Understanding the Levels of AI
AI is a subfield of computer science focused on enabling machines to solve problems like humans.
Specialized task AI (e.g., Watson, ChatGPT, Deep Blue).
Human-level reasoning and adaptability.
Surpasses human intelligence.
Future developments may redefine or expand these categories.
We at Codesmith cultivate technologists who are at the intersection of society and tech, able to meet this moment and thrive in an ever-changing world. Ready to become a modern software engineer?
Recall that computer science (CS) is the field that studies how computers and computations can be used to solve problems efficiently. Artificial intelligence (AI) is a discipline of CS that focuses on creating computer models that “imitate intelligent human behavior … [enabling computers] to perform complex tasks in a way that is similar to how humans solve problems.” Yet, AI is a broad umbrella term. We can further break down AI into 3 levels or kinds of AI based on capabilities: Narrow AI, General AI, and Super AI. Within these levels of AI, there may be additional types of AI based on functionalities. In this article, we will define these levels and types of AI to deepen our understanding.
This is the AI level we are currently at today. Narrow AI, also called Weak AI, is trained to perform a single or defined task. It focuses on a specific subset of cognitive abilities.
Example: IBM categorizes Watson® as Narrow AI. IBM Watson® is a computer system originally developed to answer questions on the "Jeopardy!" quiz show. It utilizes natural language processing, information retrieval, knowledge representation, reasoning, and machine learning. In 2011, it won first place and received a prize of 1 million dollars.
Under Narrow AI, there are functional types of AI.
Reactive Machine AI are systems equipped without memory and designed to accomplish a highly specific task. It only has the current available data to work with and can't rely on prior results. This type of AI is rooted in statistical math, with the ability to analyze huge amounts of data to derive an output.
Example: IBM Deep Blue is a computer system designed for playing chess by analyzing the board pieces and predicting the probable outcomes of each move. It utilized 32 processors to perform parallel computations, evaluating 200 million chess positions per second. In 1997, it became the first machine to defeat a human chess grandmaster under standard tournament conditions. IBM Deep Blue won via brute force by calculating winning chess moves.
Team picture at the 1997 ACM Chess Challenge, Kasparov vs. Deep Blue. Source: https://www.ibm.com/history/deep-blue (accessed 5/14/2025)
Limited Memory AI are systems equipped with memory capacity. It can use past and present data for decision-making and obtaining an output. However, the previous events and results can only be used for a specific amount of time. Limited Memory AI can absorb new learning data, improving its performance over time.
Examples:
Google's Waymo self-driving car at the Computer History Museum. Source: Photo by Igor Shalyminov on Unsplash (accessed 5/14/2025)
General AI or Artificial General Intelligence (AGI), also called Strong AI, can use previous learnings and skills to perform new tasks in a different context without human guidance and training. Hence, AGI can execute any intellectual task that a human can. So it's on par with human cognitive abilities. As of now, this is just a theory.
Theory of Mind AI is a functional type of General AI that can understand the thoughts and emotions of other entities, such as humans. Stemming from psychology, Theory of Mind (ToM) is the concept of human capacity for empathy and recognizing that other people have their own thoughts and feelings, which influence their behavior.
For instance, if you notice your friend is hangry, you might offer your friend a snack to soften the situation. With this understanding, Theory of Mind AI can simulate human-like interactions and relationships. They would be able to uniquely respond to individuals based on intentions and feelings.
Potential applications of Theory of Mind AI could be in empathetic healthcare assistants, for patients with neurological or psychiatric disorders, and it could also improve human-machine collaboration, showing the potential "that humans and machines blending together are more powerful than either working alone."
Although some LLMs display early signs of ToM-like abilities, such as ChatGPT appearing to commiserate if you express that you are frustrated, these responses are currently achieved through pattern recognition and statistical inference rather than true empathy. And while advanced LLMs can reliably predict a user's mental state, they are lacking in the ability to correctly predict the user's behavior based on that mental state and to judge whether the user's behavior is reasonable.
A table of evaluation results of ToM capabilities in LLMs by Ai2. Source: https://allenai.org/blog/applying-theory-of-mind-can-ai-understand-and-predict-human-behavior-d32dd28d83d8 (accessed 5/14/2025)
Super AI or Artificial Superintelligence (ASI), will transcend the cognitive abilities of humans. They can think, learn, act, and feel on their own. At that future point, we have reached AI singularity, the scenario where AI surpasses human intelligence and can evolve itself autonomously. This would lead to exponential technological advances and unpredictable changes to human civilization. Currently, ASI is highly theoretical.
The term "singularity" was first used in physics, including in Albert Einstein's Theory of General Relativity to describe the point of infinite density and gravity inside a black hole's center where nothing can escape. It is now commonly used to indicate a tipping point in which our existing models of understanding or reality break down.
Self-Aware AI is a functional type of Super AI that can understand its own internal state and has its own beliefs, needs, emotions, and traits. It would be "sentient," having a sense of self, consciousness, and subjective experiences.
Today, researchers are still working to fully understand consciousness in humans and animals. How would we know that AI has gained self-awareness? A research team of computer scientists, neuroscientists, and philosophers developed a checklist of 14 key criteria to test for AI consciousness. For instance, the criterion of agency, the ability to make conscious decisions for action. Another criterion is embodiment, which is taking form in physical space or relative to other digital systems.
Self-Aware AI would be able to self-improve and interact with humans and environments in new ways, leading to rapid advancements in technology and other fields. It can also make more ethical and contextual decisions in areas such as law, creativity, and healthcare.
A humanoid robot. Source: Photo by Gabriele Malaspina on Unsplash (accessed 5/14/2025)
We have now learned that AI is more nuanced than a catch-all concept. There are distinct levels or kinds of AI: Narrow AI, General AI, and Super AI. Within each level, there are specific functional types of AI. More kinds and types of AI would probably be created and defined in the future.
Explore CS Prep further in our beginner-friendly program.
Get more free resources and access to coding events every 2 weeks.
Connect with one of our graduates/recruiters.
Our graduates/recruiters work at:
Diana Cheung (ex-LinkedIn software engineer, USC MBA, and Codesmith alum) is a technical writer on technology and business. She is an avid learner and has a soft spot for tea and meows.
Connect with one of our recruiters to learn about their journeys.
Our graduates/recruiters work at:
Copyright @ 2025 Codesmith | All Rights Reserved
Terms and ConditionsPrivacy PolicyRegulatory Information
