AI, Machine Learning, and Deep Learning: Definitions and Relationships
Artificial Intelligence (AI) is a field of computer science that gives machines the methods and algorithms to mimic human reasoning and decision-making. Machine Learning (ML) is a subset of AI: algorithms that learn from data rather than following hard-coded rules. In this post, I sort out the differences between AI, ML, and Deep Learning, and help you work out which one you actually need.
Artificial Intelligence and Machine Learning
AI-generated Art with Jasper, December 2022
I update this article periodically with new ideas, so click here and save this blog post to your favourite Pinterest board. Pinning it will ensure you can refer to this detailed article later. PIN
I like the concise definition of AI on Wikipedia :
Artificial intelligence (AI) is intelligence demonstrated by machines, as opposed to natural intelligence displayed by animals including humans.
AI turns up in more places than you might expect. It can detect tumour cells or anomalies in medical scans [1], expert systems streamline online hotel bookings, face recognition unlocks your phone screen, and chatbots walk you through an online help desk.
AI agents such as the Roomba vacuum cleaner use sensors to understand their environment and make autonomous decisions while cleaning your floors, then use actuators to act on what those sensors tell them. Connect physical AI agents like this to the internet, and you get the Internet of Things (IoT). Ideally, IoT devices exchange data over the network to create intelligent environments such as smart homes or home security systems.
Self-driving cars, chatbots, and recommendation systems are implemented with AI algorithms. Even your coffee machine or dishwasher could contain algorithms to make the device more user-friendly, or “more intelligent.” These are embedded systems built on specialised microprocessors tailored to specific applications, and they keep getting more complex and capable. If you want to go deeper, read the paper by Per Lynggaard and Knud Erik Skouby (2016) on smart cities, homes, IoT, AI, and embedded systems: Complex IoT systems as enablers for smart homes in a smart city vision.
Artificial Intelligence (AI) is a rapidly growing field. These are just a few examples of the most popular AI apps today:
- Digital assistants: Digital assistants, such as Siri, Alexa, and Google Assistant, use Natural Language Processing (NLP) and Machine Learning (ML) to understand and respond to voice commands, answer questions, and perform tasks, such as setting reminders and playing music.
- Recommendation systems: Recommendation systems, such as those used by Amazon, Netflix, and Spotify, use ML to analyze user data and provide personalized recommendations for products, services, and content.
- Image and video recognition: Image and video recognition apps, such as Snapchat, Instagram, and TikTok, use DL algorithms to analyze and manipulate images and videos in real-time, allowing users to add filters, effects, and animations.
- Autonomous vehicles: Autonomous vehicles, such as self-driving cars and drones, use ML and DL to process sensor data and make decisions in real-time, enabling them to navigate and interact with the environment safely and efficiently.
- Healthcare: AI is being used in healthcare for various applications, such as medical imaging analysis, drug discovery, and personalized treatment planning.
- Fraud detection: AI is being used by financial institutions and cybersecurity companies to detect and prevent fraud and cyber attacks.
More complex systems need more sophisticated, computationally demanding methods that work with enormous amounts of data. Historical data, system or monitoring logs, user visits, visual data, user inputs, or tags can all be used to train AI systems.
Machine Learning: Algorithms and Use Cases
Machine Learning (ML) is a subset of AI that trains algorithms to learn patterns from data and make predictions or decisions without being explicitly programmed for each task. For instance, you can use a customer’s past purchases to predict when they’re likely to buy again and recommend relevant products. Movie and music recommenders work the same way — they learn from historical user data with the help of ML algorithms.
In brief, ML algorithms observe patterns in data and learn from them. You organise the data by defining features and providing labels, so the algorithm has something structured to learn from, and it can process that data efficiently given enough computational resources — disk space, memory, and time. There’s a tradeoff between how well an algorithm learns from the data you give it and how well it generalises to data it hasn’t seen. In the most extreme case, the algorithm learns the data “by heart” — overfitting — and performs poorly on anything new. To adjust to new data, you might need to retrain the model or add new labels or categories. I dig into data-size tradeoffs and overfitting in more depth in later posts.
I am affiliated with and recommend the following great ml_books at no additional cost to you.
Machine Learning Crash Course for EngineersA concise, illustrated guide to machine learning with practical Python examples for technical professionals. Covers algorithms, model building, and performance evaluation with applications in signal processing, energy systems, and robotics. |
|
|
|
Designing Machine Learning SystemsA practical guide focused on the challenges of operationalizing ML systems in production, covering data engineering, ML lifecycle automation, and responsible AI practices. |
|
|
|
Hands-On Machine Learning with Scikit-Learn, Keras, and Tensorflow. Concepts, Tools, and Techniques to Build Intelligent SystemsRecent advancements in deep learning have made machine learning accessible to novice programmers. Aurélien Géron’s updated third edition of his bestselling book provides practical examples using Python frameworks like Scikit-Learn, Keras, and TensorFlow. |
|
|
|
Python Machine Learning. Machine Learning and Deep Learning with Python, scikit-learn, and TensorFlow 2This third edition of the bestselling Python machine learning book combines a strong theoretical foundation with practical applications, now updated for TensorFlow 2, GANs, and reinforcement learning. It offers clear and intuitive explanations that delve deeply into both the theory and practice of machine learning. The revisions ensure that readers are equipped with the latest best practices and advancements in the field. |
|
|
|
Here are some examples of Machine Learning algorithm and their applications:
- Linear regression: Linear regression is a simple and widely used ML algorithm that can predict a continuous value, such as a house’s price, based on size and location.
