Different types of AI systems: A primer

Author’s note: Artificial intelligence (AI) brings both transformational potential and significant risk to businesses and society. This article, discussing the different types of AI, is the first in a series of blogs aimed at providing senior leaders and managers in mid-market organizations with a practical working knowledge of artificial intelligence (AI). If there are specific topics you wish for us to address in the future, please comment below.

This article was co-authored by Benson Chan of Strategy of Things and Peter Williams of Peter Williams Consulting.

An executive overview of AI types

According to IBM, artificial intelligence (AI) is “technology that enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity and autonomy”.[1] However, artificial intelligence is not one technology but many different types of AI technologies. Each type of AI possess unique capabilities well suited for specific types of tasks and goals.

AI enables businesses to enhance decision-making, automate operations and create new value for their customers, suppliers and partners. Unfortunately, there is a lot of hype and inflated expectations surrounding AI, leading to confusion and disappointing results. In reality, each type of AI is good at certain things, and not so good at other things. Applying the right type of AI to the things it is good at leads to good outcomes while minimizing risks. This article helps senior leaders and managers navigate the AI hype by providing an understanding of the different AI types.

Executive Imperative: Understanding the nuances of the different AI types

For today’s executives, understanding the different types of artificial intelligence is not just a technical necessity but a strategic imperative to drive growth, manage risks, and unlock competitive advantage.

Failing to understand these different AI types and their limitations can have serious consequences. Managers and executives may misjudge the potential of an AI system, leading to costly overpromises or underutilized tools. Similarly, overlooking the limitations of certain AI models can expose businesses to risks such as poor performance, biased outcomes, or operational inefficiencies. For example, in 2021, the real estate marketplace company Zillow used AI to try to judge the housing market as part of its house buying program. Flawed buying algorithms failed to anticipate a market downturn, having been trained on data relating only to a booming property market. The company lost $500 million, had to lay off 2000 people and shutter its buying iBuying subsidiary.[2]

On a broader level, a lack of understanding can stall innovation, as leaders may hesitate to embrace AI due to misconceptions or fear of failure.

Overview of the different types of artificial intelligence

AI encompasses a diverse and interrelated set of technologies with different capabilities to simulate human intelligence. These different types of AI, and their relationship to each other, is shown in Figure 1. A brief overview of the five of the major types of artificial intelligence follows below.

Different types of AI — Figure One. Different types of Artificial Intelligence technologies.

Expert Systems.

These systems, developed in the 1970s, are one of the earliest forms of AI. Expert systems mimic the decision-making ability of a human expert in specific domains by following a set of pre-defined rules. Expert systems incorporate a knowledge base which contains facts and rules, and an inference engine that applies the rules to the knowledge base to derive recommendations.[3] Users pose questions to the expert system, which then searches its knowledge base and applies the rules to answer questions and make recommendations.

Applicability: They are well-suited for structured environments with clear rules, such as compliance checks, medical diagnosis, or troubleshooting technical issues.
Strengths: Reliable, explainable, and highly accurate within their domain.
Weaknesses: Limited adaptability; they cannot learn or evolve without human intervention.
Ideal Use Cases:
- Tax preparation software that ensures compliance with regulations.
- Customer support systems that provide scripted solutions.
- Early diagnostic tools for diseases or equipment failure.
Real-World Example: TurboTax uses rule-based systems to guide users through tax filing.
What It Should Not Be Used For: Unstructured or ambiguous environments where the rules are not clearly defined.
State of Maturity: Highly mature but less commonly used in modern AI development due to their rigidity.

Overall Assessment: Expert Systems are great for static, well-defined processes but cannot adapt to new challenges.

Natural Language Processing (NLP).

Natural Language Processing is the technology behind the interaction between humans and computers through natural language. NLP focuses on enabling computers to understand, interpret, and generate human language. It combines computational linguistics, machine learning, and deep learning to process and analyze large volumes of text and spoken words. NLP enables computers to understand human language as it is spoken or written, allowing for tasks such as translation, sentiment analysis, and automated responses.

