What Is Machine Learning?

The introduction of Machine Learning has been a critical step forward in the development of Artifical Intelligence (AI) systems. As a result, it has received a lot of attention from researchers, hoping to test its full potential. Unfortunately, it has also become a popular topic for the press, which has resulted in some miscommunication and confusion as to what Machine Learning actually is. Today we hope to alleviate some of this confusion, as well as clarify some common misconceptions.

What is Machine Learning?

At its most basic level, Machine Learning is a concept where software is given the ability to mimic the human learning process. Getting slightly more technical, Machine Learning algorithms obtain input data and apply statistical analysis to predict the expected output. As new data becomes available, the predicted outputs will be updated.

Common Misconceptions

Machine Learning is a new development

Machine Learning algorithms have been around for a lot longer than many people realise.  Take AI pioneer Arthur Samuel for example.  Back in the early 1950s, Samuel was able to create a system that helped an IBM computer improve its ability to play checkers the more games it played.  So why the sudden rise in interest?  Faster hardware has provided the ability to process the large datasets required.  The availability of these large datasets has increased exponentially as a result of the drive for Big Data.  Time has allowed for the gradual improvement and refinement of some algorithms.  Code and algorithm library sets are now readily available for developers to utilise.

Machines learn autonomously

When talking about any topic related to Artifical Intelligence it is important to remember that computers do not understand what learning is. They are only as intelligent as their programming allows them to be, behind every AI system or Machine Learning algorithm is a cumbersome training process. A human programmer still has to design a learning architecture for a Machine Learning algorithm to follow, before it is presented with a large library of training data. If certain scenarios were not present within the training data the machine will not know how to deal with the situation.

Take the idea that an autonomous car could be trapped in a circle by simply following the rules of the road. As far as the cars AI is concerned, its training is telling it that when faced with markings, as outlined in the image below, the car should know it must not cross the line. Any human driver would know that this line is perfectly safe to cross, but to the AI it becomes an impenetrable barrier.

The algorithm is the most important aspect of Machine Learning

Algorithms are undeniably essential components in the development of Machine Learning software, however, this does not automatically make them the most important feature. In reality, the quality and quantity of the training data being supplied to the system are at least as important as the algorithm itself. Developing a powerful algorithm without access to high-quality training data would be like buying a car without access to fuel, a total waste of time. Of course, unlike the car analogy, a Machine Learning system will continue to improve in accuracy as more and more training data is fed into it.

Machine Learning is the beginning of the end for humanity

When you see coverage of AI in the media it is almost always portrayed as being superior to humanity, the Terminator franchise being a rather prominent example. Whilst understandably a result of the machines versus humanity who will win narrative, it is massively misrepresentative of reality.

You cannot compare the capabilities of a machine to those of a human. Machine Learning systems can process large chunks of complex data millions of times faster than what a human would be capable of, however, these systems would be rendered useless when faced with a crying baby. Humans have what is referred to as Natural Intelligence, which gives us the ability to face and deal with unseen life situations. We can decide on when, what and how to act according to the present situation, which is something no Machine Learning algorithm could achieve.

There is no doubt that machines are getting smarter with each passing day, but the human brain is such a complex tool that achieving any kind simulation of it would be a herculean task. So it seems that machines achieving human-level intelligence won’t be happening any time soon.

Zircon have steadily become more and more involved with systems that utilise Machine Learning, and as such have become well versed in the techniques that are available to developers. We know that having a clear definition of Machine Learning is certainly a good place to start, but it doesn’t provide much guidance on the potential approaches to developing a Machine Learning system. Therefore, in the next entry to this blog series, we will take a look at the different methods and techniques that are available and how they suit certain problem types.

Considering Machine Learning?

If you are looking to dip your toes into the ocean of Machine Learning, Zircon’s dedicated Machine Learning and AI team could be the resource you need.

More From The Blog

IR35, Here it Comes Again…

IR35, Here it Comes Again…

IR35, Here it Comes Again...In 2021 the reform to IR35 Off-Payroll rules is to be rolled out to the private sector. As before the reform will only affect companies that do not meet the following attributes: an annual turnover below £10m fewer than 50 employees or a...

Solving the Resource Conundrum

Solving the Resource Conundrum

Solving the Resource ConundrumPicture this. One minute all is fine and dandy, you have access to all the resources you could possibly need, then bam an unexpected challenge arises. Suddenly you find yourself lacking the capacity to meet the new need. What are your...

Quality – An Aid to Produce Consistent Rubbish

Quality – An Aid to Produce Consistent Rubbish

Quality - An Aid to Produce Consistent RubbishAnother year has passed, and myself and a colleague have hosted a BSI auditor for our annual ISO9001/TickITplus check-up, and in fact this was more than the regular check, in that it was our 3-year re-certification audit,...

The Hazards of Legacy Systems

The Hazards of Legacy Systems

The Hazards of Legacy SystemsBeing the owner of a software system with a dedicated customer base sounds like the kind of position one would like to find themselves in. At least until it gets superseded and you have to face dealing with a legacy system. Many developers...

How to Test Without Access to The Test Environment

How to Test Without Access to The Test Environment

How to Test Without Access to The Test EnvironmentIn many of our previous articles, we have expressed the importance of achieving a high standard of testing. Potentially blocking this achievement, several factors can come together to affect the quality of your...

The Technical Workshop – How To Make Them Work For You

The Technical Workshop – How To Make Them Work For You

The Technical Workshop - How To Make Them Work For YouAnyone experienced in product design will understand just how valuable a facilitated workshop can be. Bringing together a project's key stakeholders into a single space allows for the exploration of diverse...

Developing Software for Safety Related Systems

Developing Software for Safety Related Systems

Developing Software for Safety Related SystemsSoftware systems should always be both robust and reliable, however the moment you introduce a safety element, this need for reliability increases significantly. The level of safety required is governed by the severity and...

How to Choose an Outsourcing Partner

How to Choose an Outsourcing Partner

How to Choose an Outsourcing PartnerHaving recognised a need to outsource, and worked your way through the initial preparations, you are now in a strong position to seek out a suitable partner. Choosing an outsourcing partner is no trivial affair, so taking the time...