Artificial Intelligence and Machine Learning – Are we Missing the Human Point? [Part 1]

Reading Time: 4 minutes

Part one in our latest blog series explores the difference between Artificial Intelligence and Machine Learning, how it can deliver value and some practical considerations around implementation.

There has been a dramatic increase in awareness around Artificial Intelligence and Machine Learning – which appears to be the technology buzzword of the moment. While there are some good opportunities for Machine Learning and Artificial Intelligence, there are a number of challenges particularly from a data security perspective and ethics. In this new blog series, we aim to explore the differences between Machine Learning and Artificial Intelligence, demystify some of the common myths associated with these approaches and outline some of the critical points that organisations need to consider before jumping in headfirst – all within an information security context.

  1.  Isn’t Machine Learning the same as Artificial Intelligence? 

No they are different, although the terms are commonly confused and often used interchangeably. To quote Forbes:

Artificial Intelligence (AI) refers to devices that are designed to act intelligently, however AI is often classified into one of two fundamental groups – applied or general. Applied AI is far more common and refers to systems designed to intelligently trade stocks and shares, or manoeuvre an autonomous vehicle for example. Generalised AI refers to systems or devices which can in theory handle any task, and these are much less common.  This is the area that has led to the development of Machine Learning (ML), which is often referred to as a subset of AI.

Machine Learning’s key differentiator is that the device learns how to do a task, rather than is programmed to complete the task, which requires training. A common example of this is when a ML system is used to detect brain tumours in MRI scans. They were shown 1000s of images of brains with and without tumours, and throughout were told if a tumour was present. After the learning phase, the system could easily identify whether a brain had a tumour. This example shows that ML is very good at complex image tasks so long as there is a relatively simple answer. 

  1.  What is the track record of Machine Learning and Artificial Intelligence in delivering value?

It’s too early to tell. Although Machine Learning has been transformative in some fields, effective and accurate Machine Learning is challenging because often there is just not enough data available with clear results; as a result, many Machine Learning programs often fail to deliver their expected value.

Today there is much hype around AI and ML, and as a result, business Executives are generally receptive. On the other hand, some people’s expectations of what Machine Learning can do in practice can far exceed what is possible or even reasonable.

Today ML and AI is being employed in a handful of industries to support low complexity, high volume processes, but ultimately the human often has to be involved to validate any decision. For example, in the healthcare example given previously, they are using ML to detect tumours but still rely on a human radiographer to validate the results before surgery.

If you speak to your digital assistant at home, ML is used to convert your speech to text, after which a basic AI engine is used to pick out key words to respond to you. It is significantly more challenging to apply ML in a text-based environment due to complexities with language and syntax, so the need to involve a human becomes even more critical to ensure accuracy of decisions.

  1.  Is Machine Learning easy to implement?

Not really. There are two parts to any ML project, firstly you may need a very powerful computer with very specialised software. Secondly, you need a team experienced in finding the correct data for the ML engine to learn with. This ongoing problem contributes to a backlog of Machine Learning inside the enterprise. In fact, there is at least a ten-year backlog of Machine Learning projects locked inside large companies.

Data scientists often need a combination of practical applications skills, as well as in-depth knowledge of science, technology, and mathematics. Recruiting them will require you to pay big bucks as these employees are often in high-demand and know their worth. Given the recent emergence of ML and AI, service providers have staffing shortages and those who do have skilled data specialists to support a deployment will charge a high premium.

  1.  Does Machine Learning require organisations to have large infrastructures?

Yes. Machine Learning can require vast amounts of data processing capabilities. This is additive to the existing systems processing data across the environment, which will be already drawing on processing resources. Legacy systems often can’t handle the workload and buckle under pressure.

Often companies rent servers from suppliers of data farms. While people might be happy to send a text from their home assistant, companies are still reluctant to send vast amounts of confidential data outside of their organisation.

Boldon James is actively researching AI and ML techniques to identify areas where Machine Leaning can help in data classification. For more information please contact us.

Later in the series, we will talk about AI and ML for security and data classification…