I am fascinated by self-driving cars. Slightly horrified actually. That video of the guy fast asleep behind the wheel of his Tesla hurtling down a road at highway speed is enough to make anyone hold their breath just a little and go, “How stupid!” And then, after letting it sink in… “How amazing!”
If the video was about a guy fast asleep in the passenger seat of a taxi you would think nothing of it. He too is using a “self-driving car”, isn’t he?
We have grown accustomed to trusting strangers with our lives on the premise that they are human – and know what they are doing. Not only to drive us around but also with managing our money.
As humans we will entrust an investment manager with our life savings, with the confidence that, as a trained expert with years of experience, they too know what they are doing.
Trusting machines with the same responsibility is still a huge a leap of faith for most people – but why should it be? Given the same training, trial, error and real-world experience as a human, why would the outcome be any less acceptable?
The building blocks of artificial intelligence
Let’s consider the ordeal of getting to the airport in peak-hour traffic. Not a simple task, given the many obstacles and dangers along the highway. But after the umpteenth trip, an experienced taxicab driver can do this without having to “think” about it much, using their brain (a literal neural network), connected to their eyes and muscles (a sensor network), to steer a car (a driving algorithm in the sub-conscious).
Machine learning is similar. A sensor network (cameras, LIDAR, GPS) feeds an artificial neural network to process the data using algorithms, each with a specific purpose.
Unfortunately, the subject matter of data science is so convoluted with complex terminology that it becomes difficult to see the forest for the trees (if you pardon the pun referring to AdaBoosting using K-means clustering, assisted by a support vector machine with principle component analysis to isolate the occurrence of K nearest neighbour objects matching orientation of similar gradients in a homogeneous dataset of spatially related green and brown objects...).
Data science doesn’t have to be rocket science. The basic capabilities of machine learning are the same as you would expect a human to have in the same situation. Expressed in simple human terms, the key algorithms in machine learning are responsible for:
- Detecting objects (e.g. potential obstacles in the area).
- Recognising distinct objects and patterns (like the minibus taxi ahead).
- Locating an object of interest in the scene (in the emergency lane where it doesn’t belong).
- Predicting where the object is going to be next (90% confidence it will cut in in front of us in the next few seconds).
- Reinforcement through trial and error (that’s the third one in ten minutes… Now 99% sure the next one will do the same).
The same capabilities and associated algorithms are as useful in personal financial management as they are in self-driving cars because detecting risk, recognising patterns, tracking trends, predicting the future and learning from mistakes is all in a day’s work for a financial adviser.
If we can successfully apply these algorithms in financial services, we can indeed create autonomous personal financial management – or a self-driving financial plan, if you will.
In fact, maybe a machine can be better at retirement planning than a human? Consider the inherent human handicap of a short memory, limited pattern-recognition abilities and the tendency to follow popular trends over personal circumstances. Maybe an algorithm will do a better job?
For starters, the machine has an extremely large dataset to draw experience from – including experience from a network of connected “brains” exchanging trial and error data at the speed of light. With its near-perfect memory, it can see patterns over time and can predict what may be lying ahead.
It is objective and unbiased, and not fazed by small bumps in the road. It doesn’t have a biological clock ticking, so it can make better long-term decisions. When the time comes to act fast, it can do many things in parallel, considering many courses of action to avert a catastrophe. It is dedicated to one master and that is you, serving your personal interests with infinite patience.
Imagine if your personal financial adviser was all that. Some humans are like that. The exceptional ones. As sophisticated as today’s machine learning is, it still does not compare to a truly talented human with a mastery in their field. The machine can emulate but it cannot innovate – yet.
If I had the choice, and the financial means, I would, of course, prefer a human expert to dedicate 100% of their time to me and my goals. Who wouldn’t want their own chauffeur or dedicated financial manager? But I don’t have the means, and neither does the majority of society. So for me, the most exciting development in AI is the ability of machines to learn from the most capable humans on earth – and one day bring those once-exclusive capabilities to me at a price I can afford. If the progress with self-driving cars is anything to go by, that day for financial services is very near indeed.
Rudie Shepherd is head of digital innovation and platform management at Alexander Forbes Empower.
This article forms part of finweek’s Collective Insight series titled “How technology is impacting on financial decision-making” published in the 24 October 2019 edition of finweek. To read the entire series, get a copy of this magazine here.