This is a contributed post written by Jas Singh in his role as chief technology officer for mobile & online health platform company Medelinked. The organization aims to empower people to take control of their own health, from building their health profile to discovering, connecting to and sharing their health and wellness information with trusted healthcare providers.
A ‘rash’ of predictions
The rash of predictions around IT for 2017 have included an almost compulsory reference to Artificial Intelligence (AI) as a technology about to revolutionize all aspects of the connected world, not least in my own area of focus, that of dragging the medical practice into the twenty-first century, which promises to catapult a host of connected diagnostic devices into the Internet of Things (IoT).
The problem is, whatever people claim about AI, the reality is that the systems that are currently being developed (and that have the most potential to impact healthcare or any other sector) are based on machine learning — and machine learning is subtly different to AI.
In case you think I am being needlessly pedantic, let me explain.
Put simply, current AI systems lack intelligence and, in a lot of cases, struggle to master even the simplest challenges without human-provided wider context. So, without intelligence, they won’t be the imagined magic answer for so many of our current big data management and IoT analytics challenges.
The need for human-provided context
To perform what is a simple machine learning task to decode the simple answer-question format, AI machines like Google Deep Mind or IBM Watson have to be fed terabytes of data and natural language examples to help them. When Deep Mind helps the UK’s NHS in a project to identify tumours, it’s doing so based on a massive amount of data about what a tumour looks like and it is benefitting from a neural network model that allows it to ‘learn’ from the data to increase its accuracy.
But, again, it is a narrowly-defined task and it is only because of human hand-holding and ‘training’ that these machines can deliver apparently impressive performances, but they are a long way away to thinking for themselves.
Artificial? Yes. Intelligent? Not so much.
Obscuring machine learning
To claim these and other applications are full AI is counterproductive and misleading and risks obscuring the significant contribution that machine learning, along with natural language processing and cognitive computing, can make.
Fundamentally, what we have today in machine learning is the ability to take a topic, a narrow set of instructions and feed these into a neural network to identify patterns based on massive amounts of data. This technology has been around for many decades but it’s the availability of cheap computing power and a constant steam of data from IoT devices that makes it more relevant and useful as we are now able to quickly identify patterns that a human brain alone can’t.
So, technologies based on machine learning developed with a specific context and problem in mind have a lot to offer currently in terms of improving services, decreasing costs and delivering insights.
In the future, systems may eventually be truly intelligent and able to operate without context. But to do that they will be endowed with capabilities more like the feared Skynet from the Terminator film series. But current systems are not nearly advanced enough to master simple tasks on their own, let alone pose existential threats to humanity (although, as Messrs. Gates and Musk have warned in the past, it’s a prospect we should take very seriously indeed).
Whether the current rush of claims for AI is an attempt by companies to bolster valuation by claiming such capability to unleash another cycle of raised customer expectations, changing business models and general disruption, I’ll leave you to ponder.
But, I’d argue, in the meantime, as we wrestle with the challenges posed by the Internet of Things, the connected technology industry needs to use the term judiciously, set realistic expectations about AI’s promises and be clear with customers about what it and machine learning can and cannot deliver.