Pretending that opaque, error-prone ML is part of the grand, romantic quest to find artificial intelligence is an attempt to distract us from the truth.
One of the most useful texts for anyone covering the tech industry is George Orwell’s celebrated essay, Politics and the English Language. Orwell’s focus in the essay was on political use of the language to, as he put it, “make lies sound truthful and murder respectable and to give an appearance of solidity to pure wind”. But the analysis can also be applied to the ways in which contemporary corporations bend the language to distract attention from the sordid realities of what they are up to.
The tech industry has been particularly adept at this kind of linguistic engineering. “Sharing”, for example, is clicking on a link to leave a data trail that can be used to refine the profile the company maintains about you. You give your “consent” to a one-sided proposition: agree to these terms or get lost. Content is “moderated”, not censored. Advertisers “reach out” to you with unsolicited messages. Employees who are fired are “let go”. Defective products are “recalled”. And so on.
At the moment, the most pernicious euphemism in the dictionary of double-speak is AI, which over the last two or three years has become ubiquitous…
The most interesting development of the week had nothing to do with Facebook or even Google losing its appeal against a €2.4bn fine from the European commission for abusing its monopoly of search to the detriment of competitors to its shopping service. The bigger deal was that DeepMind, a London-based offshoot of Google (or, to be precise, its holding company, Alphabet) was moving into the pharmaceutical business via a new company called Isomorphic Labs, the goal of which is grandly described as “reimagining the entire drug discovery process from first principles with an AI-first approach”.
Since they’re interested in first principles, let us first clarify that reference to AI. What it means in this context is not anything that is artificially intelligent, but simply machine learning, a technology of which DeepMind is an acknowledged master. AI has become a classic example of Orwellian newspeak adopted by the tech industry to sanitise a data-gobbling, energy-intensive technology that, like most things digital, has both socially useful and dystopian applications.
That said, this new venture by DeepMind seems more on the socially useful side of the equation. This is because its researchers have discovered that its technology might play an important role in solving a central problem in biology, that of protein folding.
Proteins are large, complex molecules that do most of the heavy lifting in living organisms…
The apparent plausibility of GPT-3’s performance has led – again – to fevered speculation about whether this means we have taken a significant step towards the goal of artificial general intelligence (AGI) – ie, a machine that has the capacity to understand or learn any intellectual task that a human being can. Personally, I’m sceptical. The basic concept of the GPT approach goes back to 2017 and although it’s a really impressive achievement to be able to train a system this big and capable, it looks more an incremental improvement on its predecessors rather than a dramatic conceptual breakthrough. In other words: start with a good idea, then apply more and more computing power and watch how performance improves with each iteration.
Which raises another question: given that this kind of incremental improvement is made possible only by applying more and more computing power to the problem, what are the environmental costs of machine-learning technology?
The easing of the lockdown on July 4 has had its predictable effect — alarming rises in numbers of new infections in many parts of country. These have now reached more than 4,000 new cases a day, attributed by the head of the government’s track-and-trace operation to social-distancing rules being “routinely flouted“ in virus hotspots.
Nothing in this is surprising. People are desperate to get back to some kind of normal behaviour — hugging friends and family, meeting, drinking, dancing, going to clubs, all the things they used to do. What everybody finds hard to realise, still less to accept, is that that ‘normal’ to which we long to return is no longer available. That train has left the station. The pre-pandemic past is indeed a different country.
When the virus first reached these shores, I had a conversation with a member of my family who saw it as just another kind of flu — more dangerous, certainly, but something essentially familiar. I tried — and failed — to persuade her that it was much more significant and far-reaching than that. Reflecting on the conversation afterwards, I thought that the analogy I should have used was that of the First World War — in the sense that the world post-1918 was unrecognisably different from the world as it was in 1913. And, as the depth and reach of the Coronavirus became clearer with every passing day, that seemed to be quite a persuasive analogy.
