Week 37


Google’s Pixel launch this week dominated the tech agenda.   The company announced a range of hardware products covering a range of form factors from VR headset and smart home speaker to two high end Pixel phone models.  The latter are built around a Snapdragon 821 quad core processor running Android 7.1 and will ship with an integrated personal digital agent called Google Assistant:


Veteran pundit Walt Mossberg saw the significance of the announcement in terms of its relevance to Google’s post-mobile future which seems to be very much focussed on bringing artificial intelligence to the mass market:

a huge, bold move that will pose new challenges for the other major players in the tech industry. … because the company made it clear that the array of new Google-made hardware devices it rolled out … are important vessels for the technology it believes is the key to the entire future of tech: artificial intelligence.

Time saw the potential upside to the Android platform but also signalled potential problems ahead in further alienating Google’s existing hardware partners.  The decision to effectively compete with their these partners on consumer products brought back memories of the ultimately unsuccessful attempt to partition Psion into a hardware and software division exactly 20 years ago.  The latter went on to become Symbian while the former effectively shrank from view within a few short years.   Existing top end Android OEMs are likely to take a similarly suspicious view of developments since they are unable to compete with Google on software integration and will be left under greater pressure than ever to differentiate on a base Android proposition.   Judging from their public statements, Google don’t seem to see any problem.  The company has had several attempts over recent years to build consumer hardware notably through their Motorola acquisition, ChromeCast and Nexus.  Arguably none of them have been particularly successful.    Perhaps this time the addition of advanced AI will make the difference.

Coincidentally Samsung announced they had acquired Viv, a leading edge AI assistant proposition built by the creators of Siri.  Viv has been covered in this blog in the recent past – the world premiere of the technology was noteworthy:

In terms of what Samsung will do with Viv, this statement from the company’s Mobile Business CTO provides some background context – it’s a third party service ecosystem play not dissimilar to Amazon’s Alexa platform:

It’s enough to tempt you to try out some Apple Mac smelling salts:

Artificial Intelligence

The news that a Russian AI startup called Luka has created a “memorial chatbot” using the Telegram message archive of Roman Mazurenko, a dead friend and colleague, was widely reported in the mainstream press.  The Verge published a long post article on the subject providing a tantalising glimpse into the development process.  The most profound realisation here is that a reasonably compelling messaging proposition, which many have found uncannily close at times to the real Roman, was built using standard open source software technology freely available today.  The Verge provided this remarkable sample conversation.  This development genuinely raises the prospect of “memorialisation chatbots as a service” as a grief support tool for the recently bereaved.   Thinking about it, there’s probably way more than enough material in nearly three years of weekly blog posts to build an import digest chatbot that outlives its creator…:

Luka had been using TensorFlow to build neural networks for its restaurant bot.  Using 35 million lines of English text, Luka trained a bot to understand queries about vegetarian dishes, barbecue, and valet parking.  Using more than 30 million lines of Russian text, Luka built its second neural network.  Meanwhile, Kuyda copied hundreds of her exchanges with Mazurenko from the app Telegram and pasted them into a file.  … Only a small percentage of the Roman bot’s responses reflected his actual words. But the neural network was tuned to favor his speech whenever possible. Any time the bot could respond to a query using Mazurenko’s own words, it would. Other times it would default to the generic Russian.

Relatedly Duolingo just “introduced chatbots to its app that allow you to have AI-powered conversations.”

Nick Bostrom’s keynote at IPExpo emphasised the importance and challenges involved in building advanced AI systems with built-in human value alignment from the outset.  In particular, it would be necessary to create a digital moral framework with clear constraints to ensure the AI could not reprogram itself to ‘improve’ its value alignment algorithms.  This telecoms.com podcast attempt compiled after IPExpo attempts to explain Bostrom’s talk and muses on the difficulty simply in trying to specify any such framework:

To get a sense of how fiendishly complicated this might be, it is well worth checking out this thought-provoking MIT-hosted resource which provides an illuminating glimpse into the dramatic scope for variation in human value judgements between supposedly normal human beings.  Basically, life or death decision making scenarios that may ostensibly seem clear cut from a moral perspective prove to be anything but in reality and we all sit on a judgement spectrum.   Here’s mine.