Applicability – best suited for activities where interaction with written or spoken natural language is needed or necessary
Strengths: real-time processing of user requests, facilitates communication and interaction across different languages, processing and analysis of large documents
Weaknesses: struggle with accents, slang, or ambiguous phrases; can misinterpret sarcasm or context-specific phrases; limited understanding of complex queries; lack understanding of idiomatic expressions or cultural context; limited ability to discern the different meanings of words in different contexts.
Ideal Use Cases
- Personal virtual assistants carrying out duties or tasks
- Sentiment analysis determining the sentiment (positive, neutral, negative) behind text
- Chatbots interacting with customers for sales inquiries or technical support
- Machine translation translates text from one language to another
- Document summarization of large documents into concise versions while retaining key information.
Real World Example: Amazon’s Alexa can control smart home devices, play music on command, and provide weather updates by interpreting user queries; Google Translate allows users to translate text in over 100 languages; Sephora using a chatbot on their website to help customers find products based on their preferences.
What it should not be used for: Situations requiring deep contextual understanding or emotional intelligence, sensitive communications of topics requiring human oversight, and legal documentation
State of Maturity – Highly mature but challenges remain regarding understanding context and cultural nuances

Overall Assessment: Natural Language Processing provides significant benefits by enabling businesses to harness the potential of human language for various applications, but adopters must be aware of its limitations in processing and understanding accents, nuances and contexts in languages.

Machine Learning (ML).

This approach is the foundation of modern AI. ML enables systems to learn from data and improve performance over time without being explicitly programmed. Unlike expert systems that use rule-based reasoning and explicit knowledge representation, ML relies on analyzing statistics from vast datasets to identify patterns and relationships.

Applicability: Ideal for tasks where patterns can be identified in data, such as fraud detection, customer segmentation, and predictive analytics.
Strengths: Adaptive, scalable, and effective for data-driven decision-making.
Weaknesses: Requires large datasets and can behave like a “black box” with limited explainability; vulnerable to bias from partial or incomplete datasets.
Ideal Use Cases:
- E-commerce platforms predicting customer preferences.
- Financial institutions detecting fraudulent transactions.
- Predictive maintenance for manufacturing equipment.
- Reinforcing and fine-tuning physical models over time.
Real-World Example: Netflix’s recommendation engine uses ML to suggest content based on user behavior.
What It Should Not Be Used For: Scenarios where interpretability and transparency are critical, such as regulatory compliance or safety-critical applications.
State of Maturity: Highly mature and widely deployed across industries.

Overall Assessment: ML and DL (discussed below) are powerful for large-scale data applications but may require significant investment in data preparation and computing infrastructure.

Deep Learning (DL).

Deep learning is a subset of ML that uses single or multilayered neural networks to process large amounts of data. Neural networks mimic the human brain, allowing them to learn and make decisions in a way that’s more similar to human cognition. For example, to recognize an image, the first layers of a neural network detect low-level features like lines and edges from raw data. The information from the initial layers is then passed on to the deeper layers which combine the lower-level features to recognize higher level concepts like objects and scenes. Unlike machine learning which requires some manual intervention to transform the raw data into relevant information before it is used by ML models (“feature engineering” [4]), deep learning automatically extracts features from the raw data.

Applicability: Best suited for tasks requiring high levels of accuracy and complexity, such as facial recognition, natural language processing, and autonomous driving. It’s particularly effective for complex tasks such as image and speech recognition.
Strengths: Exceptional accuracy for complex problems, able to uncover subtle patterns in data.
Weaknesses: Computationally intensive and requires significant amounts of labeled data.
Ideal Use Cases:
- Image classification in healthcare for detecting diseases.
- Voice assistants like Siri or Alexa.
- Autonomous vehicle systems.
Real-World Example: Tesla’s self-driving car technology relies heavily on deep learning.
What It Should Not Be Used For: Small datasets or real-time applications with limited computational resources; in the absence of human oversight, mission or safety-critical applications.
State of Maturity: Advanced but resource-intensive, making it challenging for smaller firms to adopt.

Generative AI.

Generative AI creates original content such as text, images, video, audio, or software code in response to user requests. One well-known example of generative AI are models like GPT (Generative Pre-trained Transformer). Generative AI uses sophisticated deep learning models, particularly large language models (LLMs), trained on vast amounts of raw unstructured data to identify patterns and relationships. This knowledge is then used to predict the next item in a sequence. For example, generative AI models can predict the next word in a text string based on the previous words and the context those words were used in. Similarly, generative AI models can create a sharper version of the image based on an analysis of the original image.