But actually that still doesn’t get the measure of the change that we are now living though. The most fundamental change that we — humankind — will have to accept is in our conception of our relationship with nature. This thought was sparked by reading “From The Anthropocene To The Microbiocene“, a long essay by Tobias Rees in Noema magazine, a publication of the Berggruen Institute.
The thrust of the essay is that from Aristotle to Thomas Hobbes we humans thought of ourselves as part of nature — as just animals with a capacity for reason. But with Hobbes, we started to think of ourselves as apart from the natural world (where lives were famously “nasty, brutish and short”). And this distinction was steadily reinforced by the rise of science, the Enlightenment , capitalism, democratic politics, and so on. Nature was something that we could master, control and exploit (and despoil). As it happened, this hubristic belief in our intrinsic superiority was ultimately going to be our downfall as the pursuit of economic growth led to the collapse of the biosphere on which human life depends.
The significance of the Coronavirus, on this view, is that it interrupts our inexorable rush to climate catastrophe by reminding us of the extent to which our post-Hobbesian hubris was a delusion. We find ourselves unable to overcome and control this manifestation of part of the natural world. And getting a vaccine will not solve it, though it may make living with it more manageable. But these viruses are part of the human future from now on. They’re here to stay.
All of which means that our view of nature as something separate from us, was delusional. What we have to learn to accept is that we’re part of nature too. Given that we’ve had 400+ years of believing something very different, it’s not surprising that people are finding it difficult to come to terms with what lies ahead. There might be many lockdowns ahead until that penny finally drops.
Since we can’t beat nature, shouldn’t we be thinking of (re)joining it?
At last, the tech titans’ nerd immunity shows signs of fading
My OpEd piece in today’s Observer on last Wednesday’s Congressional Hearings on Big Tech.
The most striking thing about Wednesday’s congressional interrogation of the leaders of Apple, Google, Facebook and Amazon was the absence of deference to the four moguls. This was such a radical departure from previous practice – characterised by ignorance, grandstanding and fawning on these exemplars of the American Way – that it was initially breathtaking. “Our founders would not bow before a king,” said the House antitrust subcommittee chairman, David Cicilline, in his opening remarks. “Nor should we bow before the emperors of the online economy.”
If we wanted a radical departure from the legislative slumber of previous decades, this looked like it. And indeed, to a large extent, it was. One saw it, for example, in the aggressiveness of the questioning by the Democrats. At times, one was reminded of the proceedings of the US supreme court, where the justices constantly interrupt the lawyers before them to cut off any attempt at lawyerly exposition. The implicit message is: “We’ve done our homework. Now get to the point – if you have one.” It was like that on Wednesday.
The Democrats had done their homework: they had read the torrents of private emails that the subcommittee had subpoenaed. And, like any good prosecutor, they never asked a question to which they didn’t already know the answer.
The tech titans were mostly flummoxed by this approach…
This blog is also available as a daily email. If you think this might suit you better, why not subscribe? One email a day, delivered to your inbox at 7am UK time. It’s free, and there’s a one-click unsubscribe if you decide that your inbox is full enough already!
New customers fill seats at Barcelona opera house concert
To mark the re-opening of Barcelona’s Gran Teatre del Liceu opera house, the UceLi Quartet played a livestreamed performance of Puccini’s I Crisantemi (Chrysanthemums).
Well, even if you can, it makes cover-art designers (justifiably) cross.
Waterstones, the (excellent) bookselling chain, has offered its apologies to book designers after some newly reopened branches began displaying books back to front so browsers could read the blurb without picking them up.
It was understandable but slightly “heartbreaking”, said designer Anna Morrison, who mainly designs covers for literary fiction, said she could see why it was happening, but it was still “a little sad”.
“There is a real art to a book cover. It can be a real labour of love and it is a bit disappointing to think our work is being turned round.”
She’s right. One of the joys of going into a bookshop is the blaze of colour and artwork on book covers that confronts you.
