Privacy Security

Amazon Alexa

IoT is a huge topic and one that is hugely relevant for all computing students I feel. There is a tremendous number of data points generated and a scary level of security in general – so very many loopholes!

“The Gartner Hype Cycle of Emerging Technologies“ was covered in detail yesterday so I won’t repeat myself too much here. But with regards to IoT, it is too difficult to classify the whole branch on the graph. Certain elements have achieved massive market penetration such as the Amazon Echo in the USA, but does it have real utility?

Despite the skill growth as shown below the overwhelming majority still use it simply as a timer whilst cooking, playing music or giving weather information. Certainly that is the way we have used it over the last two years – the skills just end up being frustrating and way more time consuming than simply activating the action via your phone. Counter to this though has been the wonderful effect it provided to my Grandad, who in his 90’s had an amazing experience accessing audio books, radio and news with one.

It and many other devices have a massive privacy problem ( and Understanding these points beyond the surface level have helped us decide to get rid of the device. But without sounding extreme, three times have we received targeted email adverts about a conversation we have recently had – topics we have never searched via Google or social media! Is it simply inferred from other data points or was it listening?

The issue of privacy due to the multitude of data generated clearly provides benefits to a data scientist looking for an ever more valuable or interesting data set.

Another issue I’ve read about previously is the availability of IoT devices in smart home set ups. Smart door locks have backup battery systems for example but the volume of dependencies in the systems are extreme. The diagram from our lecture slides shows how many different points of contact there could be in a typical cloud architecture setup.

A simple network error could mean the heating goes off and your pipes freeze for example. The bigger point though is the invisible level of digital maintenance required. The number of settings, options and updates that need to be carried out for a multi device home can be absurd. We of course dream of the future and a frictionless experience but this hasn’t happened yet.

In combination with more and more advanced AI systems, IoT devices that have vision capabilities are dominating world news on a regular basis. From the riots in Hong Kong to daily life in China, facial recognition is having some scary consequences and brings to the fore technology ethics.

Just today I read (via a link from Twitter – how a massive US restaurant chain is testing a system that will surely have a negative effect on staff. “The program will sync with existing cameras and use machine learning to “analyze footage of restaurant staff at work and interacting with guests,” and send statistics to managers.“!

The price of components are so low now that creation and prototyping of IoT devices has become so much easier. The home of the ‘component industry’ is Shenzhen. “Shenzhen: The Silicon Valley of Hardware” was something I watched a couple of months ago that provided me with great insight on how much creativity and entrepreneurship is in the space. I wonder if it is this element that will help some IoT devices overcome the hype cycles trough of disillusionment.

DEVICE TYPE – Smart speaker that has personal assistant functionality

SENSORS – The Echo typical use case is via voice interaction using microphones and speakers. ( “The main control is the seven-microphone array built into the top, which uses beamforming technology and noise cancellation to “hear” your voice (i.e., record it and send it to the cloud).” The article also provides the following hardware information: ” A rectangular, side-mounted board above the speakers contains the brains and communication components, including the following [sources: Cunningham, Detwiler, Ifixit]:

  • TI DM3725 ARM Cortex-A8 Core Digital Media Processor
  • A TI power management integrated circuit (IC)
  • 4 gigabytes (GB) of NAND flash memory
  • 256 megabytes (MB) of LPDDR1 random access memory (RAM)
  • A Qualcomm Atheros QCA6234 dual-band 802.11n WiFi and Bluetooth 4.0 module.“

CONNECTIVITY – Connected via wireless internet, the device has no functionality if not ‘online’ and connected although the app that accompanies the device is still accessible. The speaker also has Bluetooth capability for playing other audio from your phone or tablet.

DATA – The speaker connects with the Alexa system and processes voice data locally and in the cloud. I perosnally haven’t found it to be so effective but the system apparently adapts to learn your voice commands better.

DATA FREQUENCY – According to Privacy International ( data is sent to the cloud after every activation/usage.

VALUE PROPOSITION – With over 80,000 skills (or 3rd party apps) the device aims to solve a multitide of problems. The simplest extraction would be, to quote their webpage, “Make your life easier at home”.

SECURITY – The above article link also describes in detail how the microphones must always be on to be able to respond to the wake or activation word. In theory these snippets are stored locally and are deleted at regular intervals but the process is not totally transparent and the devices involvement in serious crime cases proves how vulnerable you are. Many feel they have nothing to hide but you never know what the future brings. The new third generation device has made a slight change to tackle customer scepticism as seen in the picture below from their website. Amazon is a technology behemoth and so the volume of different data points they can infer or predict customer behaviour must be immense.

I found it interesting to dig into the hardware side of things, something I’ve rarely done and know little about. My focus in the last couple of years has revolved around chatbots that are web based. The technologies have a lot in common and a lot of similar use cases. The biggest problem I have encountered with chatbot development is an even bigger problem in voice interfaces. The promise is so great with the technology because it is so natural – in theory interacting with the device doesn’t have to be learned – you just ask it what you want to do. But this ‘natural interface’ also raises expectations to a human level. So when it doesn’t work, users frustration is much higher than if a laptop crashes for example – laptops don’t get shouted at for being dum as much as personal assistants do!

The experience of using and researching this device will continue to help me throughout my studies as a frame of reference whilst I learn and gain more data science skills. For example running an experimental deep learning algorithm on a large data set of voice snippets to reveal behaviours is quite an exciting challenge I hope to take on.

Leave a Reply