- Decision trees: Decision trees are used for classification and prediction tasks and can be easily interpreted by humans. They are used in many applications, such as credit scoring and medical diagnosis.
- Random forests: Random forests are an extension of decision trees that can improve accuracy and reduce overfitting. They are used in many applications, such as fraud detection and image classification.
- Support Vector Machines (SVMs): SVMs are robust algorithms for classification and regression tasks. They are used in many applications, such as face recognition and text classification.
- Clustering: Clustering algorithms group similar data points based on their features. They are used in many applications, such as customer segmentation and anomaly detection.
The choice of algorithm depends on the task at hand, the available data, and the performance requirements.
If you want to see ML applied in the wild, here’s a sample of published research: banking supervision using K-Nearest Neighbours (KNN), support vector machines (SVM), tree-based models, and ensemble techniques [2]; rheumatoid arthritis prediction with XGBoost [3]; and wind energy prediction with Artificial Neural Networks, Support Vector Machines, Regression Trees, and Random Forests [4].
ML is a powerful approach to AI, but it’s not the only one. Other approaches include rule-based systems, expert systems, evolutionary algorithms, and fuzzy logic systems. ML’s popularity took off thanks to the availability of large amounts of data, faster and cheaper computing power, and advances in algorithms and models — particularly Deep Learning, which has achieved breakthrough results across many domains.
Deep Learning: Neural Networks for Complex Tasks
Structured data isn’t always a “must” for building AI systems — sometimes you can extract features automatically from raw data [5]. Deep Learning algorithms sit a level up in complexity: they typically need more data and more computational power than classic ML. Deep Learning is a subset of ML that learns models from data while building a hierarchy of concepts. It builds a “network” loosely inspired by neurons in the human brain. The Artificial Neural Network (ANN) consists of multiple input, output, and hidden layers. More layers means a deeper network — hence the name — and more capacity for complex tasks. A widely cited survey [5] on Deep Learning applications and methods explains the algorithms, their architecture and complexity, generalisability, hardware requirements, and parameter tuning in more depth.
Deep Learning has seen rapid growth and adoption across various industries, and countless applications are available today. Here are some examples:
- Image and video recognition: Deep Learning is widely used for image and video recognition tasks, such as object detection, facial recognition, and gesture recognition. Applications include self-driving cars, security systems, and healthcare.
- Natural Language Processing (NLP): NLP is the field of AI that deals with human language, and Deep Learning has made significant advances in this area. Applications include language translation, chatbots, and sentiment analysis.
- Recommendation systems: Deep Learning algorithms can analyze large amounts of user data to make personalized products, services, and content recommendations. This is used by companies like Amazon, Netflix, and Spotify.
- Fraud detection: Deep Learning can analyze large datasets and identify patterns that indicate fraudulent activity, such as credit card fraud, insurance fraud, and cybersecurity threats.
- Healthcare: Deep Learning analyses medical images, identifies diseases, and develops personalized treatment plans. It’s also being used to create new drugs and treatments.
Is ChatGPT also created with Deep Learning technology?
Yes. ChatGPT is a large language model built on the GPT (Generative Pre-trained Transformer) architecture, a deep learning approach that has produced strong results across natural language processing tasks such as language generation, translation, and sentiment analysis.
The GPT architecture is a multi-layered neural network built around a self-attention mechanism that captures dependencies between different parts of a text sequence. The model is pre-trained on a large corpus of text using unsupervised learning, then fine-tuned with human feedback: OpenAI confirms that ChatGPT was fine-tuned from a GPT-3.5 model using reinforcement learning from human feedback (RLHF), the same technique behind its InstructGPT models.
That GPT-3.5 lineage traces back to GPT-3, which has 175 billion parameters, as OpenAI disclosed in the original paper. OpenAI hasn’t published parameter counts for the models powering ChatGPT since — its GPT-4 technical report explicitly withholds architecture and size details, citing the competitive landscape. Whatever the current number, ChatGPT can still produce incorrect or inappropriate responses, especially on sensitive or complex topics, so I’d verify anything it tells you before relying on it.
These are just a few examples of the many applications of Deep Learning. We can expect to see even more exciting applications with ongoing research and development.
Deep Learning vs Machine Learning: Architecture, Data, and Performance Differences
Here are five key points that highlight the differences between Deep Learning and Machine Learning:
- Architecture: One of the main differences between Deep Learning and Machine Learning is their architecture. Machine Learning algorithms typically use linear models or shallow neural networks with a small number of layers. In contrast, Deep Learning algorithms use deep neural networks with many layers, allowing them to learn more complex data representations.
