significant enhancement is the introduction of Large Language Models or LLMs which were themselves an extension of machine learning and data modeling. With automated “learning,” as we call it, we could write code that provided machines instructions on how to handle data. Prior to this, we had to write a custom parser, normalization schema, data transforms, and data enrichments to prepare data to be use by the analysis layer. With machine learning, we could allow the automation scripts to derive understanding from the data we presented it, by which the data could be modeled, or perhaps a better way of saying it would be to say the data was mapped with significant values identified and characterized.

Through these LLMs we could take disparate data sets, let the automation map out their values and characteristics, and again automate the process of finding relationships. This increased our scale and speed for processing large data sets, and expanded the different data sets we had access to.

The third and perhaps most sophisticated component is the natural language interface to these LLMs. Again, in the early days, we had to craft our data searches very carefully, in order to make sure the right scope of information was included and to make sure we organized the logic properly so that the comparative functions would get the formula right.

Today this has been replaced by data transforms so you can interact with the LLMs through a “chat” like functionality, which is where the name Chat GPT comes from. Sending commands and search requests to the LLMs is like chatting with a being that gives the impression of being autonomous and intelligent.

The fourth significant enhancement is what we refer to as the generative function of todays LLMs. Basically, we have coded the LLMs to be able to craft search results. This capability is actually what a lot of people think Google is doing when you run a search via that engine. It is similar, but also different. When you search via Google, you think you are tasking a giant analytics engine to review a bunch of information and provide you with the correct response. That’s not how Google works. It works by searching a summarization of information to find matches to the key words you have supplied, and it presents results based on pre-defined algorithms that rank them.

The same is fundamentally true about the LLMs and AI chat bots, except they don’t just return to you what exists, but they will also return to you what they can construct from what exists. If you think of the AI chat bots as engaged in a game of fill in the blank, that’s essentially what they are doing. Your search is defining the result it needs to compile from the various sources it has access to. Because it’s a computer it has to present you with a complete result or some other pre-defined response, so it fills in the blanks.

This game is also based on pre-defined algorithms that have been created to enable the chat bot to understand data sources, compare them, and know what and how to present back to you.

And so what we have today is truly a capability with many use cases. Now that we have solved the challenge of data ingestion, expanded our capacity to understand disparate data sets, enhanced our ability to find relationships across data, added a natural language interactive interface, and provided it with the ability to “generate” an answer to our questions…we have a thing we call AI.

But to return to one of my original points, it is not really artificial intelligence. We may be generous in calling it simulated intelligence, but really I prefer the pre-existing term, data analytics.

Bottom Line

I’d like to leave you with this understanding: AI didn’t come out of nowhere, and it is a technology without a solution. But it is also not what it claims to be. It is not a source of truth, nor is it wise nor fully understanding.

The only true and proper use case for AI is to analyze data for summarization purposes. It’s best if we leave it at that.