As a developer or code dabbler, you might be well aware that AI is a nuanced topic, and that there’s a difference between AI, machine learning, and deep learning. If you didn’t know, that’s also completely reasonable — even experts have a hard time agreeing on definitions for these terms.
Being able to distinguish between the different systems that often fall under the umbrella term of AI is more than just a flex to use when a conversation turns to the topic of AI. It’s a crucial step towards being an informed developer and thoughtful AI practitioner today. Here are the similarities and differences between AI, machine learning, and deep learning that you need to know about.
What is artificial intelligence?
It’s challenging to nail down a definition of artificial intelligence because it’s tough to define intelligence. AI is its kind of intelligence: “AI refers to intentionally-constructed systems that interact dynamically with the world, typically driven by some kind of large data model,” says Ada Morse, Codecademy Instructional Designer in Data Science.
It’s easy to identify AI systems in our daily lives. Computer vision in a self-driving car detects objects and its surroundings so it can make decisions about where to go. A recommender system analyzes data about the TV shows you’ve streamed and produces suggestions for what to watch next. Virtual assistants like Alexa and Siri pick up the words you say and deliver a response.
A common misconception that people have about AI is that these systems and algorithms can understand, think, or reason. AI is trained to be good at a particular thing we optimize it for, so it has a very specific type of “intelligence,” Ada says. “Human beings, while we can specialize in things and there are subjects we might be more naturally prone to, our intelligence is more general.” Whether or not an AI system could replicate the broad capabilities of human intelligence (often referred to as “artificial general intelligence” or AGI) remains to be seen.
That said, AI is not entirely artificial either, because several human decisions have to be made behind the scenes to make AI do what we want it to. “We often think that because it’s an algorithm it’s entirely artificial, and therefore somehow has the purity of mathematical knowledge, versus messy human decisions,” Ada says. “It’s a combo, which is why ‘artificial intelligence’ is such a weird term.”
What is machine learning?
Machine learning is the process of algorithmically detecting patterns in datasets and constructing a model (accurate or inaccurate) of some aspects of reality. Machine learning algorithms are used to train computers to perform lots of tasks in our daily lives. For example, the autocorrect feature on your smartphone is a kind of machine learning called natural language processing, which predicts what words you intend to type next. Your bank likely uses machine learning to determine whether your credit card transactions are accurate or a sign of fraud. And your email inbox uses machine learning to determine which emails to flag as spam.
This chart shows the input and controls that make up an AI system. As you can see, it takes a combination of human decisions and labour, data, real-world input, and algorithms to build an AI system. Want to learn more? Click the chart to check out all of our AI courses and paths.
To recap: AI is an entire system that interacts dynamically with the world, and one component of an AI system might be a machine learning model or algorithm. But a machine learning model on its own doesn’t make up the entirety of an AI system. An AI system includes hardware and software inputs (like computer vision sensors) as well as human-defined controls and fine-tuning that determine how it learns from and interacts with the world. Some AI systems have even used non-machine learning models from fields like game theory.
What is deep learning?
Deep learning is machine learning, where the algorithm is a structure called a neural network. Neural networks are also trained on datasets, and when they encounter novel situations or scenarios, they can make predictions based on that previous dataset. If you know some Python, NumPy, and machine learning basics, you can create a neural network in our path Build Deep Learning Models with TensorFlow.
The idea of “neural nets” that can replicate the behaviour of neurons in the human brain was first introduced way back in 1944. Nowadays modern neuroscience casts doubt on the idea that perceptrons, the artificial models of biological neurons that simulate decision-making, are a proxy for biological neurons. Nevertheless, the analogy has stuck. Deep learning includes multiple layers of neural networks (hence why it’s called “deep”). Thanks to the advancement of high-speed processors, neural networks have the essential computational capabilities to seamlessly merge into our everyday human experiences.
AI refers to intentionally-constructed systems that interact dynamically with the world, typically driven by some kind of large data model.
ChatGPT, for example, uses a complex deep-learning model to predict what words come next and generate human-like text. In healthcare, deep learning models are used to detect abnormalities in medical scans and predict health outcomes. And even in gaming, deep learning is used to enhance graphics and develop non-player characters (NPC) that exist in a game but aren’t controlled by a human.
When we encounter deep learning in our daily lives the output is often so accurate and complex that it can seem convincingly human. “If you talk to some modern-day practitioners of AI, they’ll sort of make the assumption that AI is identical to deep learning — in part because of this false analogy that neural networks are somehow doing something closer to human intelligence than other algorithms,” Ada says.
So, why does this matter?
The difference between AI, machine learning, and deep learning goes beyond terminology. According to Ada, the way we utilize and integrate AI into our lives, as well as how we regulate it as a society, will become a critically significant issue in tech and the world in the years to come.
As a developer, you need to understand the limitations and risks of AI so you can thoughtfully choose the right AI solution for a product you’re building. By gaining a deeper understanding of how the AI tools you employ work, you can exercise a greater sense of responsibility in utilizing this technology.
To use a coding analogy: If you’re using a Python function without understanding its inner workings, you may misuse it and introduce bugs into your code. “It’s the same just on a bigger level for AI,” Ada says.