Machine Learning is Not the Future- It's Here!
As with every New Year, the annual tradition of predicting that year’s technology trends is as eagerly anticipated as the exchanging of Christmas gifts. Many will predict, unsurprisingly, that data science and machine learning will be a bigger part of many IT departments’ plans. Those CIOs that are yet to adopt such technologies will need to start soon, and those that have experimented with them need to adopt them.
Automated behavioural analysis can detect and track suspicious activity on networks, spotting ransomware before it has a chance to encrypt mission-critical data
What these predictions are likely to lack, however, is exactly how this should be done. Machine learning and AI are now the buzzwords to drop, much like “cloud” and “big data” have been in the past, to the point that they become almost meaningless and derided. What CIOs need is a better understanding on where they should focus their efforts and where data science can help now, rather than vague predictions of the impact it may have at some point in the future. We know that more than half of CIOs are prioritizing analytics in their budgets—this money needs to be spent effectively.
Security and Machine Learning
No industry is immune to cybercrime—the ransomware attacks on hospitals in the US earlier last year show that these criminals are willing to put anyone at risk in order to achieve their aim. And according to the Verizon DBIR Report, 89 percent of breaches had a financial or espionage motive. The mischief makers and hobbyist hackers are long gone—the overwhelming majority of breaches and attacks are looking to steal identities and money. Cybercrime is a big business and—like any business—is always looking to increase its revenue and seeking for new opportunities.
However, machine learning and cybersecurity tools are now working together and providing businesses with the best proactive, detective, and reactive security possible. Much like the precogs in Philip K Dick’s The Minority Report—albeit without the terrible ethical issues—machine learning means the ability to identify threats before they happen, meaning a change of stance from reactive to proactive.
Machine learning tools can help to fill the hiring gap that currently exists—fewer data analysts are needed if algorithms are doing some of the work for you. It can also work to defend against the most common attack—ransomware. Automated behavioural analysis can detect and track suspicious activity on networks, spotting ransomware before it has a chance to encrypt mission-critical data and start demanding bitcoin to decrypt it. The other big selling point for machine learning is speed. By bringing together data from an organisation’s clients, vendors’ customers and more, it is possible to predict and react more quickly to anomalous behaviour and conditions.
Profitability and Machine Learning
Good use of data is the key to profitability. Often, however, the tools used are too blunt and take too long to produce results. A quarterly profit and loss report may tell you that you are three months into problems that desperately need to be fixed. Measuring profitability is more than tracking your bottom line.
In order to understand profitability, it’s important to understand the details—how profitable is each client and each project? How ‘billable’ each member of staff is? No one, however conscientious, is 100 percent billable, and changing demands on staff means that this number of billable hours available from staff, or ‘billability’, changes every week.
This granular level of information is vital and relatively straightforward for a business that employs just a handful of people. But becomes increasingly difficult to scale as a company grows—not just bringing the data together, but analyzing it and creating actionable decisions based on the data is quickly impossible. Machine learning can help by not only analyzing the data and helping with these decisions, but changing the decisions based on previous outcomes.
Not just Big Data—Overwhelming Data
The sheer volume of data that an IT estate can produce means that automation is necessary to make sense of it at all. CPU and memory data can identify underperforming computers. IP addresses can be used to make location-based recommendations for scheduling tasks. Failed logins and web traffic are vital to detecting security breaches and security risks. Patching data can pinpoint where vulnerabilities exist to prioritise where action is needed. There are many ways that an IT department can use data if they have the right way to analyse it.
Big data means just that. A single server can generate over fifty types of raw data records every five minutes, from monitoring the CPU, memory, network, security, and more. This single server will create over 400,000 data records in a month. A thousand servers can therefore create over 400 million data records per month. Meanwhile, a typical device can have over 200 connections to ten different websites per hour. So for a thousand devices, that translates to 144,000,000 connections per month. Detecting patterns that could mean risk to the company in that data is no small task.
“Don’t roll your own crypto,” is a mantra for IT, thanks to the things that could go wrong and the ease that vulnerabilities could be introduced. Similarly, CIOs, while keen to get the benefits of machine learning as quickly as possible, should consider how much of their limited resources this data would take to analyze and create meaningful, actionable decisions. They are right to be allocating budget, but the scale of the task is only going to increase into the future. With pressure on budgets, the decision should be where this need should be outsourced to, rather than whether to outsource it or not.