A subset of generative AI, called Agentic AI, operates autonomously and proactively to complete tasks and actions. In doing so, it collects information, makes decisions, solves problems, enlists other agents, and triggers business processes, all without human intervention.

Applicability: Content creation, idea generation, and enhancing productivity in creative tasks. For agentic AI systems, applicability includes advanced business process automation and supply chain optimization.
Strengths: Creative, versatile, and can operate across multiple media formats. Agentic AI systems can clink to operational systems to trigger responses and learn from those.
Weaknesses: Risk of generating inaccurate or biased outputs, and requires careful oversight. Training data may need to be updated over time to allow the system to recalibrate itself.
Ideal Use Cases:
- Marketing teams generating ad copy or visuals.
- Software development generating boilerplate code.
- Content creators developing scripts or designs.
- Supply chain optimization (agentic AI)
Real-World Example: OpenAI’s ChatGPT used for drafting emails, reports, or code.
What It Should Not Be Used For: Mission-critical tasks requiring high accuracy and reliability, such as medical diagnoses or safety critical processes.
State of Maturity: Rapidly advancing, with growing adoption but challenges in reliability and dependability. Hallucinations (where the AI system invents answers that have no relation to reality) are a major issue.

Overall Assessment: Generative AI (and agentic AI) offers creative possibilities but needs oversight (“human in the loop”) to avoid errors or biases.

Summary of the different AI types

Key aspects of the different types of AI is summarized in Figure Two below.

AI Type	Best For	Not Suitable For	Example Applications
Expert Systems	Structured environments with clear rules.	Unstructured or ambiguous environments where the rules are not clearly defined or are still evolving.	Scripted customer support Scripted equipment diagnosis and troubleshooting
Natural Language Processing	Interaction with written or spoken natural language is needed or necessary.	Situations requiring deep contextual understanding, cultural understanding, emotional intelligence.	Personal virtual assistants Sentiment analysis Chatbots Machine translation
Machine Learning	Tasks where patterns can be identified in data	Scenarios where interpretability and transparency are critical, such as regulatory compliance or safety-critical applications.	Customer preference recommendations Fraud detection Predictive maintenance
Deep Learning	Tasks requiring high levels of accuracy and complexity.	Small datasets or real-time applications with limited computational resources; mission or safety-critical applications without human oversight.	Image classification and recognition Autonomous vehicles
Generative AI	Content creation, idea generation, and enhancing productivity in creative tasks. For agentic AI, autonomous task execution.	Mission-critical tasks requiring high accuracy and reliability	Content (text, graphics, music) creation Software code generation Supply chain optimization (agentic AI)

Figure Two. Summary table of the different types of AI.

Next Steps: Applying this knowledge

leaders and managers can put this understanding of the different types of artificial intelligence systems to use in the following ways:

Understand the nature of the problem or task you are considering AI for
Review the table of different AI types and match the right AI approach to your specific application
Ask questions to understand the type of AI technology used in AI-enabled solutions offered by vendors to confirm it matches the challenge or task you are trying to address
Look “under the covers” of the marketplace solutions to differentiate between genuine AI solutions and those that may be using the term “AI” as a marketing buzzword without delivering substantive AI functionality
Ask questions to understand and consider potential risks associated with using AI on your challenges and tasks

In the next article, we will discuss key parameters business leaders and managers must consider in matching and assessing the AI-problem fit.

Conclusion

AI offers immense potential for mid-market companies, but understanding the landscape is crucial to making informed decisions. Each type of AI has a role to play in driving efficiency, enhancing customer experiences, or unlocking new revenue streams. By understanding the different AI types and aligning the strengths of AI with your business needs and resources, business leaders and managers can confidently take the next step toward innovation.

References

[1] “What is artificial intelligence (AI)?”, IBM, August 9, 2024. Link.

[2] “Zillow to shutter home buying business and lay off 2,000 employees as its big real estate bet falters.” T. Soper, GeekWire, November 2, 2021. Link.

[3] Expert System. Wikipedia. Link.

[4] “What is feature engineering?”, IBM, January 20, 2024. Link.