Machines can process massive amounts of data that humans can't, but sound governance structures are needed to ensure positive results.
When you “work through” a problem or issue that requires a decision, you likely feel as if you’re going through a linear checklist. But that’s not how the human brain operates; it processes in a non-linear pattern. And this is essentially how deep learning, a subset of artificial intelligence (AI), works too.
Deep learning works like the human brain
Deep learning, at its essence, learns from examples — the way the human brain does. It’s imitating the way humans acquire certain types of knowledge. Because deep learning processes information in a similar manner, it can be used to do things people can do – for example, learning how to drive a car or identifying a dog in a picture.
Deep learning is also used to automate predictive analytics – for example, identifying trends and customer buying patterns so a company can gain more customers and keep more of them. You know those sections on retail sites that show items “frequently bought together” when you’re purchasing a new screwdriver? Those are based on predictive deep learning algorithms that have considered both your current search and past buying patterns to suggest additional products you might also need.
Other applications include numerous everyday encounters and activities, such as virtual assistants, fraud detection, language translation, chatbots and service bots, colourization of black-and-white images, facial recognition and disease diagnoses.
A simple example of a neural network’s application is in parsing speech. The network takes sounds from raw audio, which combine to make syllables, which combine to make words, which combine to make phrases that prompt actions. The machine learns that this particular sound means that it should pull up a credit card balance and the more times it’s asked the same thing, the more accurate it gets.
Deep learning has applications across industries
Neural networks are not new; they’ve been around since the 1940s. In 1943, two computer scientists introduced models of neurological networks, recreated threshold switches based on neurons and showed that even simple networks of this kind are able to calculate nearly any logic or arithmetic function.
The first computer precursors were developed by a computer scientist who was tired of calculating ballistic trajectories by hand. Today, more than 70 years later, deep learning has exploded in sophistication and use, primarily because of expanded computing power (along with greatly reduced costs per unit of power), better modelling and the availability of data. Deep learning requires massive amounts of data. Currently, it’s estimated that the data we generate every day is 2.6 quintillion bytes. And it can analyse massive datasets far faster than a human. Machines don’t suffer from monotony or fatigue.
Are there risks with deep learning?
Let’s answer that question using the example of autonomous vehicles. Deep learning has given us these self-driving cars, but they seem unlikely to eliminate all road accidents, something that would be akin to a self-driving utopia. In fact, a recent study from the Insurance Institute for Highway Safety (IIHS) says that autonomous vehicles might prevent only about a third of all crashes. Still, that's more successful than people.
Yet concerns about widespread adoption may also include an increase in accident rates in the early days of rollout as the technology learns, moral decisions being left to manufacturers, and difficulty in attributing responsibility for accidents. And then there’s hacking, because, after all, deep learning is simply technology encased in a vehicle. In March 2019, two “white hat” hackers (the good guys) needed only a few minutes to go through the browser of the infotainment system to get inside a Tesla’s computer, run their own code, and have the car respond to their commands.
We must also consider the use of deep learning from the consumer’s viewpoint. If it doesn’t “work”—for example, a phone fails to unlock—it can create an unhappy or frustrated customer, which defeats the purpose. Compounding the issue, because of the complexity of the neural networks in deep learning, it can be difficult to know where or why the system went awry. Often described as the black box of deep learning, data scientists are working to improve the visibility and transparency around how deep learning models work.
Models can also have bias unintentionally built-in — and these deep learning models are being used for significant decisions, including who gets loans, jobs or parole. Deep learning needs to have clear guardrails with appropriate governance structures.
Deep learning is the future of business
Deep learning has given us image-based product searches – Pinterest, for example – and efficient ways to sort fruits and vegetables to reduce labour costs. The former is more of a consumer convenience, while the latter is a true business case for productivity.
Significant resources are being put into deep learning in financial services, in which it is used to detect fraud, reduce risk, automate trading and provide “robo-advice” to investors. According to a report from the Economist Intelligence Unit (EIU), 86% of financial services firms plan to increase their AI-related investments by 2025.
Source: Beena Ammanath Executive Director, Global Deloitte AI Institute and Trustworthy AI/Ethical Tech Leader, Deloitte;
Kay Firth-Butterfield Head of Artificial Intelligence and Machine Learning; Member of the Executive Committee, World Economic Forum