Although Machine learning (ML) is often described as a branch or subset of artificial intelligence (AI), ML is an application of AI. Without specific programming, algorithms and historical data are applied to allow machines to imitate human learning and make predictions based on analysis. The greater the input, quality, and diversity of data, the more accurate the outcome. There is not one specific technology used in ML. Python is currently the primary language used in developing ML programs and there are numerous platforms available, including Microsoft Azure ML Studio, Amazon SageMaker, and popular open-source options such as KNIME Analytics Platform.
Some of the most prevalent algorithms used in ML include linear regression, logistic regression, and decision tree algorithms. There are three main types of ML algorithms, with supervised learning the most commonly used. Supervised learning feeds historical output data, or results, into ML algorithms to provide an understanding of the required future output. Supervised ML is commonly used for practical applications, including data classification. Other uses for supervised learning include fraud detection, sales predictions, and loan application assessments.
Unsupervised learning does not provide an example of the required output and is used when the end result is not necessarily known. Unsupervised learning can be used for anomaly detection, data visualisation and for sorting and labelling data that is not otherwise already associated. A hybrid type of ML known as semi-supervised learning may also be used, with some labelled data provided along with a large amount of unlabeled data. Lastly, reinforcement learning works through providing positive and negative responses based on outputs. Reinforcement learning is essentially a trial-and-error based approach. ML is present but not always obvious in our lives. Voice activated virtual assistants like Alexa, Cortana and Siri are everyday examples of ML. One billion voice searches are made each month, with 20% of all internet searches now voice based (50+ Voice Search Stats Everyone Needs to Know in 2022, 2022). Most smartphones are pre-installed with a virtual assistant allowing us to use voice in order to find information over the internet. Many homes are also equipped with smart speakers, allowing household members – and guests to some degree – to conduct voice searches as well as direct home automation tasks, such as lighting and climate control.
ML is also used in a number of social media applications, using location services and search history, for example, in targeting advertising and service offerings to the user. It is also used for spam detection and for facial recognition of images. Other uses include online maps, which use ML to estimate trip duration and to determine the fastest route when guidance is requested. Educational institutions also employ ML to detect potential plagiarism. In the future, it is expected that ML will drastically improve healthcare services. From analysis on waiting room times through to the use of algorithms to predict likelihood of heart failure, ML is also increasingly being used for imaging and diagnostics through the use of Convolutional Neural Networks, a type of deep learning. A deep neural network has already been used to diagnose some neurological conditions 150 times faster than human specialists (Health IT Analytics, 2018). Deep learning is a subfield of ML based on artificial neural networks, imitating the way humans learn. Deep learning is expected to provide advanced tools, including cloud-based development environments, that will allow users access to libraries of pre-built algorithms.
With many employers moving towards the automation of manual processes and an increase in ML technology use, there is a shortage of skilled employees who can drive and work alongside ML technology. The World Economic Forum, The Future of Jobs Report 2020 discusses the acceleration of technological advancement owing to COVID-19 lockdowns and the 2020 global recession. The report discusses the need to reskill and upskill workers, and highlights predicted surges in employment types as well as estimates for industries most likely to be affected by unemployment. The report estimates 85 million jobs will be replaced by AI machines by 2025.
Art, entertainment, and recreation as well as accommodation and food services roles (hospitality) are the most prominent industries already affected by unemployment as well as the most at risk in future. The use of ML, particularly in the hospitality industry is largely portrayed as a positive, with cost savings, enhanced customer experience and reduction in human error highlighted as the major benefits. ML in these industries is increasing, with typically human roles being replaced. Transport and accommodation bookings, as well as concierge services can already be accessed through chatbots, with predictive analytics being used to deploy targeted advertisements to consumers. Computers are available around the clock and are not frustrated with repetitive tasks, but they cannot replace creative thinking and social connection.
Machine Learning will continue to affect and enhance our daily lives even if we are unaware of it. Smart homes are becoming increasingly popular allowing for increased energy monitoring and optimisation, automation and remote control of lighting and appliances. In educating our children, ML is being used through e-learning and allows students to sit exams remotely. ML is also being used in adaptive learning, modifying educational content offered based on each student’s performance. Interactive toys and games are increasingly being released including Minecraft and Hello Barbie. Our elderly population now have access to wearable sensors that monitor health and detect falls or heart events and will contact emergency services on their behalf. ML is also being used to research aging itself and to develop potential medications to slow the process and related diseases (De Winter 2021).
Machine Learning is not perfect as it performs based on the quality of the data used along with probability. Along with all of the positives, there are negatives. A degradation of privacy is likely to continue. Businesses are applying ML for customer identification, offering product suggestions based on customer preferences, and collecting more and more customer data in order to do it. Facial recognition, use of location services and examination of our expenditure is already occurring, and our legislation is not keeping up with the rapidly changing application of ML technologies (Dingwall, 2021). There are risks associated with ML, as we allow machines to make decisions on our behalf. This is often discussed with reference to self-driving vehicles and the consequences of decisions made by vehicles if they are confronted with various accident scenarios.
Issues with a lack of training data, lack of quality and diversity of data used has an impact, with an example being facial recognition software and its inability to identify people of colour. However, ML is adaptive, and the potential risks can and should be mitigated.
De Winter G (2021) Machine Learning Identifies Anti-Aging Drugs, Medium website, accessed 09 July 2022. https://medium.com/predict/machine-learning-identifies-anti-aging-drugs-7c6560f9e0c8
Dingwall B (2021) Laws to Consider When Implementing Machine Learning Algorithms in Your Business. Legalvision website, accessed 09 July 2022. https://legalvision.com.au/laws-implementing-machine-learning-algorithms/
Health IT Analytics (2018) What Is Deep Learning and How Will It Change Healthcare? Health IT Analytics website, accessed 09 July 2022. https://healthitanalytics.com/features/what-is-deep-learning-and-how-will-it-change-healthcare
Petrov C (2022) 50+ Voice Search Stats to Help You Rethink Your Strategy in 2022, techjury website, accessed 04 July 2022. https://techjury.net/blog/voice-search-stats/
Created using original code and creative design by team Summit