What Are The 4 Types Of Artificial Intelligence?
There are four main types of artificial intelligence (AI), which is the flexible intelligent behaviour exhibited by machines or software. Each of these types is used in specific applications, but all are aimed at creating something that simulates human thought.
Background information: The key factors behind AI include machine learning, deep learning and natural language processing (NLP). These technologies allow machines to learn over time without needing explicit programming, taking in data on new situations and reacting accordingly.
Supervised Learning
Supervised learning is a type of machine learning where you have input variables for training purposes, along with their output values — also known as “labels”. Using this data set, the model finds patterns between the input variables and output labels, so it can predict future values of unknown instances.
Many people have used supervised learning before, usually without realizing it.
For example, if you’ve ever let your Facebook newsfeed algorithm guess which posts you will probably want to see or posted a tagged photo on Instagram that includes both yourself and another person, then you’ve made use of supervised learning.
In these cases, the AI has been trained by humans to recognize certain patterns in data (posts or images), and then use that knowledge to make guesses about other information (which posts/photos you’re likely to find interesting).
Unsupervised Learning
Unsupervised learning is when there are no output variables and the model finds hidden trends within the input data.
This type of machine learning is often used in market segmentation and identifying groups of users based on their behaviour. This works by considering clusters of data, with similar information grouped together.
Another good example is how Netflix can successfully recommend shows and films to you based on your previous selections — it has been using unsupervised learning for years now.
Semi-Supervised Learning
In semi-supervised learning, the model can access both labelled and unlabeled input data sets. It then trains itself to automate finding patterns between them that most accurately predict the output labels. The more input data available, the better its predictive power.
Sometimes companies don’t want to invest too much time and money into labelling all their training data, but still want to make use of machine learning. Semi-supervised learning allows them to get the best results without needing too much-labelled data.
Reinforcement Learning
Reinforcement learning is when an agent (a model) gets feedback as it goes about its task, and learns from that feedback. This type of AI can be applied to things like video games or robotics, where the agent has a goal and then tries different actions within its environment till it reaches that goal.
Learned behaviour: There are two main types of reinforcement learning: policy and value function-based.
Policy-based reinforcement learning is where an algorithm determines which action to take in a given situation by estimating which action is most likely to bring about a good result. That’s based on the agent’s current policy, which is a state-action mapping that can be updated with each action taken.
In contrast, value function reinforcement learning adjusts the agent’s behaviour in accordance to how valuable it deems certain actions within its environment. It does this by estimating expected future rewards when certain actions are performed in a given state.
With reinforcement learning, you don’t have any labelled data for training purposes, but that doesn’t mean it doesn’t need input information — just without predefined target values to train from. Rather, you’re providing direct feedback from your environment to let it know what works and what doesn’t. Once agents have been taught using reinforcement learning, they can then apply this knowledge elsewhere in their environment.
By combining these 4 types of machine learning together, one AI agent could learn from its own experience and feedback to automate a task within certain constraints.
For example, you could have an algorithm complete a job application form if you tell it what tasks need to be completed and the format in which they must be done. Based on past feedback from employers, they would learn how to write a resume or cover letter that is most likely to impress them based on the information given.
In summary: Supervised learning is when data sets with labelled output values are fed into algorithms so that predictions can be made about future unknown instances by finding common patterns between them Unsupervised learning is when data sets are not labelled, but the algorithm is still able to find hidden trends within them Semi-supervised learning utilizes both labelled and unlabeled datasets Reinforcement learning is when an agent learns by getting feedback from its environment in order to optimize behaviour based on limited information so that it can achieve goals with increasing accuracy.
