This article is sponsored by
Artificial intelligence (AI) and machine learning (ML) are two of the hottest technologies today due to their huge potential to derive new insights and value out of data. According to IDC, machine learning and artificial intelligence will produce 30 - 40 per cent in cost savings and productivity improvements in operations management by 2020.
These advantages have sparked both intense interest as well as concern amongst organisations, given that late adopters will struggle to remain competitive. Yet whilst organisations are eager to harness its potential, at present, this remains a somewhat futuristic concept for most. Few projects have been deployed at scale and hype has not helped either, spreading confusion about how well the technology works and what it can actually do.
To demystify these technologies and learn about the most innovative technology implementations today, I spoke with one of the leading experts in the field: Jason Loh, Analytics Practice Lead, SAS AP.
Image: Jason Loh, Analytics Practice Lead, SAS AP
Demystifying AI and ML
For non-experts, one of the first sources of confusion is often the various terminologies surrounding ML and AI. But it's not just the subtleties in their meaning that is confusing, explains Jason, often these terms are used interchangeably. "If you talk about the common academic definition of machine learning - it simply describes the task of having a machine, usually referring to computers, to perform learning, referring to extracting patterns from data that describe behaviour or events, to achieve a specific outcome. A common example would be offering product recommendations based on previous customer reviews"
Deep learning, another term you have likely heard mentioned, is a very specific class on ML. This refers to ML based on deep neural networks, which means, having many layers performing learning in multiple steps. Image recognition is often performed this way, by processing a hierarchy of features where each layer looks for more complicated objects. For example, when auto-tagging friends in photos, the first layer of a deep network might be trained to detect facial features like the relative position of the eyes, noses and cheekbones, with other layers recognizing hair and eye colour, as well as who are your closest friends and the people you tag most regularly.
Out of all terms related to ML, Jason is keen to emphasise one in particular: analytics. "The popularity of terms like AI and ML have risen and fallen over the years, but the word analytics brings an additional meaning, focusing on what is going on behind the data." He argues that whilst ML may perform a task and deliver a decision, users sometimes do not understand how the decision was arrived at, which can lead to problems for the organisation.
One of the big misunderstandings linked to this, is the idea that ML can magically derive insights from data, explains Jason. "Organisations often think they have loads of data so surely ML must be able to produce terrific insights, but it does not work like that. The first issue is whether a human could actually derive insights from the data even when given limitless resources, in other words, is the data computable or answerable in the first place?"
The second big misunderstanding is related to the concept of the problem definition, which Jason labels one of the toughest things to do well. "Organisations need to understand the constraints of their data," says Jason. "We do not have 'General AI' that offers right answers to every situation, so we must be precise when defining our aims. We also need to consider whether there are ways to collect better data or ask better questions."
Despite these misunderstanding, Asia is already home to many of the most innovative applications of AI and ML explains Jason. "In Asia, we get the coolest examples that are also extremely practical, which is the best combo you need; practical but innovative."
Asia's Most Innovative Applications of AI and ML
Cybersecurity: ML is already a common feature of many cybersecurity solutions, but what if you could apply it to images, actual screenshots of user interfaces. Logs can be analysed, but they can also be manipulated or falsified. Monitoring user activity through screenshot analysis, powered by ML, adds another innovative layer of security.
Industrial safety: How can you ensure that employees are wearing safety gear such as hardhats or not following physical safety policies? Or what if there is an unusual noise in an industrial area like a loud crack or a hissing noise? Whilst humans can monitor areas, the cost of sensors is dropping and these sensors can feed data, whether it's audio or images to identify anomalies or specific patterns, alerting employees accordingly.
Product development: Using smartphone development as an example, all the major companies are very interested in something called perceptual quality. This refers to the customers' perception of whether a product or particular feature is good or not. For example the CPU speed, product weight or feel compared to competitors. Even if the hard facts dictate that one product is objectively better than another, the subjective experience of the customer may lead them to think otherwise. Analysing comments online and applying natural language processing to identify sentiment can help identify aspects of a product to be tweaked or improved in future products.
Digital marketing: There's great interest in better understanding customer preferences and directing more relevant, targeted marketing towards them. We have all seen recommended products online, but with so much competition and noise, organisations need to make the most of every opportunity they get by reducing the amount of guesswork involved. This is a particular challenge when recommending new products that no one has purchased before, because there is no existing data about similar consumers; often referred to as a cold start. ML can help overcome this challenge by analysing larger datasets to help identify more detailed buying personas in far greater granularity.
Data for good: In Indonesia (and globally), SAS has been working with the United Nations Global Pulse on a project that aims to use data and ML for good. Using indicators on social media, SAS was able to identify economic fluctuations in the prices of essential items such as rice, chilli and petrol. SAS was also able to overlay broader economic data to identify which regions were likely to be most affected by price fluctuations, then advise local governments to support their communities in advance to reduce the negative impacts. Jason commented that, "When we do ML, we tend to spend a lot of time in front of the computer crunching data, but the best kind of work is when we can make a real difference to peoples' lives - truly makes you feel like a hero!"
How SAS Supports the Use of AI and ML
Whilst the possible applications of AI and ML are incredibly varied, the one thing they have in common is their reliance on data. "The task of getting data and making it useful for whatever purpose through ML or basic visualisations is a huge task," explains Jason. "Based on our experience working with global customers and hearing their challenges, the SAS Platform has been developed to excel in three key areas: data, discovery and deployment."
According to Jason, deployment is often the area that most organisations overlook. "After spending months building analytics and machine learning models, organisations often run into challenges deploying them. As a result, SAS allows customers to produce reliable and accurate models and if circumstances or data changes, organisations can ensure models are updated and governed well for maximum effectiveness in operations. SAS is not just a platform that provides good analysis or discovery, we help customers to confidently and reliably implement AI/ ML throughout the entire ML lifecycle."