The AI industry is expanding at an unprecedented pace. Analysts estimate that the global AI market could reach $2.4 trillion by 2034, growing at a CAGR of more than 26%. [1]
Meanwhile, worldwide AI spending is expected to surpass $2.5 trillion in 2026, up 47% from the previous year. The growth is driven by heavy investments in AI infrastructure, cloud computing, specialized chips, foundation models, and enterprise software. [2]
As organizations continue investing heavily in AI, understanding the different types of AI has become essential for businesses, developers, students, and technology enthusiasts alike.
However, there is no single perfect way to categorize AI. That’s why I have mentioned different types of AI by capability, functionality, and model type.
This guide provides valuable insight into where the technology stands today and where it is likely to evolve over the coming decades.
Did you know?
Generative AI reached 53% adoption in only three years, making it one of the fastest-adopted technologies in history, faster than both the internet and personal computers. [3]
Table of Contents
1. By Capability (Most Common Classification)
AI can be classified according to its level of intelligence and cognitive ability. Instead of focusing on how an AI system is built or how it learns, this approach evaluates what the system is theoretically capable of accomplishing.
This is the most widely recognized classification of AI and is commonly used in research, industry, and public discussions about the future of artificial intelligence.
1.1 Narrow AI (ANI)
Artificial Narrow Intelligence, also known as Weak AI or Narrow AI, is the most common and only type of artificial intelligence that exists today. Unlike the human brain, which can learn and perform many unrelated tasks, ANI is designed to excel at one specific function or a limited set of closely related tasks.
An ANI system can recognize faces, translate languages, generate text, recommend movies, or drive a car, but it cannot seamlessly transfer its knowledge from one domain to another without being redesigned or retrained.
Example
Modern AI assistants like Gemini, Claude, and ChatGPT are examples of Narrow AI. While they can answer tricky questions and generate computer code, they cannot make autonomous life decisions or acquire human-level understanding across every domain.
In fact, Google searches, Face ID unlock, and Netflix recommendations are powered by Narrow AI. A smartphone often runs dozens of AI models locally to perform tasks like camera enhancement, voice recognition, and predictive typing.
A single autonomous vehicle can rely on multiple specialized AI models operating simultaneously for perception, planning, localization, and safety monitoring.
1.2 General AI (AGI)
Artificial General Intelligence, also known as Strong AI, is a theoretical form of AI that can understand, learn, reason, and perform virtually any intellectual task at a human level.
Unlike Narrow AI, which is designed for a single purpose, AGI would be capable of applying knowledge across multiple domains without requiring task-specific training.
This type of AI can transfer knowledge from one field to another, a capability known as generalization. It could continuously learn from new experiences, adapt to unfamiliar situations, make independent decisions, and improve its own understanding over time without being explicitly programmed for every new task.
Example
Since AGI has not yet been achieved, there are currently no real-world examples.
However, in the future, AGI could compose music, conduct scientific research, diagnose diseases, and solve complex business problems — all using the same underlying intelligence.
This idea is also driving the development of general-purpose robots. Tesla is building Optimus, Figure AI is developing Figure 03, and Boston Dynamics is working on the electric Atlas robot.
1.3 Superintelligent AI (ASI)
Artificial Superintelligence is a hypothetical form of AI that would surpass human intelligence in virtually every cognitive domain, including reasoning, problem-solving, strategic planning, creativity, scientific discovery, and emotional understanding.
This type of AI could continuously improve itself. Instead of relying solely on human researchers to develop new algorithms, an ASI system could potentially redesign its own architecture, learn more efficiently, and rapidly increase its intelligence through self-improvement.
This concept, sometimes referred to as an “intelligence explosion,” suggests that an AI system could evolve far beyond human capabilities in a relatively short period.
However, it is important to note that ASI does not exist today and remains a theoretical concept discussed by AI researchers, philosophers, and futurists.
Hypothetical Examples
ASI could help solve some of humanity’s biggest challenges, from curing diseases and developing clean energy to planning and managing complex interplanetary missions. It could also run millions of virtual experiments simultaneously, generate new hypotheses, and advance scientific research on its own.
Meanwhile, it also raises significant concerns about AI safety, governance, and ensuring that superintelligent systems remain aligned with human values and objectives.
2. By Functionality (Traditional Classification)
Now, let’s categorize AI by how it perceives information, processes knowledge, and makes decisions.
