ARTIFICIAL INTELLIGENCE has been defined as “a constellation of technologies that extend human capabilities by sensing, comprehending, acting and learning—allowing people to do much more”, but I find it helps if you break the actual words down a bit further:
- Artificial: made or produced by human beings rather than occurring naturally, especially as a copy of something natural.
- Intelligence: the ability to learn, understand, and make judgements or have opinions that are based on reason, data or experience.
Now you can quickly see that every AI-enabled machine, robot, or computer system has a de-facto human parent and that it must be trained by data or experience.
If you have young children, this is easy to understand. Imagine asking your 3- or 4-year old child to drive a car. They have no experience, therefore they would do it very poorly. But, if you ask them to recognise animals or objects (“cats”, “broccoli”), name people (“mummy”, “daddy”) or even to respond to a question (“what color is your pencil?”), then these are all things that they can do, with a little practice. The same is true of machines and artificial intelligence; they are informed by experience and by training data.
But as soon as we imply that “AI” is simple to understand, others will say it is incredibly complex. And they would be right. That is because the field of artificial intelligence focuses on understanding core human abilities and designing machines and software to emulate our vision, speech, language, decision making, and other complex tasks. As humans, we are also complex beings. We demonstrate much variety in our actions and responses for machines to learn. The goal for machines is to take on human intelligence, including all the variety.
That said, it is fair to say that AI today is best applied to ‘narrow’ tasks that may be hard for humans to do but easy for machines to manage. Typically these are repetitive, physically hard, boring, time-consuming, or highly-systemised tasks that employees are quite happy to outsource. That said, a new wave of technologies is fast entering every workplace to replace more advanced human skills as well. This next generation of technology employs complex algorithms and includes machine learning, natural language processing, computer vision, machine reasoning and decision making, as well as RPA and mechanical robots. Undoubtedly, the field of data science and analytics is changing too. In this way, new technologies help humans to process vast data sets faster (and extract the signal from the noise) and read through long documents in seconds (and extract key bits of information), which would take humans many minutes or even hours. In this way, artificial intelligence is giving humans machine intelligence.
Counterintuitively perhaps, AI is also being democratized. Whereas a year or two ago you would have needed to pay thousands of dollars to attend a top university to learn AI skills, the leading technology giants (such as Amazon, Google) and some forward-thinking academic institutions (like the University of Helsinki) have published MOOCs (Massive Open Online Courses), which are free for anyone to enrol in. Neo4J also offers downloadable textbooks for free, and the University of San Francisco has published a series of videos (circa 24 hours of more technical content) on CNNs, RNNs, Computer Vision, NLP, Recommendation Systems, Pytorch and much more. This self-paced learning and low barrier to entry (i.e. no need to be a Python or statistics guru) for (re)training or upskilling certainly helps to offset the popular narrative of jobs being ‘lost’ to Industry 4.0 and to reinvigorate those individuals whose roles are displaced by automation.
Since these courses are available for free and online, companies need a more agile, proactive and creative approach to upskilling their workforce for the future. Redeploying and retraining existing talent makes sense on so many levels. As well as this, the Fourth Industrial Revolution is also changing workforce demographics, and employees’ shifting expectations are challenging what it means to “go to work”. As the aging global workforce continues to contend with the advancement of automation, it’s clear that employers and individuals will need to adapt to new technology, adjust to ways of learning and stay competitive. It is my belief that in the same way that people can create web or mobile applications today (without the need for development skills) in a way that wasn’t possible ten years ago, AI will also become a “Lego-bricks” technology in the future.
So how is AI useful today?
If you consider common applications of AI, like natural language processing (meaning that a computer can understand ‘natural language’ inputs from a human), you can eventually replace keyboards as the primary way of interfacing with apps. One day we will all realise how inefficient keyboards really are as a way to get information in to a machine! Voice inputs can also trigger a voice output (as illustrated in the famous Google Assistant Phone Call with a hair salon) and let someone learn new things. For example: Pensions are a good way to save for retirement. An NLP-enabled platform could quickly:
- Give guidance as to what someone should be contributing to their pension;
- Provide a live update on what someone’s pension pot is worth today; and
- Allow individuals to make transactional decisions to increase their contributions.
Indeed, at Mercer we’re already building consumer apps with these technologies. For example, through the “Mercer Superbot” in Australia an individual can get financial advice through Facebook Messenger. Another example of AI use in pensions is Mercer’s automated Defined Benefit (DB) transfer robo-para planner. Mercer Jelf Financial Planning have partnered with Wealth Wizards to launch this service.
Artificial intelligence has many applications, but in the pensions and financial advice industry, it could at least be applied in these ways:
- Personalisation of communications: Employers and pension providers struggle with engagement. However, they could use machine learning to tailor communications. By correlating a user’s persona, personal data and financial history (ideally including data beyond their current employer’s data, which for most is an increasingly narrow point of view as people move from employer to employer more frequently) and referencing this with the behaviour of similar personas, marketers can automate communications and make them highly personalised. Since machines can process vast data sets quickly, they can also go beyond simple persona-based marketing to a very high degree of micro-personalisation, at scale.
- Educating Members: Whilst a few companies have moved from traditional paper and text forms of communication to personalised videos, natural language processing provides a new way for individuals to interact in real-time. They can ask questions and get instant answers. No more complicated modellers or doing Future Value calculations on their own.
- Prediction and Analytics: The biggest wins in AI today are often discussed as those ‘narrow’ tasks in which humans would perform very poorly. A good example of this is evaluating large data sets. In the pensions industry there is a vast amount of data that is rarely activated. AI can provide the edge and has the potential to highlight populations in employers’ schemes that will not be able to retire (due to insufficient savings) or those that will face other decisions, like what to do when saving into a pension is no longer efficient. The loop can then be closed back to personalisation and education to drive new actions.
If this all still sounds futuristic, many people simply don’t realise that their day to day life is already being run by artificially intelligent technologies, and they don’t seem to mind! Google’s search algorithm is using machine learning, and spam filters – one of the earliest use cases of AI – use artificial neural networks to detect and block spam from reaching an end user’s inbox. It’s time for humans to leverage machine intelligence, not continue to be afraid of it.