Automated decision-making (ADM) definition
Automated decision-making (ADM) is the process of making decisions using algorithms, artificial intelligence, machine learning, or rule-based systems without human involvement. ADM aims to simplify decision-making by handling very large volumes of data and processing thousands of decisions in milliseconds
Automated decision-making is based on predefined rules and models, ensuring decisions are made by automated systems without any human intervention. The systems generate relevant outputs according to the specific input data provided, ensuring all decisions are made based on the same criteria without deviation. Due to its efficiency, ADM can help reduce manual labor costs.
See also: hyperautomation, machine learning, supervised machine learning, unsupervised machine learning
Types of automated decision-making
Rule-based systems
Rule-based systems use standard if-then statements for decision-making. These systems are developed by domain experts who encode explicit logical rules. The outcomes are deterministic and transparent.
Machine learning systems
Machine learning systems, or machine learning models, are trained using data patterns to make predictions or classifications. After processing historical data, they make predictions based on their learned patterns. Machine learning systems can be supervised, unsupervised, or use reinforcement learning.
Hybrid systems
Hybrid systems combine multiple decision-making systems to balance out each system’s strengths and weaknesses. Most commonly, hybrid systems take parts of rule-based systems and machine learning systems to make more sophisticated decisions.
How can automated decision-making be used?
- Job recruitments and candidate filtering.
- Insurance and credit assessments.
- Social media content moderation.
- Dynamic pricing management.
- Financial fraud detection.
- Ad and content recommendations personalization.
What are the risks of automated decision-making?
- Bias and discrimination. Using historical training data, automated systems can perpetuate recorded biases.
- Negative feedback loop. If the data used for training contains errors, the systems may replicate these errors in their decisions.
- Accountability. Due to a lack of human intervention in the decision-making, it can be difficult to determine who holds accountability for the outcomes.
