Skip to main content Skip to Footer

BLOG


August 24, 2018
What happens when the "things" in your IoT start thinking?
By: Harshu Deshpande

In 2008, there were already more objects connected to the internet than people, causing enormous excitement about how the internet of things (IoT), with its (theoretically) smart, communicating devices, would change the way we live and work.

I remember thinking it wouldn’t be long before my home would notice when I was running low on coffee—and order my favourite brand automatically. Shops would react to my presence with personalised ads and offers. Machines would not only predict when they needed maintenance, but also contact the company responsible and book an appointment during production down time.

Today, while all of this is possible—none of it has become mainstream.

The reason we’re not there yet is that, until very recently, sensors in the IoT have been "dumb." They’d know if something was "on" or "off"; "right" or "wrong". They’d take a reading. But they couldn’t remember it… or think about it… or do much to remediate it.

Previously, doing something intelligent with sensor data required sending the data to and consulting with an external system—or getting human input. The thinking had to be done at the "centre": back at base camp. Which was expensive—requiring custom-built solutions, a well-maintained network and high-speed internet connectivity.

Thinking at the edge

But now, for the first time, "things" can be intelligent in their own right. In the last six months, the tech giants have released devices with sufficient processing power to think for themselves.

  • Google has realised that its customers need to run deep neural models similar to those powering popular apps like Translate, Photos and Assistant, without being tethered to the cloud. The answer? Its own computer chip: the tensor processing unit—up to 80 times more efficient than standard processors and specifically designed to run deep learning models.
  • Amazon has released the world’s first machine learning video camera, AWS DeepLens, that can detect and recognise objects, faces and actions (Is this person brushing their teeth or playing the guitar? Is this a dog, a cat, or your toddler?).

And this is just the start. Soon, every sensor we embed in our operations, infrastructure, shopping centres and homes—could think for itself.

It sounds like a simple change: swapping out dumb for intelligent sensors. But the ramifications will be profound in terms of both what this new "Internet of Thinking" can help us achieve and how we’ll need to change our IT architecture to get there.

What does it mean?

When devices themselves can respond intelligently to what they’re sensing at the speed of thought, several very cool things happen:

  • Machine learning and AI are affordable – costs are significantly reduced because a lot of the raw processing of data can be done on the device, and we don’t need to build a network or a highly complex data lake. This means, smart sensing devices will be accessible to households and Mum and Dad businesses. Your home security system will not only be smart enough to ignore your cat tripping a movement sensor, but also knows the difference between your teenager or a burglar climbing in the window.
  • Innovation is democratised – As the cost of building bespoke hardware and devices to perform on device processing cost comes down significantly, smaller business will have access to large-enterprise innovation, enabling companies to deliver unique services closest to the end customer. Small businesses will be able to monetise their own developments (say, convenience store fridges that automatically call their owner if the power goes out overnight) by selling them to competitors or other industries with similar pain points.
  • Speed makes all the difference – when devices can make decisions "in the moment," it unlocks a whole new raft of possibilities. For example, healthcare company NeuroPace is already embedding intelligent sensors in a patient’s head to spot and stop seizures. The sensor identifies brain activity abnormality and sends pulses to stop a seizure within milliseconds of sensing it. By avoiding the need to consult an external system, or waiting for input from the patient, the sensor stops dangerous seizures before a patient knows anything is happening. After the first year, while it learns the patient’s typical brain activity, the device reduces seizures by 44 percent.
  • No connection needed – smart sensors can operate without being connected to the cloud. In Australia, this is great news for farmers on remote properties who’ve been limited in using agriculture solutions because much of the farm is out of phone or internet range. Soon, intelligent soil pH sensors on farms will tell farmers when, and how much, lime should be applied in a particular paddock—without any cloud connection. Equally, mine sites will be able to run smart safety, asset management and maintenance solutions autonomously.

One of the big winners will be autonomous vehicles that need to respond to new information (a child running into the road): 1) in real time and 2) even when the connection to the cloud is lost.

How do we get there?

Rather than focussing on building data lakes and in-house AI capabilities, CIOs need to start thinking about a future with edge-based architecture, where data is only brought back to the centre (core systems or a data lake) if it is valuable.

There’s a lot of upside to decentralised processing, including real-time responses, lower costs (specifically reduction in network costs), greater resilience and better quality, pre-processed data coming into your organisation. But supporting the instant insights and actions needed to create intelligent solutions at scale will require:

  • Shifting your thinking to the edge and moving away from the previous approach of collating all your organisations’ data in a central data lake and then driving insights. You can now look at processing data in the location that it is gathered and then at collating only when it has proved valuable or if the insights need to be combined with other data sets around your organisation.
  • Extending your existing infrastructure from the cloud to the edge to deliver intelligence everywhere. At a time when many companies have grown accustomed to software-driven solutions as their go-to strategies, you still need a renewed focus on hardware-based solutions at the edge.
  • Managing the lifecycle of edge devices – How will you update them as and when the algorithm or model evolves? How will you maintain them in a harsh external environment? Have you accounted for the lifecycle costs of these devices?

The rise of the Internet of Thinking means the scenarios futurists have been talking about for a decade are within our grasp. It’s a paradigm shift that few people have really got their heads around. I’m really looking forward to working with Australia’s early adopters. Because, no matter what your industry, this is going to change the game.

Popular Tags

    More blogs on this topic

      Archive