Sam Harris strikes a similar tone of concern in this recent TED talk in which he raises the spectre of an arms race between nation states and leading companies to achieve the ultimate prize of what Bostrom refers to as human level machine intelligence (HLMI) which is essentially another term for AGI (Artificial General Intelligence). Harris suggests any current complacency over the eventuality of ASI being a long way off is dangerously misplaced.  He emphasises the exponential nature of the intelligence escalator and brings up Bostrom’s cautionary note that an ASI may treat us with the same consideration we give to ants.  Which is namely some curiosity and degree of coexistence but unhesitant extermination when they get in the way of something we’re trying to do.

When we get there we will have to contend with the possibility of mass scale technological unemployment from “full brain emulations” or ems.  This post reviewing economist Robin Hanson’s book entitled “The Age of Ems” tries to paint a picture of a “post-human” landscape.  It’s a topic that Bostrom touches upon in his book, Superintelligence but in his clear view AGI will very quickly give way to ASI and acceleration way out of our reach up the intelligence escalator.   The use of any terminology that suggests AGI will be anything other than a blink in the eye of this acceleration is potentially missing the point.  Still, if we survive the transition, at least there’ll be no need of the likes of Donald Trump:

Once ems will be doing most of our jobs, humans will have to create a new civilization for themselves. Hanson imagines ems will be congregating in a few megacities full of technology, while the humans will live essentially retired in suburbs. The ems will be running all the businesses and government.

Recent advances in Reinforcement Learning (RL) combined with Deep Learning (DL) are likely early milestones on the road to AGI and beyond.   This early draft of an exhaustive technical book covering Reinforcement Learning is a useful and highly detailed reference point.  Another resource courtesy of Denny Britz provides a more practical guide to understanding how to use RL/DL techniques by linking to various implementations many of which are powered by Google’s TensorFlow platform.  These sort of approaches are being used today by a growing number of startups. CBInsights previewed 60 using DL today across healthcare, robotics, security and ecommerce:


For now consumers can make do with the Amazon Echo which will be going head to head with Google Home over the coming year.  The latter presumably leverages TensorFlow too.

Cloud and Digital Transformation

FirstRound post on why startups need to put Customer Experience at their core starting by rethinking customer support:

“We’re in a world that’s increasingly intrigued by AI, bots and self-service models. Support interactions between humans will become increasingly rare — and even more atypical from those who deeply know and care about the product they represent. But the moment you have a genuine, kind and human interaction that resolves your problem, it’s surprising and delightful.”

Chanel’s digital transformation of the catwalk incorporated some questionable data center chic.


Canalys chief Steve Brazier warns that “the world’s largest cloud players are in danger of becoming too big to fail“:

“at least one of them is going to stumble and fall, taking customers’ off-premises systems with them. … And depending on when this happens, the potential consequences for all businesses could be catastrophic.”:

The solution he suggests is to seek to re-cultivate obsolete hardware skills:

“Hardware engineers were getting increasingly old, as a workforce, and there is a genuine danger, he claimed, of the skill set dying out.”

The Information on how data analytics is changing startup funding driven by this fundamental dynamic:

Art by Sam Lessin.

IP Expo referenced above was a good opportunity to understand the current state of the art with respect to cloud computing.  It was way busier and well-attended than last year.  Particular highlights were keynotes from Nick Bostrom on the road to AGI, Mark Russinovitch (Microsoft Azure CTO) on cloud and Gavin Jackson (MD AWS Europe) on the history and future of AWS.   This image from the Microsoft presentation provides a vivid illustration of the changing corporate IT skillset:


The Internet of Things

Quartz on one of the world’s biggest distributed denial of service (DDOS) attacks ever and how it appears have been launched from a botnet of compromised IoT devices (in this case webcams).  The target was the personal blog of Brian Krebs, a top computer security researcher and the armada of infected hosts assaulting it was on a biblical scale:

“The size of the attack was nearly double that of any Akamai had seen before. … The number of machines in the latest botnet is still unknown, and could be as large as a million.”

The Register on why the pendulum is swinging back to hardware in the form of custom processors.   Hardware, in short, is starting to eat software largely off the back of IoT.


Fortune magazine on why software patents are “in peril” following a landmark ruling against controversial so-called “patent troll” outfit Intellectual Ventures.