While this classification is primarily theoretical, it provides a useful framework for understanding the evolution of AI from simple reactive systems to the advanced intelligent machines envisioned for the future.
2.1 Reactive Machines
Reactive Machines are the most basic type of AI in the functionality-based classification.
These AI systems are designed to respond only to current inputs without storing memories or learning from past experiences. They make decisions solely based on the information available at the present moment.
Because they have no memory or learning capability, Reactive Machines produce the same output whenever they encounter the same input. They cannot improve through experience.
Reactive Machines may seem simple by today’s standards, but they played a key role in AI’s early days. They showed that computers could beat humans at certain tasks by evaluating the current situation and selecting the best possible action, without needing memory or human-like thinking.
Examples
- IBM Deep Blue: This famous chess-playing computer defeated world chess champion Garry Kasparov in 1997 by evaluating 200 million chess positions per second without remembering previous games.
- Automatic doors and motion sensors: These systems respond immediately when movement is detected but do not remember previous events.
- Basic industrial robots: Many factory robots repeatedly perform the same welding, painting, or assembly tasks based solely on current sensor inputs.
2.2 Limited Memory AI
Limited Memory AI is the most widely used type of artificial intelligence today. It can store, analyze, and learn from historical data for a limited period of time.
By combining past information with real-time inputs, these systems can make more accurate predictions, improve their performance, and adapt to changing situations.
However, their “memory” is task-specific rather than human-like. They do not possess lifelong memory, consciousness, or a complete understanding of the world.
Examples
Most modern AI applications (including chatbots, fraud detection systems, image recognition models, recommendation engines, and self-driving cars) fall into this category.
These systems are trained on massive datasets and use previous observations to recognize patterns, identify trends, and make smart decisions.
2.3 Theory of Mind AI
Theory of Mind AI is a theoretical type of artificial intelligence that would be capable of understanding human thoughts, intentions, emotions, beliefs, and social interactions.
It would be able to interpret why people behave in certain ways and adjust its responses based on their emotional and psychological state.
The concept is inspired by the human cognitive ability known as “theory of mind,” which allows people to understand that others have their own beliefs, emotions, intentions, and perspectives.
A true Theory of Mind AI would go beyond processing words or images: it would understand context, recognize emotional cues, predict human intentions, and interact with people in a more natural, empathetic, and socially intelligent manner.
Although researchers have made great progress in emotion recognition and conversational AI, no true Theory of Mind AI exists today.
Potential Applications
- Healthcare and mental health support: AI could recognize emotional distress, monitor patient well-being, and help therapists provide more personalized care.
- Education: AI tutors that recognize when students are confused, frustrated, or disengaged and adjust teaching methods in real time.
2.4 Self-Aware AI
Self-Aware AI is the most advanced and entirely hypothetical stage of artificial intelligence in the functionality-based classification. If it ever becomes a reality, it would be conscious, understand its own existence, and experience emotions, much like a human being.
A truly self-aware AI would be able to form independent goals, reflect on its own decisions, understand its strengths and limitations, and potentially experience subjective awareness.
However, these characteristics remain the subject of ongoing debate among AI researchers, cognitive scientists, and philosophers because human consciousness itself is not yet fully understood.
3. By AI Model Type
AI model types describe the underlying computational methods used to build intelligent systems. These models differ in how they represent knowledge, learn from data, make decisions, and solve problems.
Over the past seven decades, AI models have evolved from rule-based symbolic systems to statistical machine learning, deep neural networks, and modern hybrid architectures that combine multiple approaches.
3.1 Symbolic AI
Symbolic AI, often referred to as Good Old-Fashioned AI, is one of the earliest approaches to artificial intelligence.
Symbolic AI represents knowledge using symbols, facts, relationships, and logical rules. Instead of learning from data, it solves problems by applying predefined rules and logical reasoning.
Experts manually create the knowledge base by defining facts, rules, and relationships. For example, “IF a patient has a fever AND cough, THEN consider influenza.” A well-known example of Symbolic AI is MYCIN, an early medical expert system that used approximately 600 expert-defined rules to diagnose bacterial infections and recommend treatments. [4]
Symbolic AI can be further categorized into five sections:
3.1.1 Rule-Based Systems
Rule-Based Systems are one of the oldest and most fundamental forms of artificial intelligence, forming the foundation of Symbolic AI.
These systems make decisions by following predefined IF-THEN rules created by human experts. Every rule specifies a condition and a corresponding action.
