How Real is Artificial Intelligence?
Today we hear of artificial intelligence, or AI, being implemented everywhere – from smart phones through to domestic appliances through to robotic devices. Businesses claim to be based on it. But given the way that artificial intelligence has changed over the years it is valid to ask what it is today, and indeed, is it just marketing hype?
The term AI was coined back in the 1950’s, with the idea of making intelligent computer systems that “think” and make decisions like a person would. Typical of that time was the Turing test, with the aim to have a program that could converse in a manner indistinguishable from a human. Many of these aims have still not been met, but they have led to ideas that are in use today. Machine Learning (ML) is a simpler form of AI that “learns” as it identifies new patterns in data, with both terms becoming somewhat interchangeable. This drives the modern form of AI, made practical with modern computing and the internet. It is a far cry from approximating intelligent human beings.
It matters little whether this is called artificial intelligence, deep learning, machine learning, statistics or data science. What is important is that the richness of data now available can lead to better decisions, more efficient businesses and ultimately a better society. This requires using the most appropriate tools combined with a professional approach, regardless of what names they go by.
Not unlike their human counterparts, decisions made by AI systems are not always sensible. The current generation of point-and-click AI tools such as Google’s TensorFlow can produce a great fit for your data, but is it meaningful? AI systems lack “common sense”, both statistically and otherwise, and can be easily confused by unexpected input data, causing unpredictable results.
AI tools are powerful and have their uses, but meaningful results are not just a button push away.
Data Analysis Australia has been using AI / ML techniques for years for the benefit of our many clients. Regression methods, the workhorse of statistics and a core analytical technique used in most ML platforms, are used to predict the outcomes of an event based on the relationships between variables in a dataset. When the input variables change frequently, statistical rules can be set up to automatically re-run the regression models and provide automated recommendations as to whether the process is 'in control' or whether something has changed and should be investigated. The key to getting this right lies in the rules the statistician sets up, and ongoing monitoring and maintenance is critical.
Automation is also useful for clustering and classification into groups using inherent structures in data, but again the statistician or analyst needs to build in the statistical rules to optimise the "AI" outputs. For example, are 3 clusters or groups sufficient, or is it 10? The right answer will be different every time, and even then, meaningful interpretation of the clusters or groups requires human intervention. Our team combines expert judgement with such clustering algorithms to group customers into sensible different personas based on demographics, values or behaviours. Data driven approaches can uncover new classes of customers or groups and refine existing (and potentially incorrect or outdated) industry “norms”.
Machine learning can also be very effective with images, with techniques to automatically tag and extract features. At Data Analysis Australia, we’re combining automatic classification of satellite images with our forecasting know-how to model and better understand how cloud cover impacts on solar panel efficiency on roof tops and the subsequent impacts on the energy grid.
Much of modern AI involves understanding data – sometimes simple, sometimes complex. While AI/ML has its limitations, we provide the conduit between the smart algorithms and your data, and the human touch needed to draw the relevant and valid insights for the problem of interest.
At Data Analysis Australia we do artificial intelligence with human understanding.
October 2018