Machine learning can be described as a branch of artificial intelligence (AI) and computer science that focuses on the use of data and algorithms to imitate the way humans learn. The idea of machine learning is to use data to gradually improve its accuracy and eventually reach parity with human intelligence.
Arthur Samuel, a researcher at IBM is often credited for coining the term, “machine learning”, with his research around the game of checkers. While some researchers termed machine learning as a primitive form of AI, the wrr report released in November last year not only defines AI as a ‘system technology’ but goes on to elaborate machine learning (ML), computer vision, natural language processing (NLP), speech recognition and robotics as the five main branches of artificial intelligence.
Machine learning can even be termed as the most dominantly used branch of AI and an important component in the growing field of data science. Machine learning allows scientists and technologists to train algorithms using statistical methods with the goal of making classifications or predictions. The technology is also responsible for uncovering key insights within data mining projects.
For most consumers, machine learning is deeply embedded in the products and services they use every day. While ML was used to build machines that can play the game of checkers or chess initially, the technological development around processing power and storage has led to newfound innovation such as recommendation engines from the likes of Spotify and Netflix or self-driving cars made by the likes of Waymo.
ML: How does it work?
Machine learning, as mentioned earlier, is all about using available data to draw conclusions or inferences. The central idea of machine learning is to extract the data into meaningful use cases. While there are many ways to reach that conclusion, one of the widely recognised methodologies is the one described by UC Berkeley. This learning system of a machine learning algorithm is broken down into three main parts:
- A Decision Process: This first step is arguably the most important since it directly deals with how the data is used. Generally, the data available for a machine learning algorithm can be either labelled or unlabelled. With the purpose of using machine learning algorithms to make a prediction or classification, the algorithm will be used to produce an estimate or a pattern about the data. This process is also called the decision process in the industry.
- An Error Function: Once a pattern is determined from the data, an error function will serve the role of evaluating the prediction made by the model. An error function can make a comparison with known examples to determine the accuracy of the model.
- Model Optimisation Process: While the decision process and an error function allow a machine learning model to reach an inference or conclusion, it is also important to better optimise the model. A machine learning researcher will often evaluate how the model can fit better to the data points in the training set. The weights are then adjusted to reduce the discrepancy between the known examples and the model estimate. Once this is done, the algorithm will repeat the process of evaluation and optimisation, autonomously updating weights until a threshold of accuracy is met.
ML categories: What you need to know
Machine learning is an expansive subject in the field of data science but it can also be classified into three primary categories. This easy classification into supervised learning, unsupervised learning and semi-supervised learning makes it easier to understand machine learning as a concept. Here is what you need to know about these categories:
- Supervised learning: Supervised learning, also known as supervised machine learning, is defined by the use of labelled dataset to train algorithms used to classify data or predict outcomes accurately. With supervised learning, the model will adjust its weights as input data is fed until the model has been fitted appropriately. This approach helps organisations solve real-world problems at scale. Some of the popular supervised learning methods include neural networks, linear regression, support vector machine (SVM), and more.
- Unsupervised learning: Unsupervised learning, also known as unsupervised machine learning, is defined by its use of an unlabelled dataset. This type of learning uses machine learning algorithms to analyse and cluster unlabelled datasets. These algorithms discover hidden patterns or data groupings without the need for human intervention. This ability of unsupervised machine learning to discover hidden patterns, similarities and differences, makes it an ideal solution for exploratory data analysis, customer segmentation, image and pattern recognition.
- Semi-supervised learning: As the name gives away, semi-supervised machine learning serves as a middle ground between supervised and unsupervised learning. It uses a smaller labelled data set to guide classification and feature extraction from a larger, unlabelled data set. When data scientists lack enough labelled data set to train a supervised learning algorithm, they can solve the problem with a semi-supervised learning model.
Use cases in the real world
As mentioned earlier, machine learning forms the backbone of services like recommendations on Netflix or speech recognition. A consumer might knowingly or unknowingly use a service or product that relies on a machine learning model. Here are a few examples of ML that you might encounter every day.
- Speech Recognition: Speech Recognition, also known as automatic speech recognition (ASR), is a capability that uses natural language processing (NLP) to process human speech into a written text. The digital assistants found on smartphones like Google Assistant, Siri and Amazon Alexa, rely on speech recognition or speech-to-text to provide more accessibility around texting.
- Customer Service: During the pandemic, labour shortage and hospitalisations forced people to stay home, businesses turned to machine learning to replace those stricken employees. Online chatbots replaced human agents to help with the customer journey by answering frequently asked questions. Messaging bots are now common on e-commerce sites, apps such as Slack and Facebook Messenger, and can even employ virtual or voice assistants.
- Computer Vision: Computer Vision is a branch of AI that also relies on ML for its data set. This technology allows computers to derive meaningful information from digital images, videos and other visual inputs. Computer vision is used for actions like tagging photos on social media, adding metadata in the automotive industry and for labelling radiology images in healthcare.
- Recommendation engines: The most popular use case of machine learning can be found in recommendation engines. These AI algorithms use past consumption behaviour to discover data trends and offer relevant add-on recommendations to customers. In the case of online retailers, this will be done at the checkout process while platforms like Netflix will do the same at the end of a binge session.
Machine learning has made our lives easier as technology advances and is used by the likes of technology companies to improve their products and services. However, the implementation of machine learning also raises a number of ethical concerns and remains one of the major challenges for business.
- Singularity: Singularity, also known as Technological Singularity, is an event where the AI surpasses human intelligence in the near or immediate future. Also referred to as superintelligence, it is defined as any intellect that can surpass the best human brains in every field. While superintelligence is not imminent, it raises a number of questions as to how humans will react when the machine surpasses their intelligence. Some believe that humans should never let machines become superior in terms of intelligence.
- Impact on Jobs: One of the immediate challenges for businesses is to convince their own employees, consumers and their partners that AI is here to supplement them and not replace them. While every new disruptive technology is known to eliminate some specific roles, the impact of AI could be far-reaching. AI could eventually write this same article one day and thus eliminate the role of a technology writer. The only way to balance this situation is to help individuals to upskill themselves and transition to new areas of market demand.
- Privacy: Another challenge with machine learning and artificial intelligence is in the context of data privacy, data protection and data security. While more data is better as the approach followed by tech companies often at the cost of user privacy, there are new algorithms and models that can learn from less but structured data to reach a similar conclusion. Legislation like GDPR can help protect the privacy and personal data of people as technology sweeps into every walk of life.
- Bias and Discrimination: This is the biggest challenge facing technology companies like Microsoft and Google that rely extensively on AI and ML right now. We have already seen instances of bias and discrimination across a number of intelligent systems. This has raised many ethical questions and the challenge here is how to safeguard against bias and discrimination when the training data itself can lend to bias.
- Accountability: There is still no meaningful legislation to regulate AI practices. As a result, there is no real enforcement mechanism available to ensure that AI is practised in an ethical manner. While countries like the US, the UK and European Union are debating around the best way to regulate AI, these legislations could result in both accountability and elimination of bias.