If you’re here, there is a good chance that you already know the difference between Artificial Intelligence, Machine Learning, and Deep Learning and that it’s the last thing you’re looking for.
But what about the use of AI, Machine Learning, and Deep Learning in self-driving cars?
Autonomous systems are said to use a lot of Machine Learning and Artificial Intelligence.
But where is the frontier?
Artificial Intelligence intervenes in a lot of fields.
When looking at this graph...
Machine Learning and Robotics seem to be in different fields.
Computer Vision and Natural Language Processing seem to be apart from Machine Learning.
Problem-solving, Planning and knowledge representation are not part of Machine Learning at all.
Is there a way to concretely segment the AI field?
First, AI is simply one discipline of the vast Computer Science field.
Second, AI is made of these 6 fields we just saw, and they’re all individual disciplines.
Computer Vision is the science of studying images; a simple “if-then” algorithm performing on images can be considered Computer Vision.
It doesn’t need to use any sophisticated learning.
The same goes for Natural Language Processing which studies text, Planning that studies shortest path algorithms and all the others.
Machine Learning is also an individual discipline as we can simply apply Data Science to any dataset and get an output from it.
🔺The confusion comes when we mention autonomous robots that can understand human language, detect people on cameras, compute intelligent paths to a destination and build a 3D map of their environment while they navigate it.
In this example, all the disciplines mentioned above are used, and we can all put it behind the words Artificial Intelligence.
What has Machine Learning to do with all that?
Machine Learning comes in the play when we mention the science of learning from data.
The AI process is pretty simple: in a defined context, we compute algorithms to output a solution to any given input.
Here, the shortest path algorithm finds the shortest way from A to F.
The context is the graph, the goal is F, the input is all the path durations.
Machine Learning is something different. To solve a problem, we reason in two steps: Learning (from data) and Prediction.
The role of a Machine Learning algorithm is to generate a trained model that can solve a specific problem (classification, regression or clustering).
The prediction is then made on new data using the trained model.
In Machine Learning, we learn a rule from thousands of examples.
In Artificial Intelligence, we hand-design the rule based on mathematics; there is no learning.
To sum it up, and also include recommender systems and reinforcement learning algorithms to it, here’s a map.
To link it with autonomous robots, we can choose to use AI or Machine Learning in any task we want, including autonomous robots.
If we want to understand human intentions, we can use complicated text semantic analyzers or state-of-the-art Deep Learning models.
For specific tasks, like perception, we often combine both.
We can, for example, use Deep Learning for obstacle detection and Bayesian Filtering (Robotics) to track the detected obstacles.
In this case, we don’t use Machine learning everywhere but combine AI and its subset to have better performance.
Something to keep in mind is that eventually, Deep Learning may become the default choice for every task we want to complete.
Today, it is still not the case and a lot of algorithms have used AI for decades without any learning needed.
📩If you’re interested in Self-Driving Cars and the use of AI inside, I invite you to join the daily emails and receive my Self-Driving Car MindMap as a welcome gift!