Sobering post outlining the case for why simple testing of unhandled errors “can prevent most critical failures in software:

“A majority of the production failures (77%) can be reproduced by a unit test.”

Good survey of Python data visualisation landscape covers the usual suspects such as matplotlib and ggplot and also introduces a new (for me at least) contender in the form of Altair.   Here’s a simple example of how to use it to generate a bar graph:

import pandas as pd
from altair import Chart, X, Y, Axis, SortField

# From: http://pbpython.com/altair-intro.html
file =
df = pd.read_csv(file)
top_10 = df.sort_values(by='amount',ascending=False)[:10]
top_10.plot(kind="bar", x=top_10["detail"],
 title="MN Capital Budget - 2014",
chart = Chart(top_10).mark_bar().encode(x=X('detail'),
# NB. To get this to work, I had to add an
# "https:" to the script locations in output html
html = chart.to_html()
with open('chart.html', 'w') as f:

How code profilers work with worked Python examples covering both deterministic and statistical cases:

Deterministic profilers record exact time of all function calls and returns and then process collected data to determine duration of function calls and other important parameters. …. Statistical profilers periodically take snapshots of running program and deduce where time is being spent.

Hilarious glimpse into the crazy crazy world that is JavaScript in 2016:

“Look, it’s easy. Code everything in Typescript. All modules that use Fetch compile them to target ES6, transpile them with Babel on a stage-3 preset, and load them with SystemJS. If you don’t have Fetch, polyfill it, or use Bluebird, Request or Axios, and handle all your promises with await.”

What do engineers do all day?  This Medium post attempts to articulate an answer using the analogy of a group of writers collaborating on creating a book.

A recent InfoQ article provides a fascinating insight into what Spotify engineers do all day. Interesting to see organisational culture is central to their engineering operations. At the heart of that culture is autonomy, diversity and a focus on the long term.


Observations from a tech due diligence survey provides a surprising insight on hiring.   References seem an obvious and easy thing to follow up on so it seems baffling why they’re not used more widely:

“Half of the companies get less than 10 applications when they advertise a position. Based on that, it becomes clear that building an attractive brand for developers is an asset.  I guess related with that, once companies attract a candidate, most of them try to finish the hiring process in a month. The survey also shows that developers aren’t in love with pre-screening tests. Another surprise here is that only a third of companies do backdoor references. Because in our experience, that’s one of the best predictors for failed hires.”

This listing of the world’s most valuable companies and how they have changed over the last 10 years provides ample evidence of the huge disruption that software has wrought:

The Ken is like The Information but for India.

Fortune article on the struggles of and eventual replacement of Lands End CEO highlights the challenges that any outsider Change Agent faces in entering a corporate culture in the face of institutional resistance.

No individual however great can ultimately make a change happen. They can inspire incumbents and hire newcomers that help spread the world but ultimately change comes from within:


The Sisyphus kinetic art table provides a tech-inspired Zen garden for your living room.

Why humans are “unlikely to ever live longer than 125 years“.  Our bodies and brains aren’t physically designed to last any longer than that.

A world champion public speaker provides some tips on how to do great presentations – don’t give it all away in your title, start bold and focus on three points:

Researchers at Columbia University have been trying to create an accurate high level model of human memory which captures how memories consolidate over time.  This ensures the brain operates in a highly efficient manner several orders of magnitude lower than current AI systems.

Their model shows that over time, as a person stores enough long-term memories and accumulates enough knowledge, human memory storage becomes more stable. At the same time, the plasticity of the brain diminishes. This change helps explain why babies and children are able to learn so much so quickly: Their brains are highly plastic but not yet very stable.

