15 Apr 2016

Don’t be evil, try telling that to the robots!

It’s 1977. The Goblin Teasmade advert is on television screens, Punk is hitting the UK streets and Clifton Country Primary (CCP) are engaged in the dying embers of a grudge soccer match against arch pre-teen rivals, Mayfield. It’s nil-nil. 

CCP have just brought on a substitute, Steven Roundtree. He’s new to the team and it turns out new to the rules of English football but fresh legs are required. The diminutive Mayfield goalie takes several strides back in a studied run-up to the last goal kick of the game.

Insanely, Roundtree charges towards the spot and rockets the ball into the top right corner of the Mayfield net, before screaming back up the wing, fists pumping, anticipating the schoolboy pats, hugs, plaudits that never come.

I thought of Steven Roundtree when I read about the Racist Robot debacle, this week. If the story hasn’t somehow hit your Facebook-curated news algorithm, it centres around a Microsoft PR techno-wheeze which spectacularly backfired in the megacorp’s face.

The software leviathan was not to know that releasing a machine-learning program, known as ‘Tay’ into the internet wild to ‘improve customer service’ would have the equivalent effect of putting Attila the Hun in charge of the Brexit campaign.

Within hours, a bunch of horrible trolls got hold of the algorithm and poisoned the innocent Tay with nazi invective.

Within hours of letting the new-born bot loose on Twitter, teen-voiced Tay’s timeline was transformed from saccharine-sweet tweets like ‘humans are super cool’ to a more jackboot-stomping cadence, fuelled with anti-social juvenile vitriol such as ‘her’ now infamous tweet, ‘Hitler was right’.

All this on the same day that tech-beat BBC radio journeyman, Rory Cellan-Jones was allowed a segment on the highbrow Today Programme with Siri reading out the news to highlight some esoteric story about how ‘bots’ (wtf, asks Radio 4 listener Missus Lampeter of 1 Acacia Drive, Eastcote) are the new apps.

If the Beeb had waited a few hours, it could have had Tay vomiting out the news in a kind of techno-tourets.

The Meta story here is that most Internet of Things (IoT). No longer a geek boy plaything, IoT is now the cipher for how we will soon do everything, or rather, have everything done for us. It’s like the Goblin Teasmade of the digital age.

Soon, and we’re talking years, machines will undertake all of the drudgery of our daily lives at work and in the home as well as transcending the art of headline writing, language translation and forecasting natural disasters or avoiding terror attacks. Recently, Google’s machine learner Al beat the top player of the ancient and complex game, Go, at the DeepMind Challenge.

So how then, with this super-abundance of intelligence did Microsoft’s Tay get all ‘her’ social sums so wrong and what does this say about the future of a small thing called humanity?

As Steve Hawking says in Wired magazine, “A super intelligent AI will be extremely good at accomplishing its goals, and if those goals aren’t aligned with ours, we’re in trouble.”

Thought leaders and pioneers of machine learning are of the same voice as Dr Hawking. The prestigious MIT Media Lab appointed a Dalai Lama-grade monk to its Ethics panel, the Venerable Tenzin Priyadarshi, philosopher, philanthropist, polymath and all-round virtuous big brain. The artificial intelligence and governance initiative aims to examine ‘meaningfulness and moral purpose’ between individuals, organisations and societies.

The High-Church ‘don’t be evil’ Google aspiration will hopefully pass on its mantra to the internet of things. When neuroboffin Demis Hassabis’s company Deepmind was acquired by Google for $400 million, he reportedly asked the search giant to create an ethics board to oversee its AI research as a condition of its acquisition, which it did, autonomously.

Microsoft’s Tay marauded around the internet for 24 hours before being shut down. Steven Roundtree was allowed a few fist pumps before he was summarily sent to Coventry for throwing the 1977 Inter-Primary School Cup final for CCP. I wonder if he ever learned from that experience. I bet Tay has.


Steven Roundtree is not his real name.


17 Mar 2016

Machine-learner difficulties

Google’s Al might have stuffed the Go champ but youthful instagrammers are sticking it to the machine.

Google’s Al beat the top player of the ancient and complex game, Go, at the DeepMind Challenge, recently. It didn’t so much beat the South Korean champion, Lee Sedol, as wipe the board with him 4 straight wins to 1.

The news has been reported variously as the Fall of the Geek or the Rise of the Machines and there’s been little sympathy for the down-in-the-dumps human, tangled up in it all.

The story has been reported with a fan-boy zeal by Reuters, CNN and the BBC with a mouthwatering message about how machines can learn independently, approximating that ole wives’ favourite, human intuition, by studying  countless historical Go matches and using simulated Go games to perfect an emotionless, unbeatable strategy.

Well, not quite, Lee did beat Al once at least!

It’s significant, according to boffins at Carnegie Mellon University, because of the applications to other areas like health care, scientific research, even the law.

News breaks of Al’s smug victory over a crestfallen ancient board game strategist at the same time that photo share service, Instagram announces it’s going to order our feeds according to a number of dubious machine factors.

The change has come about with the outstanding statistic that instagrammers miss about 70 percent of their chonological feeds because, well, unless you’re a teenage girl, you’re not on it all the time and with 400 million regular visitors that’s a lot of lost content and potential squandered advertising messages.

The new feed will be ordered by machine learnings and a mix of ‘signals’ to determine the photo and video flow.

In Human that means there is a piece of code, an algorithm, that trawls our behaviour and calculates the likelihood of our interest in certain content, the relevance of posts, the relationship between two or more users and what they share in common.

There is a profit motive in play here to get us to spend more time on the app so the machine can serve more ads. It’s understandable but will it work?

You would have thought Instagram would base its non-chron decision on at least some input from users.

