Revealed: how AI will change the world of finance ... sooner than you think
AI technology is already changing the finance industry with all manner of innovations – but the really impressive stuff is yet to arrive.
And when it does, insists Maxime Dumas, Principal Researcher at Croesus, it will change the way we work forever.
It all comes down to a technological approach known as “deep learning”’, in which algorithms mimic the neural networks of the human brain.
“Neurons are interconnected with other neurons on layers,” Dumas explains. “Each layer basically learns an abstraction of the previous layer. If we take face recognition, for example, the first layer could be specialized in detecting edges in the image.
“The second layer could use the edges to detect noses, eyes and mouths. A third layer could learn parts of the human face by composing the nose and the mouth. Nowadays, models are composed of hundreds or thousands of layers. That's why we talk about ‘deep’ (learning). Models can learn much more complex interactions between data because of that.
It all sounds like exciting – and revolutionary – stuff. It’s easy to see how such tech can be applied to the finance industry (and, indeed, any industry).
But surely AI implementation comes with a whole new set of obstacles and problems?
“Bringing AI research projects into production is really hard,” Dumas says. “Developers often have no idea how the models make their decisions. This is a huge issue for most applications where predictions could have an impact on human life, for example in credit risk assessment.”
“There is also the issue of bias. Machine learning systems are trained using data. If some classes in the data are misrepresented or unbalanced in the dataset, this can result in really bad situations. Most of the time, developers had no idea there was such an issue in their data, and it is not always easily detectable at first sight.”
“Private life is also a major issue. Since training machine learning models requires large quantities of data, many corporations end up with large databases of personal information about their users. There have been large public cases where things have gone wrong because of that, such as data leaks.”
“There are promising solutions on the horizon though. For example, Edge AI aims to train models at the ‘edge’, where the data is collected, in order to avoid collecting the raw data centrally. Differential Privacy can protect individual records by introducing some amount of neutral noise in the data. Finally, Synthetic Data Generation tries to build training datasets, completely free of sensitive data. This is something we know many financial institutions in Canada are exploring seriously to protect their client data.”
As for the most troubling question of all: whether AI is going to replace human wealth professionals altogether? It’s a bit of a double-edged sword, as Dumas explains.
“AI will have a lasting impact on human workers. In 2017, PwC reported that the finance industry, in the next 20 years, should expect a workforce reduction of around 8%. Jobs that can easily be automated are especially at risk. But technology-related job openings are expected to rise, and industries where human relations are important, such as healthcare and education, are also expected to grow.”