Donald Trump

An already depressing campaign inevitably reached a new nadir this week and now risks rupturing the earth’s crust and descending into magma following the biblical level of bile in the second head to head. It makes for a wholly dispiriting contrast with the tone of the election just eight years ago in 2008:


Almost the only consolation is that it will be one way or another all over in a month’s time.  In case you’ve been hiding under a rock, Donald Trump’s increasingly improbable hopes of reaching the White House cratered following WashPo’s revelations of his lewdity in the form of a hot mic tape taken 11 years ago.   It elicited a tidal wave of revulsion including many women in the US which he essentially tried to ignore and then laugh off as ‘locker room’ banter but which will surely devastate his appeal to female voters. There’s an interesting subtext here. Trump has been on record threatening to “go after” Jeff Bezos and Amazon for tax evasion which is quite something given his own refusal to come clean on his tax records.  Bezos also owns the Washington Post and they have been doggedly relentless in their pursuit of Trump.  All bets are off as to what he’ll attempt to do with WashPo if he does somehow contrive to clutch victory from the jaws of defeat.   The list of those standing in opposition to his views and candidacy is now so long he cannot surely “go after” them all. Even before the WashPo story broke The Atlantic broke with normal tradition to make the case against him.  Then again, it’s worth remembering that “Donny” Trump was the archetypal hero for Patrick Bateman, the fictional American Psycho.

What does it say about the country when we have a presidential candidate who is such a parody of ’80s excess that he literally served as the role model for a racist, homophobic, misogynistic character who is the ultimate parody of ’80s excess?


Over in the UK, events weren’t very much better. The Conservative Party seemed to be lining up to out-compete one another in Brexit tub-thumping rhetoric with a zeal reminiscent of the Tudor Reformation.   Prominent amongst the sabre rattling was a promise to crackdown on overseas students and a very controversial threat to force employers to disclose how many non-UK staff they employ. Anyone working in the UK tech industry today understands the diverse nature of a typical workplace and the immense challenge of finding the best possible tech talent.  What is less clear to outsiders but entirely obvious to anyone who knows the scene, any attempt to gerrymander employee provenance is bound to end up in failure. Any jobs lost will almost certainly not go to UK candidates.  They’ll simply be outsourced depriving the UK of any economic benefit and creating ill will abroad in the process.

Later in the week the government appeared to back down on the “name and shame” aspect of the foreign employee disclosure and claim they only wanted to “gather data to aid decision-making”.  John Naughton provided a disturbing reminder why the collection of identification data of this type can have terrible unforeseen consequences by considering what happened in the case of the Netherlands in Second World War where it was co-opted by the Nazis to brutally efficient effect to aid the Final Solution.  Perhaps appropriate then that one of the best responses to the development was James O’Brien’s radio piece in which he provided a quotation that he suggested was from Amber Rudd’s speech and turned out to be from Hitler’s Mein Kampf.  Spot the difference:

“For the state should draw the sharp line of distinction between those who, as members of the nation, are the foundation and support of its existence and greatness, and those who are domiciled simply as earners of their livelihoods of there.”

The incendiary language prompted one developer to release a Chrome plugin that does a search-replace of “immigrants” with “people”.   The cold reality is that there really are only a few options on the table for the UK to extract itself as swiftly as possible from the EU.   This table from HSBC outlines the difference between then them:

hard and soft brexit hsbc

Choosing the hard exit option would mean leaving the single market which by common consensus will severely impact the UK economy perhaps by as much as £66 billion per year.  If anything approaching that endgame comes into view, there are likely to many who demand an alternative given evidence that many who voted for Brexit did so in the anticipation it would be a zero-cost exercise.  No wonder cartoonists are having a field day with rollercoaster analogies:

A revealing study showed a clear correlation between nostalgia for a bygone Britain and voting Leave, perhaps the clearest evidence for the Little Britain effect that drove Brexit:

The clumsy public expression of neo-parochialism with lashings of xenophobia from our elected politicians has potentially very negative consequences in terms of the image of Britain abroad. Many of the places we suppose we will be “unfettered” to conduct free trade with post-Brexit took an pretty dim view of the UK government’s pronouncements and the pound has dropped in value significantly in the days since.  We’re in real danger of alienating the very people we will be counting on outside the EU and will need to collectively work hard to counter the kind of startling negative sentiment being revealed from some distinctly unlikely sources.

John Harris in the Guardian also pointed to the real danger of having the UK brand irredeemably associated with bigotry and intolerance.
Perhaps we need a new popular movement to inspire the young both inside and outside the UK along the lines of the Anti-Nazi League in the 1970’s.  A BBC documentary covering that period entitled “The People’s History of Pop, Tribal Gatherings 1976-1985” provided a superb window on those troubled times.   Here’s The Clash playing Victoria Park in 1978 as part of a Rock Against Racism gig: