AI Modeling Drives Automation Across Smart Systems

The Machine That Learns to Build Itself

In 2022, a computer scientist named Iqbal H. Sarker published a paper that read less like a technical survey and more like a blueprint for a quiet revolution. The paper, which has since been cited over a thousand times, argued that artificial intelligence had stopped being a tool we use and had become something closer to an architect. Not just a smarter calculator, but a system that could design, optimize, and even automate the creation of other smart systems. The key, Sarker wrote, was something he called "AI-based modeling" — a process where machines don't just follow rules, but learn to generate their own models of how the world works (Sarker, 2022).

This is not science fiction. It is already happening in hospitals, on farms, in financial markets, and inside the cybersecurity systems that protect your data. But the implications are stranger and more profound than most people realize. Because when a machine builds its own model of reality, it doesn't just automate a task. It automates the thinking about that task.

What Does It Mean for a Machine to Build a Model?

Sarker's paper is a sprawling survey of every major AI technique in use today — neural networks, fuzzy logic, evolutionary algorithms, reinforcement learning, and more. But the paper's real contribution is not the list of techniques. It is the argument that these techniques are not separate tools. They are a unified way of approaching a problem.

Think of it like this. A traditional computer program is a set of instructions: if this happens, do that. A human writes the rules. But an AI model does something different. It takes in data — thousands of examples of something — and infers the rules itself. It builds a simplified version of reality that it can then use to make predictions or decisions.

Sarker calls this "AI-based modeling," and he argues that it is the foundation for all truly intelligent systems (Sarker, 2022). Without a model, a machine cannot adapt. It cannot handle a situation it has never seen before. With a model, it can.

The paper breaks down the types of AI that can be used to build these models. There is analytical AI, which processes numbers and logic. Functional AI, which acts on the world. Interactive AI, which converses with humans. Textual and visual AI, which reads and sees. Each type produces a different kind of model. But the goal is always the same: to create a system that can automate not just a task, but the understanding of the task.

The Five Types of Intelligence That Drive Automation

Sarker's taxonomy is practical. He does not just name the types of AI. He maps them to real-world problems.

Analytical AI: The Number Cruncher That Learns Patterns

This is the oldest and most mature form of AI. Analytical AI includes machine learning algorithms like regression, decision trees, and support vector machines. These models take structured data — rows and columns of numbers — and find patterns that a human might miss.

In finance, analytical AI models predict stock prices and detect fraud. In healthcare, they analyze patient records to forecast disease risk. Sarker notes that these models are particularly good at classification and prediction tasks (Sarker, 2022). They are the workhorses of automation.

But there is a catch. Analytical models are only as good as the data they are trained on. If the data is biased, the model will be biased. If the data is incomplete, the model will be wrong. Sarker emphasizes that developing an effective AI model is "a challenging task due to the dynamic nature and variation in real-world problems and data" (Sarker, 2022). The model is a simplification. It is never perfect.

Functional AI: The Action Taker

Analytical AI thinks. Functional AI acts. This category includes robotics, autonomous vehicles, and industrial control systems. A functional AI model does not just predict what will happen. It decides what to do about it.

Sarker points out that functional AI is the key to building "smart systems" — things like smart grids, smart factories, and smart cities (Sarker, 2022). These systems do not just collect data. They use that data to change their behavior in real time. A smart thermostat does not just measure temperature. It models your preferences and adjusts the heating accordingly.

This is where automation becomes truly powerful. The system does not need a human to tell it what to do. It has built a model of the problem and can act on its own.

Interactive AI: The Conversationalist

This is the AI that most people have encountered. Chatbots, virtual assistants, and customer service bots all use interactive AI. These models are designed to understand human language and respond in a way that feels natural.

Sarker notes that interactive AI relies heavily on natural language processing (NLP) and sentiment analysis (Sarker, 2022). The model must not only parse words but also infer intent and emotion. This is hard. A sentence like "That's great" can mean genuine enthusiasm or sarcastic disappointment. The model has to figure out which.

Interactive AI is driving automation in customer service, healthcare triage, and even mental health support. But Sarker warns that these models are still brittle. They can be fooled by unusual phrasing or cultural references. They are getting better, but they are not yet human.

Textual and Visual AI: The Reader and the Seer

Textual AI goes beyond simple chatbots. It can summarize documents, translate languages, and even generate new text. Visual AI can recognize objects in images, detect faces, and analyze medical scans.

Sarker groups these together because they both deal with unstructured data — the messy, human-generated stuff that does not fit neatly into spreadsheets (Sarker, 2022). A photograph is not a row of numbers. A paragraph of text is not a column of data. These AI models have to extract meaning from noise.

The implications for automation are huge. A visual AI model can inspect products on an assembly line faster than any human. A textual AI model can read thousands of legal documents in seconds. These systems do not just automate a task. They automate a skill that was once uniquely human: the ability to see and read.

The Paradox of the Black Box

Here is the uncomfortable truth that Sarker's paper does not shy away from. The more powerful an AI model becomes, the harder it is to understand how it works.

