The basic procedure of machine learning is to give preparation data to a learning algorithm. The learning algorithm then generates a new set of rules, based on inferences from the data. This is in spirit generate a new algorithm, formally referred to as the machine learning model. By using different training data, the same learning algorithm could be used to generate different models. For example, the same type of learning algorithm could be used to teach the computer how to translate languages or predict the stock market.
Inferring new instructions from data is the core strength of machine learning. It also highlights the critical role of data: the more data available to train the algorithm, the more it learns. In fact, many recent advances in AI have not been due to radical innovations in learning algorithms, but rather by the enormous amount of data enabled by the Internet.
How machines learn:
Although a machine learning model may apply a mix of different techniques, the methods for learning can typically be categorized as three general types:
- Supervised learning: The learning algorithm is given labeled data and the desired output. For example, pictures of dogs labeled “dog” will help the algorithm identify the rules to classify pictures of dogs.
- Unsupervised learning: The data given to the learning algorithm is unlabeled, and the algorithm is asked to identify patterns in the input data. For example, the recommendation system of an e-commerce website where the learning algorithm discovers similar items often bought together.
- Reinforcement learning: The algorithm interacts with a dynamic environment that provides feedback in terms of rewards and punishments. For example, self-driving cars being rewarded to stay on the road.
Machine learning is not new. Many of the learning algorithms that spurred new interest in the field, such as neural networks, are based on decades-old research. The current growth in AI and machine learning is tied to developments in three important areas:
- Data availability: Just over 3 billion people are online with an estimated 17 billion connected devices or sensors. That generates a large amount of data which, combined with decreasing costs of data storage, is easily available for use. Machine learning can use this as training data for learning algorithms, developing new rules to perform increasingly complex tasks.
- Computing power: Powerful computers and the ability to connect remote processing power through the Internet make it possible for machine-learning techniques that process enormous amounts of data.
- Algorithmic innovation: New machine learning techniques, specifically in layered neural networks – also known as “deep learning” – have inspired new services, but is also spurring investments and research in other parts of the field.
As machine learning algorithms are used in more and more products and services, there are some serious factors must be considered when addressing AI, particularly in the context of people’s trust in the Internet:</p?
- Socio-economic impacts: The new functions and services of AI are expected to have significant socio-economic impacts. The ability of machines to exhibit advanced cognitive skills to process natural language, to learn, to plan and to perceive, makes it possible for new tasks to be performed by intelligent systems, sometimes with more success than humans. New applications of AI could open up exciting opportunities for more effective medical care, safer industries and services, and boost productivity on a massive scale.
- Transparency, bias, and accountability: AI-made decisions can have serious impacts in people’s lives. AI may discriminate against some individuals or make errors due to biased training data. How a decision is made by AI is often hard to understand, making problems of bias harder to solve and ensuring accountability much more difficult.
- New uses for data: Machine learning algorithms have proved efficient in analyzing and identifying patterns in large amounts of data, commonly referred to as “Big Data”. Big Data is used to train learning algorithms to increase their performance. This generates an increasing demand for data, encouraging data collection and raising risks of oversharing of information at the expense of user privacy.
- Security and safety: Advancements in AI and its use will also create new security and safety challenges. These include unpredictable and harmful behavior of the AI agent, but also adversarial learning by malicious actors.
- Ethics: AI may make choices that could be deemed unethical, yet also be a logical outcome of the algorithm, emphasizing the importance to build in ethical considerations into AI systems and algorithms.
- New ecosystems: Like the impact of mobile Internet, AI makes new applications, services, and new means of interacting with the network possible. For example, through speech and smart agents, which may create new challenges to how open or accessible the Internet.