For example, a simple rule might state: “IF a customer spends more than $2,000, THEN flag the transaction for additional verification.”
A Rule-Based System typically consists of three core components: a rule base containing all predefined rules, a working memory that stores current facts or inputs, and an inference engine that evaluates the rules and determines which actions to execute.
3.1.2 Expert Systems
Expert Systems use a knowledge base of facts and expert-defined rules, along with an inference engine that applies logical reasoning to solve problems, diagnose conditions, and recommend solutions.
A typical Expert System has three main components: the knowledge base, which stores facts and domain expertise; the inference engine, which evaluates rules and reaches conclusions; and a user interface, which allows users to interact with the system.
Real-World Examples
- DENDRAL: An AI system that helped chemists identify the molecular structure of organic compounds using expert-defined rules.
- XCON (R1): Developed by Digital Equipment Corporation, this AI system automatically configured computer hardware and substantially reduced configuration errors.
3.1.3 Logic-Based AI
This type of AI solves problems through formal logic, mathematical reasoning, and inference rather than statistical learning.
Logic-Based AI is built on formal logic systems, such as propositional logic and first-order predicate logic. It represents knowledge as logical facts (for example, “All mammals are warm-blooded” and “A whale is a mammal“). An inference engine then applies logical rules to draw conclusions (such as “A whale is warm-blooded“).
Because every conclusion follows a clear chain of reasoning, Logic-Based AI is widely used in applications where accuracy, consistency, and explainability are important.
Examples
- Prolog programming language: A popular logic programming language used in AI research, natural language processing, and expert systems.
- Automated theorem provers: These AI systems prove mathematical theorems or verify logical correctness by applying formal reasoning. [5]
3.1.4 Knowledge Representation
Knowledge Representation (KR) focuses on how AI systems store, organize, structure, and reason about knowledge. It converts information into meaningful concepts, facts, relationships, and rules that machines can understand and use for logical reasoning.
For example, if an AI system knows that “Paris is the capital of France” and “France is located in Europe,” it can logically infer that “Paris is located in Europe.”
KR uses several techniques to organize information, including semantic networks, ontologies, knowledge graphs, frames, logical predicates, and production rules.
Real-World Examples
- Google Knowledge Graph: Google organizes billions of entities (such as people, places, companies, books, and events) and the relationships between them to improve search results and answer user queries.
- Wikidata: One of the world’s largest open knowledge bases, containing structured information about millions of real-world entities and their relationships.
3.1.5 Planning & Search
Planning & Search focuses on enabling AI systems to find the best sequence of actions to achieve a specific goal.
More specifically, Planning & Search algorithms use logical reasoning, heuristics, and mathematical search techniques to explore possible solutions and determine the most efficient path toward an objective.
Depending on the complexity of the problem, it may use algorithms such as Breadth-First Search, Depth-First Search, Uniform Cost Search, A (A-Star), Greedy Best-First Search, or Minimax to navigate through thousands, or even millions, of possible states.
Real-World Examples:
- GPS navigation systems: Google Maps and Apple Maps apps use search algorithms to calculate the fastest or shortest route while considering traffic, road closures, and travel time.
- Chess AI: Chess engines analyze millions of possible moves using search algorithms such as Minimax combined with optimization techniques like Alpha-Beta Pruning to identify the strongest move.
3.2 Classical Machine Learning
Classical Machine Learning is a data-driven branch of AI in which computers learn patterns from historical data instead of relying on manually programmed rules.
It enables algorithms to automatically identify relationships, make predictions, classify information, and improve their performance through experience.
Classical Machine Learning became the dominant AI approach during the 1990s and early 2000s, replacing many rule-based systems for prediction, classification, and other data-driven tasks.
3.2.1 Supervised Learning
Supervised Learning is the most widely used type of machine learning, where an AI model learns from labeled data. During training, it receives input data along with the correct output (or label), allowing it to learn the relationship between them. Once trained, the model can accurately classify new data or predict outcomes for previously unseen inputs.
Because the correct answers are known during training, the learning process is described as “supervised.”
Did you know?
ImageNet, one of the world’s most influential AI datasets, contains more than 14 million labeled images and has played a major role in advancing computer vision research. [6]
Examples
- Email spam detection: Gmail and Outlook train AI models using millions of labeled emails to distinguish spam from legitimate messages.
- Medical diagnosis: Hospitals use supervised learning to identify diseases like diabetes, cancer, and heart conditions from labeled medical images and patient records.
