Organisations currently using AI at scale are performing an average of 11.5% better than those who are not, according to Microsoft's recent whitepaper Accelerating competitive advantage with AI. This is has increased from 5% just one year ago.
The organisation's research also found that in the UK alone there has been an 11% rise in the number of companies using machine learning within the past year.
What is Machine Learning?
Machine learning is a subset of artificial intelligence and enables systems to learn how to identify patterns and perform specific tasks - such as image recognition or natural language processing - without being programmed.
It enables individuals and organisations to automate routine tasks or process huge volumes of data.
There are two types of machine learning tasks:
Initially the programme will need to be "taught" by submitting sets of example data. For example, in the case of image recognition, an individual will need to highlight images of interest that they would like the system to learn to detect. Once enough data has been submitted, the system will start to make decisions for itself.
The programme is not given any example data and must discover patterns and structure for itself, without any set outcomes. This is particularly helpful for discovering patterns in large volumes of data where there isn’t a desired outcome, such as analysing patterns in user behaviour.
Applications of machine learning
One of the best-known applications of machine learning is in fraud detection. An organisation that represents medical professionals has been using the technology to help them detect fraudulent certifications. A machine learning programme was “taught” to detect and classify stamps on certifications and help to spot fakes.
Machine learning can be used to tailor online and consumer experiences to the individual, providing them with the most relevant experience. Many retailers use this technology to analyse customer purchase history and suggest relevant products.
Organisations are increasingly using machine learning to test for vulnerabilities in their networks and identify potential instances of malware within their systems. Earlier this year Microsoft used machine learning to thwart an attack aimed at the satellite and communications industry.
Businesses that use physical processes – such as factories or food processing plants – can use machine learning to improve effectiveness. For example, machines that cut food into slices could automatically adjust themselves according to the weight of portions produced reducing wastage. The programme could also be used to alert engineers about possible mechanical issues.
Machine learning programmes can be used to analyse data to make predictions of likely outcomes. For example, a bank may analyse loan data to determine which individuals are most likely to default.
What to consider before using machine learning
Have a plan in mind
The first step to take when engaging in machine learning is to identify a problem that you would like to solve, such as detecting fraudulent accounts.
Similarly, you should also identify a measure of what success will look like, such as a specific cost saving, a reduction in time taken to process data or a higher success rate.
Ensure there is enough data
Computers may be able to perform tasks faster and more efficiently than humans, but they still can't think better than people can. You need to ensure you have enough data - if you cannot make a decision yourself based on the information you have, a machine learning programme will not be able to do it for you.
Check you have the right data
Finally, you'll need to ensure that your data is in a format that can be inputted into the machine learning programme. As Microsoft said in Accelerating competitive advantage with AI: "This need for companies to get their data house in order is true across all sectors, with experts from the fields of finance, healthcare, retail and manufacturing united in seeing it as a critical component of any AI scaling plan".
You should also check to see if there is any bias that the system could learn from it - for example, a system used to screen potential job applications could learn to screen out females if it learns that successful applications tended to have more years of experience.
Article by James Boother. James is Sales and Marketing Director at Coeo, a Microsoft Gold Partner providing consulting and managed services for Microsoft data platform and analytics technologies. He has extensive experience within the software industry, and before joining Coeo had 15 years’ experience working as a programmer, system architect, head of technology and consultant. James regularly presents at industry and community events.”
Find out more