AI For Beginners
Everybody has to start at somewhere.
Let's start now getting into the basics of AI!
What is AI?
Artificial intelligence (AI) is the ability of machines to imitate human abilities such as logical thinking, perception or creativity. Essential characteristics of such systems are that they perceive their environment with regard to their task, process the impressions and draw conclusions from them. This gives them the ability to solve the task for which they are built, even under certain uncertainties.
For half a century, there have been several approaches to teaching machines such abilities. For example, humans systematically formulate rules for computers to solve certain tasks. Or that they put expert knowledge into a logically structured form that machines can process with simple rules. But it has been shown that these methods only produce limited useful results.
What is Machine Learning? Why is data so important?
Over the last 20 years, machine learning has emerged as the central technology for developing AI. In this process, computers learn the patterns and rules for solving a task by working through a large amount of sample data and finding the statistical relationships and rules themselves in this way. Machine learning algorithms are the programmes that are used to work through the data in order to derive the correlations for solving the task from the sample data.
These statistical patterns are then stored in a model in the form of complex formulae. The model can then apply the task it has been trained to perform to new data. The process of finding these patterns and rules from sample data using algorithms is called training.
What are Deep Learning and Neuronal Networks?
In recent years, this has made it possible to create AIs that solve problems whose rules would be too complex for humans to formulate (e.g. language generation) or are simply not known (e.g. forecasting in different areas such as predicting stock prices). Using machine learning, computers can find these patterns themselves from sufficient sample data.
Neural networks are the most powerful technique currently available among machine learning algorithms that examine data for rules. Deep learning refers to working with neural networks when they have a certain size and thus performance. Processing certain data such as images, speech or audio data tends to require this level of technical performance. Deep Learning has therefore enabled a particularly large amount of progress to be made in these areas.
The processing of sequential data by neural networks was particularly difficult for a long time. This refers to data in which one data point depends on the last, such as language or share prices. Transformers are a special architecture of neural networks that was developed precisely for solving these tasks and today underlie products such as ChatGPT.
What is Generative KI?
Most AI systems are built to recognise whether a new image shows a cat or a dog based on the learned patterns of, for example, cat and dog images. Generative AI is built to generate new data that resembles the sample data as closely as possible based on the patterns recognised from the sample data. In this way, texts, images, audio data, tabular data or new protein structures, among other things, can be generated according to given models.
Large language models are currently the talk of the town. They refer to models that have been created using transformers and trained with language data. Typically, they can be controlled via text input. ChatGPT is a concrete product based on this technology.