3.2.2 Ensemble Learning
Ensemble Learning combines the predictions of multiple models to produce a more accurate, stable, and reliable result than any individual model could achieve on its own.
This approach is based on the principle that a group of diverse models often makes better decisions than a single model.
Many of the world’s highest-performing machine learning models (including Random Forest, XGBoost, LightGBM, AdaBoost, CatBoost, and Gradient Boosting Machines) are ensemble methods.
Example
- Credit risk assessment: Financial institutions use ensemble algorithms such as Random Forest and Gradient Boosting to evaluate loan applications and predict default risk.
- Predictive maintenance: Manufacturing companies combine sensor data and multiple predictive models to anticipate equipment failures before they occur.
3.2.3 Unsupervised Learning
In Unsupervised Learning, AI algorithms learn from unlabeled data. They receive only raw data and must discover hidden patterns, structures, relationships, or groupings on their own.
The primary objective of Unsupervised Learning is to uncover meaningful insights that humans may not immediately recognize. It is particularly valuable for exploring large datasets where labeling every data point would be impractical.
Examples
- Customer segmentation: Retailers and e-commerce giants group customers with similar purchasing habits to create personalized marketing campaigns.
- Cybersecurity: Security systems detect abnormal network activity that could indicate cyberattacks or unauthorized access.
3.2.4 Probabilistic Models
As the name suggests, Probabilistic Models use probability and statistics to make predictions under uncertainty. Rather than giving a single fixed answer, they estimate the likelihood of different outcomes and choose the most likely result.
The fundamental idea behind Probabilistic Models is that many real-world events cannot be predicted with absolute certainty. For instance, a bank cannot know with 100% confidence whether a customer will default on a loan, but it can estimate the probability based on historical financial data.
Example
- Weather forecasting: Meteorological agencies estimate the probability of rain, storms, snowfall, and other weather events using probabilistic forecasting models.
- Email spam filtering: Naïve Bayes classifiers estimate the probability that an incoming email is spam based on the words and patterns it contains. [7]
3.2.5 Classical Reinforcement Learning
Classical Reinforcement Learning became a major area of AI research during the 1980s and 1990s, long before the rise of Deep Reinforcement Learning.
In this approach, an AI agent learns through trial and error by interacting with its environment. It receives rewards for good actions and penalties for poor ones. Over time, the agent discovers which actions maximize long-term rewards, allowing it to make increasingly better decisions.
Examples
- Warehouse robots: Autonomous robots learn efficient routes for transporting products while minimizing travel time and avoiding collisions.
- Traffic signal optimization: Smart traffic management systems adjust traffic light timings based on vehicle flow to reduce congestion and improve traffic efficiency.
3.3 Neural Networks (Deep Learning)
Neural Networks are inspired by the structure and functioning of the human brain. Instead of relying on statistical formulas or manually programmed rules, neural networks learn by processing vast amounts of data through interconnected layers of artificial neurons.
These neurons automatically identify patterns, extract features, and continuously improve their predictions by adjusting internal parameters during training.
Today, Deep Learning represents the most powerful and rapidly advancing field of AI, driving breakthroughs in computer vision, natural language processing, speech recognition, scientific research, generative AI, and robotics.
In fact, nearly every major AI breakthrough (including ChatGPT, Claude, Gemini, Midjourney, DALL·E, AlphaFold, and autonomous vehicles) is powered by deep neural networks.
3.3.1 Feedforward Networks
Feedforward Neural Networks (FNNs) are the simplest and most fundamental type of artificial neural network. They are called “feedforward” because information flows in only one direction: from the input layer, through one or more hidden layers, to the output layer.
Examples
- House price prediction: Real estate firms estimate property values using features like location, size, age, and market trends.
- Customer churn prediction: Telecom companies and subscription services predict which customers are most likely to cancel their services.
3.3.2 Convolutional Neural Networks (CNN)
CNNs are a specialized type of deep neural network designed primarily for processing images and other visual data.
They automatically learn visual features such as edges, shapes, textures, and objects by applying mathematical filters called convolutions. This allows them to recognize patterns in images with remarkable accuracy while using far fewer parameters than traditional neural networks.
CNNs gained widespread attention after AlexNet won the ImageNet Challenge in 2012. Modern CNN models can achieve over 99% accuracy in certain specialized image classification tasks. [8]
Examples
- Face recognition: Face ID, facial authentication, and surveillance systems.
