AI Definitions
The most sophisticated algorithms use machine learning, which enables machines to improve with experience. It is useful for establishing huge patterns and relationships, and also for streamlining processes. Deep learning is a class of machine learning, a data processing technique based on multi-layered artificial neural networks. The technique is inspired by the basic functioning of brain neurons. It has been around for more than fifty years ago, but we now have a large enough volume of data and computing power to apply it in a multitude of practical cases. (Source: APDCAT)
Explore the chart
Deep learning is a subtype of machine learning that can process a wider spectrum of data, including text, image, and speech, which requires less human intervention in the preparation of data and that often produces better results than traditional machine learning approaches.
FONT: Catalonia.AI
A set of algorithms that detect patterns, anticipate what will happen (predictive analytics), and learn how to make recommendations on what should or can be done (prescriptive analytics) using data processing. The algorithm also adapts in response to new data to improve efficiency.
- Main machine
learning methods:
- Supervised
learning - Unsupervised
learning - Reinforced
learning
FONT: McKinsey
Artificial intelligence is an advanced digital technology that allows machines to reproduce or master some of the skills that require intelligence when performed by humans. This includes technologies that allow machines
to learn and adapt, sense and interact, reason and plan, optimize procedures and parameters, to extract knowledge from large amounts of data, and operate autonomously.
FONT: Catalonia.AI
Glossari
Source: APDCAT (Report: Artificial Intelligence Automated Decisions in Catalonia,, 2020)
Algorithm Algorithms are everywhere and can be described as the precise sequence of instructions that must be followed to perform a given task. For example, the steps required to enrol in a school or to apply for social benefits. In computing, the computer follows the instructions of the algorithms with which it has been programmed.
Automated Decision Algorithm (ADA). Algorithm that makes decisions without the need for any human intervention. It is estimated that, today, 70% of financial transactions are carried out by ADAs. The potential effects
on people have highlighted important ethical issues, such as liability for decisions (for example, in self-driving car accidents). In the field of health, these should only be used to support the decision of the doctor or specialist. The machine should never decide on the treatment for a patient or diagnose a disease without the supervision of the healthcare professional.
Black box . In deep learning it is not always easy to explain the response of the algorithm or why it has made a certain decision. In these cases, the model is said to behave like a black box because it receives certain inputs and produces an output or result, but how it made the prediction or took the decision is not clear.
Artificial Intelligence (AI). This is a generic term that encompasses those algorithms that seek to endow computers with intelligent behaviour. The concept of intelligent behaviour is not clearly defined. There are two main approaches: General AI (which seeks to equip computers with the intellectual abilities of humans) and weak AI (which focuses on specific tasks such as voice recognition, image recognition, etc.).
Model. Representation of reality used by automated decision algorithms to make predictions and take decisions. This model is tailored to each individual case in a learning process based on the available data.
General Data Protection Regulation (GDPR). European regulation on the use of personal data. It came into force in May 2018. It strengthens the guarantees provided by the previous directive.
Data controller. Person or organization that determines the purpose and manner in which the data is to be processed.
Transparency. Basic principle of the GDPR. For processing to be transparent, stakeholders need to receive information that is clear and uses understandable language.
Neural network. Machine learning model inspired by the functioning of the neural networks present in living organisms.