Simple models like decision trees are transparent. You can trace exactly why the model made a certain decision. But the most accurate models — deep neural networks, for example — are black boxes. They have millions of parameters that interact in ways that even their creators cannot fully explain.

Sarker calls this one of the major research issues in the field (Sarker, 2022). How do you trust a system that you cannot understand? In healthcare, a black box model might correctly diagnose a disease 99% of the time. But if it makes a mistake, you cannot ask it why. You cannot audit its reasoning.

This is not just an academic problem. It is a regulatory nightmare. If an autonomous car kills someone, who is responsible? The programmer? The data set? The model itself? The law is not ready for this question. Sarker's paper highlights explainability as a critical open challenge (Sarker, 2022).

What the Research Does Not Prove

Sarker's paper is a survey, not an experiment. It does not test a hypothesis or produce a new algorithm. It synthesizes existing knowledge and points to gaps. That is a legitimate scientific contribution, but it means the paper has limitations.

The paper does not prove that any specific AI technique is superior to others. It does not provide a step-by-step guide for building a smart system. It does not offer a unified theory of intelligence. What it does is provide a map of the territory. It tells you what tools exist, what they are good for, and where the hard problems lie.

One of the most interesting open questions is whether these five types of AI can be combined into a single, unified model. Sarker suggests that the future of intelligent systems will require integrating analytical, functional, interactive, textual, and visual AI into a single architecture (Sarker, 2022). But he admits that this is easier said than done. The techniques are built on different mathematical foundations. They do not always play well together.

Another open question is how to handle the dynamic nature of the real world. A model trained on last year's data might be useless for next year's problems. Sarker emphasizes that AI models must be continuously updated and retrained (Sarker, 2022). But that creates its own challenges. How do you retrain a model without losing what it learned before? How do you prevent it from forgetting?

These are not weaknesses in the paper. They are honest acknowledgments of what we still do not know.

The Automation That Builds Itself

The most provocative idea in Sarker's paper is not stated explicitly. It emerges from the structure of the argument itself.

If AI-based modeling is the key to building intelligent systems, and if those systems can themselves build models, then we are looking at a recursive process. A smart factory does not just automate assembly. It monitors its own performance, identifies bottlenecks, and redesigns its workflow. A smart city does not just collect traffic data. It models traffic patterns and adjusts traffic lights in real time.

Sarker calls this the path toward "automation, intelligent, and smart systems" (Sarker, 2022). But the phrase that sticks is "automation." Because what is being automated here is not a single task. It is the entire process of optimization. The system becomes its own engineer.

This is already happening in cybersecurity, where AI models detect and respond to threats faster than any human could. Sarker notes that cybersecurity is one of the key application areas for AI-based modeling (Sarker, 2022). A security system that uses AI can learn what normal network traffic looks like and flag anomalies. It can even take action — blocking an IP address or isolating a compromised device — without waiting for a human decision.

In agriculture, AI models analyze satellite images to predict crop yields and optimize irrigation. In healthcare, they read medical scans and suggest diagnoses. In finance, they execute trades in milliseconds. In each case, the model is not just a tool. It is the decision-maker.

The Human Cost of Smarter Machines

Sarker's paper is technical. It does not dwell on social implications. But the implications are unavoidable.

If AI models can automate the thinking behind a task, what happens to the people who used to do that thinking? The answer is not simple. Some jobs will disappear. Others will change. New ones will emerge. But the transition will be painful.

The paper does not address this directly. But it does provide a framework for thinking about it. Sarker distinguishes between different types of AI based on what they do. Analytical AI replaces number crunching. Functional AI replaces physical action. Interactive AI replaces conversation. Textual and visual AI replace reading and seeing.

Taken together, these five types of AI can replace a wide range of human cognitive and physical skills. The question is not whether they will. It is how fast, and what we do about it.

Sarker's paper is a guide for the people building these systems. But it is also a warning. The systems are getting smarter. The models are getting better. And the automation is happening whether we are ready or not.

What This Actually Means

• ▸If you are building a smart system, start with the data. Sarker's paper makes clear that the quality of the model depends entirely on the quality of the data. Garbage in, garbage out. Spend your time cleaning and labeling your data before you worry about which algorithm to use.

• ▸Do not expect one AI technique to solve everything. Sarker's taxonomy shows that different problems require different types of AI. If you are building a chatbot, do not try to use a visual AI model. Match the technique to the task.

• ▸Plan for the black box problem. If your system makes decisions that affect people's lives, you need to be able to explain those decisions. Sarker identifies explainability as a major research issue. Until the field solves it, regulators and the public will be skeptical.

• ▸Build for change. The real world is dynamic. Your model will need to be retrained. Sarker emphasizes this repeatedly. Do not build a system that assumes the world will stay the same.

• ▸The automation is not coming. It is here. Sarker's paper is a snapshot of a field that is already transforming industries. The smart systems he describes are not theoretical. They are in production. The question is not whether to adopt them. It is how to adopt them responsibly.