- Medical imaging: Detecting cancer, fractures, pneumonia, and other diseases from X-rays, CT scans, and MRI images.
3.3.3 Recurrent Neural Networks (RNN)
RNNs are designed to process sequential data, where the order of information matters.
They contain internal memory that allows them to retain information from previous inputs. This enables the network to understand context and relationships across sequences such as text, speech, time-series data, and videos.
Although Transformers have largely replaced RNNs in modern natural language processing, RNNs remain an important milestone in deep learning history and are still used in certain time-series forecasting, signal processing, and embedded AI applications where computational efficiency is critical. [9]
Examples
- Speech recognition: Early versions of Siri, Google Assistant, and Alexa used RNNs to convert spoken language into text.
- Machine translation: Early Google Translate systems relied on RNNs before transitioning to Transformer models.
3.3.4 Transformers
Transformers are a deep learning architecture designed to process sequential data such as text, code, images, audio, and video.
Introduced by Google researchers in the landmark 2017 paper “Attention Is All You Need,” Transformers revolutionized AI by replacing the sequential processing of Recurrent Neural Networks (RNNs) with a powerful mechanism called self-attention. [10]
Instead of processing information one step at a time, Transformers process the entire input in parallel, enabling them to learn complex patterns from massive datasets. This architecture powers Large Language Models (LLMs) and has become the foundation of modern generative AI.
Real-World Examples
Transformers are the most influential neural network architecture in AI. They power leading AI models like ChatGPT, Gemini, Claude, Llama, DeepSeek, Microsoft Copilot, DALL·E, Stable Diffusion, and many other state-of-the-art systems.
Their ability to scale efficiently to billions (or even trillions) of parameters has made them the driving force behind the current AI revolution.
3.3.5 Graph Neural Networks (GNN)
GNNs are designed to analyze graph-structured data, where information is represented as nodes (entities) connected by edges (relationships).
They work by allowing neighboring nodes to exchange information through a process called message passing. During training, each node continuously updates its representation by combining its own data with information from connected nodes. This allows the model to capture both local and global patterns within a network.
Real-World Examples
- Social media: LinkedIn, Facebook, and X (formerly Twitter) analyze relationships between users to recommend friends, groups, and content.
- Drug discovery: Pharmaceutical companies use GNNs to predict molecular properties and accelerate the development of new medicines.
3.3.6 Autoencoders
Autoencoders are designed to learn efficient representations of data without requiring labeled examples.
They work by compressing the input data into a smaller, information-rich representation called a latent space and then reconstructing the original input from this compressed version.
The aim is to recreate the input as accurately as possible while capturing its most important features and filtering out unnecessary information.
Examples
- Image denoising: AI removes noise, blur, and compression artifacts from photographs, medical scans, and satellite images.
- Variational Autoencoders (VAEs): VAEs generate realistic images, faces, handwriting, and synthetic datasets for research and creative applications.
3.3.7 Generative Adversarial Networks (GANs)
Introduced by Ian Goodfellow and his team in 2014, GANs create realistic synthetic data using two competing neural networks: the Generator and the Discriminator.
The Generator produces fake data, while the Discriminator tries to tell the difference between real and generated data. As they compete, both networks improve. And through this competitive process, both networks continuously improve until the generated content becomes highly realistic.
Although Diffusion Models have recently become the dominant technology for AI image generation, GANs remain an important milestone in deep learning.
Examples
- Image enhancement: AI enhances photo quality by increasing image resolution, restoring old photographs, and removing noise.
- Video game development: Game studios use GANs to generate realistic textures, characters, and environmental assets.
3.3.8 Diffusion Models
Diffusion Models are a state-of-the-art deep learning architecture built to generate highly realistic images, videos, audio, and other forms of digital content.
These models learn by gradually adding noise to training data and then learning how to reverse that process. Starting from random noise, the model progressively removes the noise step by step until it produces a realistic output that closely resembles the training data.
It’s worth noting that training large Diffusion Models is extremely demanding. They require billions of images, powerful GPU clusters, and vast computational resources, making them among the most resource-intensive AI models.
Examples
Since 2022, Diffusion Models have become the leading architecture for generative AI, powering popular applications such as DALL·E, Stable Diffusion, Adobe Firefly, Midjourney, and many other image-generation platforms.
Their applications are also expanding to video generation (such as OpenAI’s Sora), 3D content creation, and scientific simulations.
