Artificial intelligence (AI)
Let’s begin by stating a simple definition of what AI is exactly:
Computers behaving intelligently and autonomously, versus computers doing what they are told
For decades, software engineers have been constructing expert systems: tightly constrained, rules-based software solutions that perform a single task very well.
There’s nothing wrong with that! In many cases it’s the right answer. Think about updating the firmware on your TV. You can look for it on demand via the Setup menus, or let the TV look for it once in a while, but it’s a problem best solved with traditional technologies. It doesn’t require intelligence.
Now think about driving a car around in a big city: that requires intelligence! We were getting no where fast with driverless cars until the rise of AI.
Types of AI
So what are the types of AI? There are three.
The first type is narrow AI: think highly focused. Apple’s Siri on the iPhone is very helpful with phone-related tasks, but try asking it "why is my partner mad at me today?". It’s not going to be very helpful.
Spam vs Ham filters for email: when is the last time you had to setup a Junk mail rule? You don’t, AI has essentially solved that problem.
Face recognition on Facebook – very focused and wickedly accurate – but the underlying ML implementation would be useless if you tried using it to fly a drone.
This is what the _narrow_ness is about: extremely performant AI solutions in a limited setting.
The next type is general AI – think of the kind of intelligence that would rival human capacities. It doesn’t exist yet.
The human capacity to adapt to the challenge at hand underscores the agility and flexibility of our intelligence. If you and I needed to learn to drive a forklift, we could. If we needed to master flying a drone, we could. This kind of generalized intelligence that can be directed at a wide variety of tasks has not been achieved in AI.
The last type of AI is super AI – now we’re talking about the kind of intelligence that we can only speculate on, since it exceeds our own.
General AI must first be achieved prior to this type– and speculation on what this might entail has led to many thrilling sci-fi books and movies.
Bottom line: we have no idea. And we wouldn’t create it – general AI computers would.
Where does Machine Learning fit in?

So if we have 3 Types of AI – where does Machine Learning fit in?
Machine Learning is a type of narrow AI: it is purpose-built and focused on a relatively small spectrum of tasks.
Remember: narrow AI is the only type of AI in existence right now, all of the technological miracles we’re immersed in are brought to us courtesy of Machine Learning.
Why Machine Learning?
Because it solves problems traditional approaches can’t
As we’ve discussed, traditional software technologies have been unable to crack certain nuts. These tend to be the areas where Machine Learning shines.
ML is not the solution to every problem – but it is often the solution where previous technologies have failed or come up short.
Remember the maxim: "To someone with a hammer every problem can look like a nail". Don’t go down that road with ML – there are plenty of areas where traditional technologies are the right answer – and will continue to be the right answer.
What is Machine Learning?
Big Data + Algorithm = Model
A function that learns by sifting through mountains of data and gets smarter at its task is a Machine Learning model. This trained model can then be deployed and serve useful as it’s exposed to new data.
Traditional software doesn’t work like this – it doesn’t evolve as its exposed to more data – it just blindly follows the rules it was designed with.
In Machine Learning for example, you can feed a Neural Network (which is a type of algorithm) a ton of images of dogs and cats, and then take that model and deploy it to the cloud (like Microsoft Azure, Google Cloud Platform GCP, or Amazon Web Services AWS) – and then present a new picture to that model and ask it if it sees a dog or cat. This kind of functionality is essentially unsolvable using previous technologies.
What is Machine Learning good at?
Replacing existing systems that require a lot of fine tuning or involve long lists of rules
If you have an existing software system the requires constant tuning or involves a laundry list of rules – you may have a good candidate on your hands. When the ongoing costs of maintaining a system are high this can be a good indication that ML may be a better fit.
Complex problems (think voice recognition, defect identification, etc.) for which traditional solutions have failed
There are many real world problems for which traditional technologies have come up short. Machine Learning is powering the phenomenal advances we’re experiencing by filling in these gaps.
Dynamic, fluctuating environments requiring constant adaptation to new data
Problems that are by definition in a high state of flux (think about new images being posted constantly and needing to be classified) are often best served by ML solutions.
Gaining insight about complex relationships from large lakes of data
Patterns in data are obvious when the data is small (think of a spreadsheet with a 100 rows). As the amount of data grows (think of a spreadsheet with a million rows) it quickly becomes impossible to spot patterns or trends. ML technologies excel in this arena.
The three major branches of ML

Going back to our example on dog versus cat images – this would be an ML model that falls under supervised learning: the algorithm was trained with data for which is was given the correct answer (thus the name supervised).
Unsupervised learning involves algorithms sifting through data for which there is no answer. A recommendation system like Netflix uses is a good example of this kind of ML implementation – the algorithms are looking at viewer’s streaming habits and identifying similarities by which they can make future recommendations. So they’re clustering subscribers together and making intelligent suggestions based on historical streaming habits.
Reinforcement learning is well known in the gaming world: a software agent operating in an environment makes observations, takes actions, and receives feedback (which may be positive or negative). The objective is to learn to maximize positive feedback by trial and error over time. Real world applications abound: games (of course), robotics, smart home devices (like the Nest thermostat), and automated stock market trading – just to highlight a few.
Conclusion
By now you should feel comfortable giving an elevator speech on:
- What is ML?
- Why should you care?
- How does it relate to AI?