What Machine Learning Is and Isn’t
Last week in my post, It’s not the size of your data that matters, it’s how you use it, I introduced the idea that there are some common misconceptions around big data and artificial intelligence. This misinformation seems to be magnified when it comes to the way these concepts can be utilised in the commercial space.
The lions share of the responsibility for this confusion has rest with companies offering big data or artificial intelligence solutions. Far too often the words ‘big data’ and ‘artificial intelligence’ are woven into elaborately worded marketing documents to increase the appeal of a particular product or service offering with little thought given to whether or not the solution is indeed utilising those processes.
On cognitiveprocurement.com our quest is to demystify this space and provide clear definitions and pragmatic use cases of where big data, artificial intelligence and cognitive computing have been put into practice and to discuss how these concepts will affect the procurement function, both now and into the future.
Buzzword Bingo
Today I want to address ‘machine learning’ another term that has sadly been passed through the marketing ringer.
Play a word association game with the term ‘machine learning’ and you’ll soon get someone talking about robots taking over the world. If the robots can learn, then it follows that they can get smarter than us and start to out do us. But this is taking machine learning, at least in the way data scientists refer to the term, a few steps too far.
The most pragmatic way to think about machine learning is to consider it as a part of the broader artificial intelligence process. More specifically, machine learning is the statistical arm of AI. It involves creating software that can learn from its past experiences. Machine learning is in fact more akin to statistics that it is to the image of AI you may have become familiar with in marketing brochures. Machine learning does not attempt any of the more ‘human’ elements of AI such as knowledge representation, reasoning or abstract thinking.
The genesis of AI
Our earliest attempts at AI stemmed from our desire to automate the decision making process. This essentially fell back on if/then scenario planning. If the stock prices drops to here then buy more stock. We named these ‘expert systems’.
Expert systems represented the first step in machines mimicking human decision making, but the process was too one dimensional to be classified as cognitive processing. Expert systems could not look at external data and were severely limited in their capacity to scale up. In essence, they could only process the code that was directly loaded into them.
Machine learning represents the next step in the evolution of the artificial intelligence movement. Moving beyond if/then scenario planning, machine learning focusses on producing algorithms that, through processing huge amounts of data, can learn (if a computer improve its performance of a task based on information it has gained from a past experience, we can say it has learned). This learning can then be applied to complete simple tasks.
It’s important to note however that the outputs of machine learning require a high level of human analysis and interaction before they are useful. Machine learning can indeed provide you with interesting data that corresponds to a recent drop in productivity at your biscuit factory, but you’ll need someone who understands the numbers and knows where to look to explain it all to you if you hope to get any value out this analysis.
Machine learning is not capable of making decisions or solving complex business problems on its own. It is entirely reliant on humans interpreting the results the algorithms return. Perhaps even more critical than the time spent analysing the data that gets spat out, is the time required to prepare the data that gets feed into the process. As we dealt with in It’s not the size of your data that matters, it’s how you use it : Garbage data in will equal garbage data out.
In order to understand the full breadth of artificial intelligence and the capability of computers to interpret data and make decisions we need to introduce the topic of ‘machine intelligence’ which is exactly what we’ll be doing next week.
