Labhesh Patel, Jumio’s CTO & Chief Scientist
Artificial Intelligence is all around us and it is a term that has quickly become embedded into our everyday lives. Likewise, the abbreviation, AI, is now so recognised as meaning artificial intelligence that many commentators don’t give much thought to what it actually stands for anymore.
However, what we’re starting to see is the acknowledgement that in many applications, AI can’t solve every problem, because of its inherent limitations. As a result, we’re beginning to witness the ‘A’ in AI become ‘augmented’.
So what is augmented intelligence? Well, it’s the fusion of human expertise and machine learning. But aren’t we at a stage where machines are as intelligent as us? The short answer is no, because what is clear, is that there are still use cases and scenarios whereby AI cannot work alone.
Celebrating the advancements of AI
It’s been a momentous few years for AI, there’s no denying it. We have seen how deep learning and image recognition technologies can diagnose possible signs of lung cancer. Google’s AI DeepMind has been beating professionals at the complicated sci-fi strategy game, StarCraft. London football club, Wingate & Finchley FC, is even using AI to coach and select its team’s formation and tactics.
These are amazing applications of AI, and in these instances, AI is at a stage whereby it can, on the whole, work with little human interaction. But there are situations whereby that human expertise is still required – this is where augmented intelligence comes into force. AI, at its core, enables humans to make faster, better and more accurate decisions. But, where the risk of failure is simply too great, or where human intuition is critical, AI cannot act alone and human expertise is still needed.
Here, we can look to the identity verification market to help explain this. AI models are incredibly successful in scanning selfies against the face on government issued IDs, such as a passport or drivers licence, to determine whether they are the same. However, there are instances where users might upload a picture with a low-res camera that could make the image blurred. Or, there might be glare on the image of the ID, distorting some elements. What’s more, with 195 countries in the world, all with different types of IDs, and with more being added on a daily basis, algorithms just wouldn’t be able to understand the nuances of specific fraud patterns for some of the more niche, or newer, IDs.
Building on the successes
It’s clear that there are still situations that exist whereby AI needs to be underpinned by humans to pick out unusual situations in the decision making process. This is exactly the forte of augmented intelligence. Think of it like Batman and his utility belt – without the technologies in this sacred belt, great things simply wouldn’t have been achieved. Together, though, it’s a different story.
This human / AI partnership is particularly important when it comes to algorithms and the need to train these models. Again, thinking back to the process of identity verification, when 10,000+ types of a specific ID are being scanned and reviewed each day, whether that’s a UK passport or a Dutch identity card, the process in itself creates a large data set. With humans, that data can then be tagged and machine learning models will eventually be able to differentiate the legitimate IDs from the fraudulent ones. When such a large set of data is tagged in this way, the algorithms get smarter and faster, and inevitably, will be able to recognise these patterns automatically.
What’s more, with augmented intelligence algorithms, they can start flagging suspicious parts of an ID, but then pass them over to human reviewers who can inspect them more closely. This process means that human agents spend their time and attention on unusual aspects of an ID, while the technology can sufficiently cover the standard checks. Ultimately, this speeds up the process and helps humans to make faster and more accurate decisions.
The advancements in AI we’ve seen in the past few years alone are astounding. The thought of having an AI-powered home assistant, or that self-driving cars would be in development, would have seemed baffling not that long ago. Yet, we’ve embraced AI with open arms. However, while some use cases need little, if any, human involvement, there are still a number of scenarios that do require a level of human expertise – as identity verification has shown us. For this reason alone, augmented intelligence has a very important role to play when it comes to AI, and will subsequently be critical to its advancement.