After a handful of outdoor companies like North Face, REI and Patagonia said they would stop advertising on Facebook and Instagram last week, several other advertisers have joined the movement, including Ben & Jerry’s, Eileen Fisher, Eddie Bauer, Magnolia Pictures, Upwork, HigherRing, Dashlane, TalkSpace and Arc’teryx.
Heavyweights in the ad industry have also begun pressing marketers to pull their dollars.
On Tuesday, Marc Pritchard, chief brand officer at Procter & Gamble, one of the largest advertisers in the country, threatened to pull spending if platforms didn’t take “appropriate systemic action” to address hate speech.
In an email to clients obtained by the Wall Street Journal last Friday, 360i, a digital-ad agency owned by global ad holding group Dentsu Group Inc., urged its clients to support the ad boycott being advocated by civil rights groups.
I’m sorry to say this, but it looks to me just like virtue-signalling. Just like all the sudden corporate support for “our brilliant NHS” when the Coronavirus panic started in the UK. Facebook’s targeted advertising system is just too useful to companies to be dropped.
Wrongfully Accused by an Algorithm
This seems to be the first case of its kind, but it’s the canary in the mine as far as those of us who regard facial recognition technology as toxic.
On a Thursday afternoon in January, Robert Julian-Borchak Williams was in his office at an automotive supply company when he got a call from the Detroit Police Department telling him to come to the station to be arrested. He thought at first that it was a prank.
An hour later, when he pulled into his driveway in a quiet subdivision in Farmington Hills, Mich., a police car pulled up behind, blocking him in. Two officers got out and handcuffed Mr. Williams on his front lawn, in front of his wife and two young daughters, who were distraught. The police wouldn’t say why he was being arrested, only showing him a piece of paper with his photo and the words “felony warrant” and “larceny.”
His wife, Melissa, asked where he was being taken. “Google it,” she recalls an officer replying.
The police drove Mr. Williams to a detention center. He had his mug shot, fingerprints and DNA taken, and was held overnight. Around noon on Friday, two detectives took him to an interrogation room and placed three pieces of paper on the table, face down.
“When’s the last time you went to a Shinola store?” one of the detectives asked, in Mr. Williams’s recollection. Shinola is an upscale boutique that sells watches, bicycles and leather goods in the trendy Midtown neighborhood of Detroit. Mr. Williams said he and his wife had checked it out when the store first opened in 2014.
The detective turned over the first piece of paper. It was a still image from a surveillance video, showing a heavyset man, dressed in black and wearing a red St. Louis Cardinals cap, standing in front of a watch display. Five timepieces, worth $3,800, were shoplifted.
“Is this you?” asked the detective.
The second piece of paper was a close-up. The photo was blurry, but it was clearly not Mr. Williams. He picked up the image and held it next to his face.
“No, this is not me,” Mr. Williams said. “You think all black men look alike?”
Mr. Williams knew that he had not committed the crime in question. What he could not have known, as he sat in the interrogation room, is that his case may be the first known account of an American being wrongfully arrested based on a flawed match from a facial recognition algorithm, according to experts on technology and the law.
Mr Williams had a cast-iron alibi, but the Detroit police couldn’t be bothered to check
He has since figured out what he was doing the evening the shoplifting occurred. He was driving home from work, and had posted a video to his private Instagram because a song he loved came on — 1983’s “We Are One,” by Maze and Frankie Beverly. The lyrics go:
I can’t understand
Why we treat each other in this way
Taking up time
With the silly silly games we play
Imagine a world where this stuff is everywhere, where you’re always in a police line-up.
The history of inquiries into race riots
A sobering (and depressing) piece by the Harvard historian Jill Lepore in the New Yorker.
TL;DR? (In case you’re busy, here’s the gist.)
In a 1977 study, “Commission Politics: The Processing of Racial Crisis in America,” Michael Lipsky and David J. Olson reported that, between 1917 and 1943, at least twenty-one commissions were appointed to investigate race riots, and, however sincerely their members might have been interested in structural change, none of the commissions led to any. The point of a race-riot commission, Lipsky and Olson argue, is for the government that appoints it to appear to be doing something, while actually doing nothing.
