Artificial intelligence (AI), machine learning, data science and advanced analytics are just some of the terms that are used a lot, often mixed up and randomly related to each other.
Without trying to take you through a six-week term at your local university, we’ll try to explain some of these terms and more importantly, establish how we at IFS interpret them. For that purpose, I want to introduce IFS Labs AI engineer, Martijn Loos, who has co-written this blog.
Defining artificial intelligence
In the previous blogs (“Artificial intelligence myths,” “Why the hype around AI?” and “Are chimpanzees more intelligent than computers?”), we have already established a definition for artificial intelligence.
AI is the theory and development of computer systems being able to perform tasks better or as good as humans. We also discussed that although the future of AI might be general intelligence, today’s version of AI is very much about specific intelligence, and that’s where companies should invest. Can we take specific tasks in our business applications or processes and augment our users with AI capability to better perform their work? An example here would be using AI technologies to help doctors interpret MRI or ECG images when diagnosing a patient. This is about augmenting the doctor, not about replacing him.
Within the field of AI, there are multiple sub-disciplines. Many applications involving AI utilize more than one of them. For example:
Computer vision focuses on automating tasks involving digital images and videos. Examples are face recognition from photos or videos (e.g. automatically being able to tag friends in your photos on social media), reconstructing a 3D image from photos or the tracking of persons on video images.
Natural language processing
Natural language processing is applied where written text or speech needs to be interpreted or generated. Examples range from generic applications like Siri or Cortana to specific chat bots such as the KLM Messenger Bot to sentiment analysis algorithms that can interpret the sentiment behind an article or tweet.
Autonomous agents are able to progress towards their goal, without interference. A good example is drones. They can be programmed in such a way that they can fly towards a goal or stay as close as possible to a person, without anyone having to control the drone, all the while avoiding possible objects in its path.
Then there are applications which use all of the above-mentioned disciplines. Think of self-driving cars. This is, by definition, an autonomous agent because there is no one controlling the car, while it still has to follow the traffic rules. It does this by using computer vision techniques which continuously process images of the surroundings. We can speak to the car to tell it our destination, it will understand where to go and map the shortest route to go there, which is done by using natural language processing.
To emphasize some of our earlier statements, this also means that a self-driving car is not one single AI that figures everything out, but rather a complex computer system that uses several smaller, specific AI features, each individually addressing a very specific need.
Key AI discipline: machine learning
Until now, we have not even talked about the most important AI discipline: machine learning. It is one of those terms that is thrown around a lot, but what does it actually mean?
Machine learning is a term to describe a set of algorithms that can learn by themselves. This is achieved by feeding the algorithm many examples (data) of the task it should perform so it can extract patterns to achieve its goal. Once the pattern is known, it can be applied to data the algorithm has not seen before. For example, to make predictions.
Neural networks are often talked about. However, it is not some separate technique but rather one of many machine learning algorithms.
A neural network is modeled to mimic the behavior of the human brain. Where a human brain contains neurons, connected with synapses that can send signals to each other, similarly a neural network contains artificial neurons which can send signals to other neurons. These neurons are usually organized in layers; a typical configuration contains an input layer, a certain number of hidden layers and an output layer.
An example of using this method is trying to identify if a picture contains a cat. To let the network learn this, many pictures containing cats are used as input for the model. The hidden layers are used to learn and identify the different characteristics of cats, such as whiskers, ears, a tail, etc.
The result will come from the output layer, returning if the picture actually contains a cat or not.
The more hidden layers there are in the network, the ‘deeper’ it is, and the more complex patterns it can represent. When we are talking about deep learning, we actually mean neural networks that have many hidden layers.
Although machine learning is its own discipline, it is also interwoven in all other disciplines of AI. Trying to learn a neural network to identify cats in pictures is a good example of how machine learning and computer image processing are related.
AI-related terms: data science and advanced analytics
Data science is the application of a diverse range of methods and techniques in order to gain insight from data. Machine learning is one of the most important techniques to achieve this, however, there are other tools as well.
When developing an AI application, data science is required. Without insights from the data, it is impossible to develop a proper, usable model. A data scientist will try to find the most suitable algorithm to use, structure and clean the data, optimize parameters and interpret results.
Another related term is advanced analytics. It is often used as an umbrella term for tools and techniques that can autonomously gain insights from data and are more ‘advanced’ than traditional analytical tools, such as business intelligence (BI) or historical data analysis. Techniques one can think of are data mining, graph analysis, big data analytics and machine learning.
When comparing data science to advanced analytics, we can ask ourselves if there is an actual difference between the two. We think the terms are very similar to each other. Often you see that advanced analytics is used in the context of a product (e.g. advanced analytics platform), while data science is used as an activity (e.g. applying data science).
If you want to learn more about IFS’s view on AI, there will be more blog posts on this topic in the coming weeks. Of course, you can always reach out to us below.
Do you have questions or comments about AI?
We’d love to hear them so please leave us a message below.