3.4 Hybrid AI
Instead of relying solely on neural networks, symbolic reasoning, or knowledge graphs, Hybrid AI integrates these technologies so they can complement each other’s strengths while overcoming individual limitations.
3.4.1 Neuro-Symbolic AI
Neuro-Symbolic AI is a hybrid artificial intelligence approach that combines the strengths of Neural Networks (deep learning) with Symbolic AI (rule-based reasoning and logic).
While neural networks excel at learning patterns from large datasets, symbolic AI is better at logical reasoning, knowledge representation, and explainable decision-making. By integrating these two approaches, Neuro-Symbolic AI aims to create AI systems that can both learn from data and reason using human-like logic.
Although still an active area of research, Neuro-Symbolic AI is gaining momentum in industries that require both learning and reasoning, such as finance, robotics, and scientific research. Many experts believe this hybrid approach could help pave the way for future Artificial General Intelligence (AGI). [11]
3.4.2 Knowledge-Augmented AI
A Knowledge-Augmented AI system enhances machine learning and deep learning models by combining them with external knowledge sources like knowledge graphs, databases, ontologies, documents, and domain-specific repositories.
Instead of relying solely on patterns learned during training, these AI systems can retrieve and use up-to-date factual information to produce more accurate, reliable, and context-aware responses.
With the rapid growth of Large Language Models (LLMs), Knowledge-Augmented AI has become one of the most important areas of modern AI research.
By using technologies like Retrieval-Augmented Generation (RAG), knowledge graphs, and enterprise search systems, AI assistants can access current and company-specific information, making them far more useful in real-world scenarios.
Examples
Search engines: Google Search engine and Microsoft’s Bing use knowledge graphs alongside AI models to provide richer and more accurate search results.
Enterprise AI assistants: Companies use AI that retrieves information from internal documents, policies, and knowledge bases to answer employee questions.
3.4.3 Graph-Augmented Neural AI
This AI combines deep neural networks with graph-based knowledge structures, such as knowledge graphs, semantic networks, and relationship graphs.
In this approach, neural networks perform tasks such as language understanding, image recognition, or prediction, while graph structures provide additional information about how people, objects, places, or concepts are related.
This allows the AI to make more informed decisions, improve factual accuracy, and better understand context.
For example, Google Search combines neural AI with knowledge graphs to better understand search queries and the relationships between entities. Similarly, Amazon, Netflix, and Spotify enhance their recommendation systems by analyzing connections between users, products, movies, and songs.
3.4.4 Agentic AI
Agentic AI can autonomously plan, reason, make decisions, and take actions to achieve specific goals with minimal human intervention.
Unlike traditional AI models that respond to individual prompts or perform a single task, Agentic AI can break complex goals into smaller steps, use external tools, gather information, evaluate results, and adapt its strategy as conditions change. This enables it to operate more autonomously and handle complex, multi-step workflows.
An Agentic AI system typically combines Large Language Models (LLMs) with technologies like memory, planning algorithms, reasoning engines, APIs, retrieval systems, and feedback mechanisms.
As more businesses automate knowledge work, Agentic AI is becoming the technology behind next-generation digital assistants, autonomous software agents, and enterprise automation platforms.
Examples:
- Customer service agents: AI handles customer inquiries, accesses company databases, processes refunds, and escalates complex cases when necessary.
- Autonomous research assistants: AI searches the web, gathers information, analyzes sources, and produces comprehensive reports with minimal human input.
- Software development: AI coding agents write, test, debug, and improve software while interacting with development tools.
Read More
- Artificial Intelligence vs Machine Learning vs Deep Learning | The Difference
- 14 Best Artificial Intelligence Programming Languages
- 13 Quantum Programming Languages To Learn
- AI market size and trend analysis, Fortune Business Insights
- Worldwide AI spending grows 47% in 2026, Gartner
- Economic footprint of AI across the private sector, Stanford University
- MYCIN: a consultation program for infectious disease diagnosis, ScienceDirect
- Automated Theorem Proving, Dr. Uwe Waldmann’s courses
- Image dataset organized according to the WordNet hierarchy, ImageNet
- Spam filtering with Naive Bayes, Kaggle
- A CNN called AlexNet won the ImageNet 2012 challenge, Pinecone
- A shift in the way machines understand language, Medium
- Attention Is All You Need, Ashish Vaswani
- A review of neuro-symbolic AI integrating reasoning and learning, ScienceDirect