It’s the old, old story. What’s the betting the same thing will happen with Boris Johnson’s “cross-government inquiry into all aspects of racial inequality in the UK”?
Lepore’s is a fine piece, well worth reading in full. Thanks to David Vincent for alerting me to it.
Segway, the most hyped invention since the Macintosh, ends production
Very good report on what once looked like a great idea, but one that never caught on. Segways were very useful for TV cameramen and camerawomen covering golf tournaments, though.
My main regret is that I never managed to try one.
This blog is also available as a daily email. If you think this might suit you better, why not subscribe? One email a day, delivered to your inbox at 7am UK time. It’s free, and there’s a one-click unsubscribe if you decide that your inbox is full enough already!
In his letter, Mr Krishna said that “IBM no longer offers general-purpose IBM facial recognition or analysis software” and “firmly opposes and will not condone uses of any technology, including facial recognition technology offered by other vendors, for mass surveillance, racial profiling, violations of basic human rights and freedoms, or any purpose which is not consistent with our values and principles of trust and transparency. We believe now is the time to begin a national dialogue on whether and how facial recognition technology should be employed by domestic law enforcement agencies.”
Amen to that. No sooner had the letter been released than cynics and sceptics were poring over it for the get-out clause. IBM was never a big player in the facial recognition game, said some, and so it’s no sacrifice to exit it: to them, Krishna’s letter was just “virtue- signalling”. Yet two days later Amazon heard the signal and announced a one-year suspension of police force use of its Rekognition facial recognition software – they say they’d like Congress to pass stronger regulation around it.
The IBM announcement and now Amazon’s are a big deal. Just ponder their significance for a moment…
And now Microsoft has joined the rush to paint a line between the company and the toxic tech. To be fair, their lawyer Brad Smith has been calling for regulation of the technology for quite a while.
With time on his hands, the Observer‘s restaurant critic turns chef
Hilarious, beautifully-written piece. I loved this bit in particular…
I decided I needed something more challenging, because I am stupid, and don’t know when to quit. The soufflé suissesse has been on the menu at Le Gavroche since about 1968. According to Michel Roux Jr, who took over from his father Albert in 1993, it’s been lightened over the years. This is shocking, because, to make four servings, the current recipe (in Le Gavroche Cookbook) calls for six eggs, 600ml of double cream, 500ml of milk, 200g of gruyère, a slab of butter, a defibrillator and a priest standing by to administer last rites.
My devout mother would have held that the last was the essential ingredient.
Worth reading in full.
Signal Downloads Are Way Up Since the Protests Began
The week before George Floyd died on May 25, about 51,000 first-time users downloaded Signal, according to data from the analytics firm Sensor Tower. The following week, as protests grew nationwide, there were 78,000 new downloads. In the first week of June, there were 183,000. (Rani Molla at Recode noted that downloads of Citizen, the community safety app, are also way up.)
Organizers have relied on Signal to devise action plans and develop strategies for handling possible arrests for several years. But as awareness of police monitoring continues to grow, protest attendees are using Signal to communicate with friends while out on the streets. The app uses end-to-end encryption, which means each message is scrambled so that it can only be deciphered by the sender and the intended recipient.
Signal has also already been tested. In 2016, the chat service withstood a subpoena request for its data. The only information it could provide was the date the accounts in question were created and when they had last used Signal. Signal does not store messages or contacts on its servers, so it cannot be forced to give copies of that information to the government.
It’s a terrific app, which has got a lot better over time. Think of it as WhatsApp for serious people who don’t trust Facebook.
This blog is also available as a daily email. If you think this might suit you better, why not subscribe? One email a day, delivered to your inbox at 7am UK time. It’s free, and there’s a one-click unsubscribe if your decide that your inbox is full enough already!
