Understanding Artificial Intelligence
Saying that Artificial Intelligence (AI) is misunderstood would be an understatement. Most people have no idea what it is, how it works or what it can do. This is made worse by the fact that contrary to other technological revolutions, AI are not physical machines but another kind of software; immaterial. Hence, in this article I wish to demystify AI by presenting its various domains with respect to its various applications, data, and methods used.
Before we dive into the various domains of AI, here is a simple definition : An AI is a piece of software that can be taught to solve problems. It differs from other kinds of software in that its behaviour is not fully decided upon by its developer but also by what it has learned through training. It is noteworthy that while most people would use the term AI, practitioners mostly use the term Machine Learning or ML instead.
Domains by kind of applications
First, let’s discuss what we can do with AI systems.
AI systems can be used for making predictions and helping with decision-making. Examples include deciding if an email belongs to the spam box or not, what would be the right price for a flight ticket or how to translate text from another language. As our world gets more complex, more human decisions are being offloaded to AI systems. Hence, AI is expected to help us solve the biggest challenges facing humanity such as climate change by finding more efficient uses of resources and energy.
Information extraction is also another strength of AI. Such AI systems can look at the equivalent of multiple libraries a day and find what you are looking for. This is essential in our world where the amount of data available is exploding in an exponential manner. For example, personalised systems such as Gogle search or YouTube can learn what you tend to search and thus what you are likely looking for.
Of course, there exist other applications that do not fit neatly into these two categories. For example, there exists AI that learn mathematical representations of words which allows us to apply algebraic operations on language.
Domains by kind of data
AI systems can also be categorised with respect to what kind of data it ingests.
AI can be used for analysing images. This includes facial recognition, self-driving cars or cancer detection in medical imagery. In 2012, it was image recognition that revived AI from its decades long hiatus and marked the start of a new boom for AI.
Natural Language Processing (NLP) is the study of AI systems applied to textual data. It is somewhat different to other AI fields as it has roots in both computer science and linguistics. NLP is concerned with problems such as machine translation, automatic summarization, chat bots, trend extraction, … It is also my field of study.
Finally, AI can also be used for tasks on numerical data. Predicting the stock market, deciding whether an individual can have access to a loan or finding anomalies in weather patterns. Working with numbers may not be as sexy as working with images or text but it still remains an important part of the study of AI.
There exists many other kinds of data that can be leveraged by AI systems. This includes graph data such as maps or electrical networks which can be analysed to improve traffic or energy usage.
Domains by kind of training
There exist many ways of creating AI systems. Supervised methods are what first comes to mind when discussing AI. Supervised learning consists in providing examples to an AI and teaching it to distinguish between them. One could show an AI system examples of cancerous skin tumours and benign skin moles and train it to recognise cancerous growth from the rest. The problem is that supervised learning requires lots of manually annotated examples for the AI to learn which can be expensive to produce.
Unsupervised AI systems can learn to organise data or find anomalies. Contrary to supervised systems they do not require human labelled data. Such systems are able to find patterns hidden in a dataset without human intervention. For example, given millions of genetic sequences of a virus population, such AI can organise them and find potential variants.
Finally, we have agent-based AI systems. Such systems are composed of agents that can interact in a virtual world or the real one and learn how to behave from its mistakes. Currently, financial markets all around the world use such systems to trade with each other at an extremely rapid rate that humans cannot compete with. Through trial and error, they have learned how to make money as fast as possible.
Once again, there exist many other kinds of AI systems. For example, semi-supervised learning consists in training an AI with limited data and letting it learn the rest by itself. While such methods are more difficult to implement they allow us to reduce the cost of annotating the data.
Is your company looking to integrate AI solutions, analyze data, or strengthen its back-end development? Contact me!
Mail : pro.judicael.poumay@gmail.com
Linkedin : Judicaël Poumay
Portfolio : www.judicael-poumay.be
- ChatGPT and Large Language Models : Key Historical Milestones in AI - 22 August 2024
- Why non linearity matter in Artificial Neural Networks - 18 June 2024
- Is this GPT? Why Detecting AI-Generated Text is a Challenge - 7 May 2024
