Artificial Intelligence (AI) vs Machine Learning (ML): What's the difference?

From the computational photography in our smartphone camera apps to state-of-the-art chatbots like ChatGPT, artificial intelligence is just about everywhere. But if you look a little deeper, you’ll notice that the terms artificial intelligence and machine learning are often used interchangeably. Despite this confusing narrative, however, AI is still a distinct concept vs ML.

The difference between AI and ML has become increasingly important in the age of advancements like GPT-4. That’s because some researchers believe we’ve taken the first steps toward making computers nearly as intelligent as the average human. Tasks like creative drawing, writing poetry, and logical reasoning were once out of reach for machines and yet, that line has now become blurred.

So with all of that in mind, let’s understand what makes AI different from ML, especially in the context of real-world examples.

What is Artificial Intelligence (AI)?

Artificial intelligence is a very broad term that describes a machine’s ability to perform complex intellectual tasks. The definition has evolved over the years – at one point, you consider perhaps scientific calculators as a form of AI. But these days, we’d need an AI system to perform more advanced tasks.

Generally speaking, anything that can mimic the decision-making abilities of a human can be classified as an AI. Banks, for example, use AI to analyze markets and perform risk analysis based on a set of rules. Likewise, email providers also use AI to detect spam in your inbox. And finally, navigation apps like Apple Maps and Google Maps use an AI system to suggest the fastest route to your destination depending on traffic and other factors.

However, all of these examples fall under the scope of “narrow AI”. Put simply, they only excel at one or two tasks and cannot do much outside their fields of expertise. Imagine asking a self-driving car to win a game of chess against a grandmaster opponent. It simply hasn’t had any training to perform the latter task, while the opposite is true for a specialized AI like AlphaZero.

The rise of Artificial General Intelligence (AGI)

Indeed, most real-world applications we’ve seen so far have been examples of narrow AI. But the depictions of AI you’ve probably seen in movies are known as general AI, or Artificial General Intelligence (AGI). In a nutshell, general AI can emulate the human mind to learn and perform a wide range of tasks. Some examples include critiquing essays, generating art, debating psychological concepts, and solving logical problems.

Of late, some researchers believe that we’ve made strides toward the first AGI system with GPT-4. As you can see in the screenshot below, it can use logical reasoning to answer hypothetical questions, even without explicit training on the subject. Moreover, it’s primarily designed to function as a large language model but can solve math, write code, and plenty more.

However, it’s worth noting that AI can’t completely replace humans. Despite what you may have heard, even advanced systems like GPT-4 aren’t sentient or conscious. While it can generate text and images remarkably well, it doesn’t have feelings or the ability to do things without instructions. So even though chatbots like Bing Chat have infamously generated sentences along the lines of “I want to be alive,” they’re not on the same level as humans.

What is machine learning (ML)?

Machine learning narrows the scope of AI as it exclusively focuses on teaching a computer how to observe patterns in data, extract its features, and make predictions on brand-new inputs. You can think of it as a subset of AI – one of the many paths you can take to create an AI.

To understand how machine learning works, let’s take Google Lens as an example. It’s an app that you can use to identify objects in the real world through your smartphone’s camera. If you point at a bird, it’ll identify the correct species and even show you similar pictures.

So how does it work? Google ran machine learning algorithms on a large dataset of labeled images. A good number of them included different types of birds, which the algorithm analyzed. It then found patterns like color, the shape of the head, and even factors like the beak to differentiate one bird from another. Once trained, it can make predictions by analyzing future images, including those you upload from your smartphone.

Machine learning techniques: How do they differ?

As you may have guessed by now, accuracy in machine learning improves as you increase the amount of training data. However, feeding large amounts of data isn’t the only criterion to make a good machine learning model. That’s because there are many different types of ML, which affects how they perform:

1. Supervised learning: In supervised learning, the machine learning algorithm gets labeled training data, which guides it toward the end result. Imagine one folder full of dogs and another filled with cats. This approach requires a fair bit of human oversight but can lead to more accurate predictions with the same amount of data.
2. Unsupervised learning: As the name suggests, unsupervised learning uses an unlabelled dataset. This means that the machine learning algorithm must find patterns and draw its own conclusions. With a sufficiently large dataset, this is not a problem.
3. Reinforcement learning: With reinforcement learning, a machine learns to make correct predictions based on the reward it gets from doing so. For example, it might learn to play chess by making random actions on a board before realizing the consequences of a bad move. Eventually, it will learn how to play entire games without losing.
4. Transfer learning: This machine learning technique uses a pre-trained model and improves upon its capabilities for a different task. For example, transfer learning can help a model that already knows what a human looks like to identify specific faces. That last bit can come in handy for use cases like facial recognition on smartphones.

These days, machine learning algorithms can crunch extremely large amounts of data. ChatGPT, for instance, was trained on nearly half a terabyte of text.

AI vs ML: What’s the difference?

So far, we’ve discussed what constitutes artificial intelligence and machine learning. But how do they differ?

Let’s take a chatbot like Bing Chat or Google Bard as an example. Broadly speaking, these are examples of AI as they can perform a variety of tasks that only humans once could. However, each of their underlying features depends on ML algorithms. For example, both can understand natural language, identify your voice and convert it to text, and even talk back in a convincing manner. All of these required intensive training, both supervised and unsupervised, so it’s not a question of ML vs AI, but how one augments the other.

FAQs