Infuriated by the impunity with which Dominic Cummings was able to flout the lockdown rules?
If you are a UK voter and have a Tory MP, why not write to him or her letting know how you feel about Cummings’s impunity and Boris Johnson’s support for it?
It’s simple to do: just go to the MySociety Write to Them and the site will check your MP’s identity by your postcode and set up a form for composing and dispatching a suitable message to him or her.
I’ve just done it. It’s called giving feedback.
The benefits of taciturnity
Portrait of Forster by Dora Carrington, oil on canvas, 1920.
In Madrid the other week a literary journalist told me the following joke. A man goes into a pet shop and sees three parrots side by side, priced at $1000, $2000 and $3000. ‘Why does that parrot cost $1000?’ he asks the owner. ‘Because it can recite the whole of the Bible in Spanish,’ comes the reply. ‘And why does that one cost $2000?’ ‘Because it can recite the whole of the Bible in English and in Spanish.’ ‘And the one that costs $3000, what does he recite?’ ‘Oh, he doesn’t say a word,’ explains the pet shop owner: ‘but the other two call him Maestro.’
This made me think, naturally enough, of E.M. Forster; and then of the fact that we were about to undergo the annual garrulity of the Booker Prize for Fiction.
Reminds me of that old adage of Abraham Lincoln’s: it’s better to keep one’s mouth shut and be thought a fool, than to open it and remove all room for doubt.
btw: When I was a student I went to E.M. Forster’s 90th birthday party in King’s College, Cambridge in January 1969. When I tell people that, they check for the nearest exit, and when I tell them that the party was hosted by Francis Crick of DNA fame, they really run for cover. But it’s true: I was a member of the Cambridge Humanists and they held the party for him. Crick was at the time the Chairman of the Humanist society.
Remind me to tell you about the Boer War, sometime …
New life and an awareness of mortality
Kara Swisher has a baby daughter — at the age of 57. Don’t know how that happened, but she’s written about the differences it has made to her life under lockdown:
I am at the highest risk of our little quarantine group, as my 15-year-old has pointed out to me more than once. I assume it is his way of whistling past the grave in hopes that the grave does not whistle back.
But whistle it does, sometimes softly, like when I had a life-threatening stroke on a long-haul trip to China five years back, or more loudly, like when my father died unexpectedly more than 50 years ago from an aneurysm at 34 years old, at the start of what should have been a brilliant long life with his three children.
That is why I am thinking more often of math. Each of us has an exact number — whether it is of years, days, minutes or seconds. We don’t know our number, but it helps to keep in mind that this number exists.
I’m now more aware that our time here is finite. So I take an extra minute I might not have before watching my sons play with their new sister at the dinner table. It is a love that I did not expect to jell so quickly and so perfectly. My sons, with their phones down, are clapping their hands, making faces and doing anything they can to delight my daughter into yet another magnificent smile. Luckily for us, she is an endless font of those.
Covid is messing with machine-learning systems
You know those ‘recommender’ systems that tell you what you might be interested in based on your browsing or purchase history? Well, it turns out that the poor dears are mightily confused by our ‘weird’ behaviour during the pandemic. For example, once upon a time the top 100 searches on Amazon, say, would be mostly for gadgets — iPhone cases, battery packs, SSDs, etc. etc. And machine-learning systems trained on these searches have traditionally been good at extracting the trends from those patterns.
And then all of a sudden everybody is interested in quite different things. “In the week of April 12-18”, says an interesting Tech Review article by Will Douglas Heaven,
the top 10 search terms on Amazon.com were: toilet paper, face mask, hand sanitizer, paper towels, Lysol spray, Clorox wipes, mask, Lysol, masks for germ protection, and N95 mask. People weren’t just searching, they were buying too—and in bulk. The majority of people looking for masks ended up buying the new Amazon #1 Best Seller, “Face Mask, Pack of 50”.