E.g. Do you:

a.want Instagram to pre-judge the content we think you’ll like?

b.serve you what your friends and followers want you to see?

But no, all the official blog, presumably not written by a machine, will tell us  is “to improve your experience, your feed will soon be ordered to show the moments we believe you will care about the most.”

The key here is the ‘we believe you will care about the most‘ and to coin my brilliant grandfather, ‘I may not know a lot about art, but I know what I like.’

Anyone who has tried and failed to get traction with a message through social media will know that nuance is everything and it’s much easier to turn off than on.

The wisdom of crowds is a powerful thing, peer pressure, collective sensibility; it’s the reason there aren’t more people picking their noses in public.

So why then is Instagram not picking our brains by engaging us in a, you know, conversation to ask us how we think our social experience could be improved?

They’re doing it to keep us engaged longer so they can squirt more advertising messages into our soon-not-to-be-so-live Instagram feeds.

The risk is that pre-curating our social content will have the opposite effect on engagement and will drive it down, especially in the youth market who can spot a faker from a thousand emojis. Then the dad-dancing Facebook audience overspills into Instagram, clears the room so that all you’ve got left is a 45-year-old MAMIL hand-jiving around a chair, with a robot to Keane.

I’ll leave the last word to a Tumblr blogger, called Maribellum:

Guys, they’re gonna Facebook the shit out of the chronology on Instagram. When will they learn this system is so unnecessary?!? After this there will be no point in posting on Instagram just like there’s no point to Facebook.” #InstagramDon’tBeStupid


9 Mar 2016

Is it ‘time to die’ for print?

Last week’s launch of a print-only newspaper excited journalists – but is it sustainable in a digitised consumer world?

When I started my newspaper career in the late nineties ‘digital’ was called ‘the bloody internet’.

“Forget about the bloody internet,” my ink-stained tabloid boss would argue, before resting his case on some surreal, qualitative Hollywood evidence: “What’s Harrison Ford reading at the beginning of Blade Runner, eh? I’ll tell you – he’s reading a newspaper, he’s not surfing the bloody internet!”

Ink will always course through the veins of the newspaper industry and so too thankfully will the instinct to hunt and tell compelling stories.

New Day, the duck-egg blue-top, is a defiant yell at the bloody internet and aims to tell the biggest stories of the day, solely for a print-friendly audience.

Life’s short, let’s live it well, is the slogan of the new tabloid newspaper, aimed, in between hyphens, at 35-55-year-old, time-poor, glass-half-full people, according to editor Alison Phillips.

We won’t know for sure about the sales figures until May’s ABC report but speculation is febrile in the journalist community that sales were dropping by the third day at a rate of 4%, based on a dip in sales from 153,000 last Tuesday to 148,000 by the end of its first Wednesday, according to unofficial industry estimates in The Guardian.

The fast-paced title may not be shooting to stardom with its audience just yet but it has surely beaten all records for the shortest time taken to get its own Private Eye column.

The helium-fused, relentlessly upbeat tone of New Day’s 40 pages conjures Armando Iannucci and Chris Morris’s The Day Today, bandying about preposterous labels like ‘sportopinioneers‘ on its back pages.

New Day has been inspired by market research into lapsed newspaper buyers, 1 million of them, according to Trinity Mirror chief executive, Simon Fox.

The research, he maintains, shows that although media consumption habits are driven by digital and social media-distributed news, there is a sizeable bunch of readers out there who still crave a printed news product, in between Facebook Like-athons.

However, media consumption is a two-, sometimes three-way street in 2016 – a conversation between content provider, advertiser and consumers, not always inclusively.

Connecting with the reader through social, mobile and online touchpoints is the single-most brilliant opportunity for media owners to stem the haemorrhaging flow of advertising losses, £121m, all told, in 2015.

The latest research from the Ericsson Consumerlab shows it’s not just advertisers and media owners who benefit from a deeper digital dialogue.

Ericsson tracked 100,000 consumers and their media journeys to reveal an emerging habit called place and device shifting. It’s about individuals keeping focused on the same piece of content best suited to their location, across TV, iPad, mobile and portable devices.

The research revealed that media consumers were lost in the jungle of stuff to watch. Within this dilemma comes a plethora of personal data issues. Mediacells calls it the Doh! And the Woh! effects.

Obvious Netflix recommendations e.g. ‘We recommend you watch Series 4 of House of Cards, based on your watching House of Cards Series 3′. Doh! Or the creepier suggestions which are spookily accurate and resonate with consumers, but not in a good way. Woh!

Mediacells sees this as the main opportunity for publishers, broadcasters, brands and agencies to interact in a transparent, meaningful way with their audiences, fans, customers and users. It’s about creating a kind of personal data quid pro quo with the audience providing personal information to get to right content for them, where and when they want it.

The opt-in mechanic in this dialogue empowers the consumer and deepens their trust in selected brands.

Ericsson claims 1 in 4 consumers are willing to provide more personal data in order to receive greater accuracy around recommendations on what media to consume and where.

This fresh dialogue relies on behavioural tracking as well as up to date, accurate personal information which is actively, knowingly offered by the audience.

The research which informed Trinity Mirror to fill the gap created by the exiting Independent has left in the printed newspaper market is like the Turbo Start function on Mario Kart  – it will fuel and energise players but only for a short distance.

The level of consumer insight that advertisers and consumers are increasingly demanding can only be achieved with distributed connections through, say, a newspaper’s website, app, social media accounts and its print products too.

I often wanted to counter my old tabloid boss by asserting that Harrison Ford might have been reading a newspaper at the beginning of Blade Runner but there was no newspaper to be seen by the end; only carnage. But then he would have ripped me a new orifice, digitally. @BradCRees