What is Artificial Intelligence?

Depending who you ask you are likely to get a different answer to this very basic question. As a global network of Artificial Intelligence experts, we feel it is our responsibility to clarify some of these  AI-related concepts which often get confused. Here is a short summary of the most popular buzzwords you may often hear but are still unclear about.

1.     Big Data

This term is often misunderstood as there is no clear definition of how much data there needs to be to classify it as being ‘big’. To put it in simple terms, if it fits into your excel spreadsheet and can be analysed with a pivot table, or can be stored on your laptop, then it is not big data. The term most often refers to unstructured complex and constantly growing datasets which require specific software and infrastructure to be processed and analysed. An example of such software is Spark, which includes a lot of data processing libraries that allow data scientists to analyse and draw conclusions from the big datasets.

2.      Machine Learning

About 60 years ago Alan Turing figured out that machines could potentially learn patterns, like humans, from information that you feed to them. For machine learning algorithms this information is data which can be presented in many different formats, e.g., numbers, words, images. The task of most machine learning algorithms is to find a pattern and to predict or guess something new by following that pattern. There are four main types of algorithms: supervised, unsupervised, semi-supervised, and reinforcement learning; but brevity we will only discuss the first two.

Supervised machine learning algorithms apply supervision on the dataset, i.e., we tell our machine learning algorithm what it is by labelling the data. The supervised learning algorithm then gets better at predicting the new unlabelled cases based on this knowledge. Due to the importance of human input into this process supervised learning algorithms are often prone to bias. E.g., if a data-labelling human classifies a person or an object wrongly, then the machine learning algorithm is likely to make the same mistake in the future.

Unsupervised machine learning algorithms are more independent in the way that they find their own meaning to the data and do not require a labelling process. A good example of unsupervised learning is clustering, which is where an algorithm can find similarities across different groups of people or objects which are not observable by a human eye. This is a tool that is often used by political parties to hyper-target and influence voters, for example during the controversial presidential campaign in the USA in 2016. Using machine learning it was possible to divide the society into groups and focus budget and messaging to those voters that were most prone to change their mind and give a vote to the republican candidate. The effects of this very effective segregation of society are observable to this day.

You can think of machine learning as a toolkit for AI. There are many machine learning algorithms that have been built by skilled data scientists and are ready and available to be used within open-source machine learning software like Python. It is important however to make sure you consult an expert in the field before drawing conclusions from an output of a machine learning algorithm as there are many important aspects such as statistical significance which, when omitted can skew the results.

3.      Artificial Intelligence

And finally, the most mysterious of it all – AI. What is Artificial Intelligence and how can intelligence be artificial? As a matter of fact, AI is the least artificial of the available technologies as it is designed to resemble the natural human brain. The idea was first proposed by Geoffrey Hinton who came up with a concept of a neural network. The neural network is a complex structure of nodes containing information which, similarly to synapses in a biological brain, can transmit signals to other nodes and extracting certain information depending on their weights. A good example of a neural network are image recognition tools, where information carried by nodes allows the system to guess with a certain amount of probability whether an image contains a certain object, or not. Most of the neural network systems have certain level of accuracy associated with their performance which describe how good they are at what they do. Similarly to a kid learning what is a fruit and what is a vegetable, the accuracy normally increases over time.

Neural networks are the most popular example of artificial intelligence, although there are other areas such as genetic programming which also fall under the definition of AI. What is important to realise is that AI systems rarely have a set of eyes or can walk. The robots that we often see in the sci-fi movies fall under a whole different area of science called robotics.

Artificial intelligence and machine learning are both based on data processing and are closely related so the terms are often confused. Accessibility of the off-the-shelf solutions makes it easy for businesses to adopt AI, but it is important to remember that expert advice is necessary to ensure your AI systems are bias-free and to avoid any risks associated with wrong conclusions drawn from an incorrectly deployed AI system.

At Brainpool, we provide Artificial Intelligence Consultancy & strategy services across industries. Thanks to over 300 best in class AI and Machine Learning experts in Brainpool Network we build products that transform industries.​ For more information, please contact our network of Artificial  Intelligence experts: [email protected]