- Feature extraction: Machine Learning algorithms often require manual feature extraction, which involves selecting and engineering relevant features from the data. Deep Learning algorithms can automatically learn feature representations from raw data without requiring manual feature extraction.
- Data requirements: Deep Learning algorithms typically require large amounts of data to train, while Machine Learning algorithms can often achieve good results with smaller amounts of data.
- Hardware requirements: Deep Learning algorithms are computationally intensive and often require specialized hardware, such as Graphics Processing Units (GPUs) or Tensor Processing Units (TPUs), to accelerate training and inference, while Machine Learning algorithms can often be run on standard hardware.
- Performance: Deep Learning algorithms have achieved state-of-the-art performance in many domains, such as image recognition, speech recognition, and natural language processing, while Machine Learning algorithms may not be as powerful in complex tasks but can still be effective for simpler problems or when data is limited.
I am affiliated with and recommend these fantastic Deep-Learning Books Books
Deep Reinforcement LearningAn overview of deep reinforcement learning covering basic algorithms to advanced topics like multi-agent RL and meta learning. Ideal for graduate AI courses. |
|
|
|
Deep Learning with PyTorch Step-by-Step. A Beginner's Guide. Volume III. Sequences & NLPLearn sequence models and NLP with PyTorch 2.x in an easy-to-follow approach. Covers RNNs, Transformers, BERT, GPT-2, and using HuggingFace for fine-tuning language models. |
|
|
|
Choosing Between Machine Learning and Deep Learning
AI and ML are not going anywhere, and the tools keep getting more accessible. From healthcare to finance to the camera on your phone, they’re already solving problems that were unsolvable a decade ago. Machine Learning is the practical entry point for structured-data tasks; Deep Learning is the right choice when your data is unstructured, large-scale, and feature engineering is impractical.
Disclaimer: I have used chatGPT by OpenAI for updating this post on 12th April 2023.
Did you like this post? Please let me know if you have any comments or suggestions.
Posts about Machine Learning that might be interesting for youRelated tools you may want to try next.
CustomGPT.AI is a very accurate Retrieval-Augmented Generation tool that provides accurate answers using the latest ChatGPT to tackle the AI hallucination problem.
Bibliography
- Ursuleanu, T.F., Luca, A.R., Gheorghe, L., Grigorovici, R., Iancu, S., Hlusneac, M., Preda, C. and Grigorovici, A., 2021. Deep Learning Application for Analyzing of Constituents and Their Correlations in the Interpretations of Medical Images. Diagnostics, 11(8), p.1373.
- Guerra, P. and Castelli, M., 2021. Machine Learning Applied to Banking Supervision a Literature Review. Risks, 9(7), p.136.
- Messelink, M.A., Roodenrijs, N.M., van Es, B., Hulsbergen-Veelken, C.A., Jong, S., Overmars, L.M., Reteig, L.C., Tan, S.C., Tauber, T., van Laar, J.M. and Welsing, P.M., 2021. Identification and prediction of difficult-to-treat rheumatoid arthritis patients in structured and unstructured routine care data: results from a hackathon. Arthritis Research & Therapy, 23(1), pp.1-10.
- Buturache, A.N. and Stancu, S., 2021. Wind Energy Prediction Using Machine Learning. Low Carbon Economy, 12(01), p.1.
- Alzubaidi, L., Zhang, J., Humaidi, A.J., Al-Dujaili, A., Duan, Y., Al-Shamma, O., Santamaría, J., Fadhel, M.A., Al-Amidie, M. and Farhan, L., 2021. Review of deep learning: Concepts, CNN architectures, challenges, applications, future directions. Journal of big Data, 8(1), pp.1-74.
- Embedded System Definition
-
Lynggaard P, Skouby KE. Complex IoT systems as enablers for smart homes in a smart city vision. Sensors. 2016 Nov 2;16(11):1840.
- ChatGPT (OpenAI)
References
- Introducing ChatGPT (OpenAI)
- Brown, T. et al., 2020. Language Models are Few-Shot Learners (GPT-3 paper)
- Vaswani, A. et al., 2017. Attention Is All You Need
- Ouyang, L. et al., 2022. Training Language Models to Follow Instructions with Human Feedback (InstructGPT)
- GPT-4 Technical Report (OpenAI)
Related Reading
Enjoyed this? Get more like it.
Weekly notes on AI tools, Python, and what I'm actually building — plus a free copy of Fantastic AI: The 2026 Toolkit.
 is a field of computer science. AI provides methods and algorithms to mimic human intelligence, reasoning, and decision-making and provide insights, which businesses could use in research or industry to build new exciting and innovative products or services. Machine Learning (ML) is a subset of AI with algorithms that learn from data. In this post, we sort out the differences between AI and ML.&method=bookmarklet){kind=link}