What’s happening is that machine-learning systems trained on normal (i.e.pre-pandemic) human behavior are now finding that ‘normal’ has changed, and some are no longer working as they should.
But machine-learning isn’t just used for recommendations. Mr Heaven found a company in London, Phrasee ( Motto: “Empower your Brand with AI-Powered Copywriting”), which uses natural-language processing and machine learning to generate email marketing copy or Facebook ads on behalf of its clients.
Making sure that it gets the tone right is part of its job. Its AI works by generating lots of possible phrases and then running them through a neural network that picks the best ones. But because natural-language generation can go very wrong, Phrasee always has humans check what goes into and comes out of its AI.
When covid-19 hit, Phrasee realized that more sensitivity than usual might be required and started filtering out additional language. The company has banned specific phrases, such as “going viral,” and doesn’t allow language that refers to discouraged activities, such as “party wear.” It has even culled emojis that may be read as too happy or too alarming. And it has also dropped terms that may stoke anxiety, such as “OMG,” “be prepared,” “stock up,” and “brace yourself.” “People don’t want marketing to make them feel anxious and fearful—you know, like, this deal is about to run out, pressure pressure pressure,” says Parry Malm, the firm’s CEO.
If, like me, you are sceptical about the claims made for machine-learning technology, this kind of thing will be music to your ears. Though I doubt if the Spotify system that thinks it knows my musical tastes has made the necessary adjustment yet.
This blog is also available as a daily email. If you think this might suit you better, why not subscribe? One email a day, delivered to your inbox at 7am UK time. It’s free, and there’s a one-click unsubscribe if your decide that your inbox is full enough already!
Using AI to find candidates for trying against COVID —
Er, for “AI” read machine learning. Usual mistake, but interesting nevertheless.
A team at BenevolentAI, a UK company that uses machine learning to aid drug discovery, had been searching through their database of all existing, approved drugs, searching for one that could be repurposed to treat the novel coronavirus. And according to this report they found one in just three days.
“Most drug companies had been looking at antiviral drugs, but we approached it from the other end and looked at what processes used by the virus could be disrupted,” said Peter Richardson, vice president of pharmacology at the company.
Protein kinases — enzymes that speed up chemical reactions in the body — seemed a promising area to look into. Some of these regulate the way substances can enter human cells — disrupt them, and the virus might be unable to get into the lung, heart and kidney cells it has been so prone to invading.
Baricitinib, a drug developed by Eli Lilley and approved in 2018, stood out because it not only inhibited kinases but also prevented the cytokine storms — the body’s own extreme autoimmune reactions that have led to so many fatalities with Covid-19. It was also likely to be compatible with other drugs being used to treat the disease, such as remdesivir. Richardson and a team of three part-time researchers identified an initial 370 kinase inhibitors, and then narrowed it down to six that looked most likely to work.
“It validated using AI for this kind of problem,” says Richardson. “It would have been impossible for the four of us to do it at that speed otherwise. If you took 250 people you still couldn’t do it at that pace because there would be too many competing ideas. You really can’t do it without an organised knowledge graph and the ability to query it.”
Interesting. I suppose they had to describe it as AI, given that the letters appear in the firm’s name. Benevolent Machine Learning doesn’t have the same ring to it.
How coronavirus almost brought down the global financial system
Another amazing long read from Adam Tooze, this time about how close the world came to a financial meltdown because of the Coronavirus. Most of it stuff I hadn’t known or understood. Tooze is a really phenomenal historian, with an astonishing grasp of how the finance industry works. * Crashed: How a Decade of Financial Crises Changed the World*, his history of the 2008 banking crisis, is terrific. And now he seems to be really on top of the Coronavirus crisis. I’ve been thinking that what we’re facing at the moment is what the world would have been like if the Spanish flu and the Great Depression had come together.
This essay, which is worth reading in full (requires a cup of coffee and some peace and quiet) is mainly about how the central bankers of the West succeeded — just — in avoiding a global meltdown. But it ain’t over yet. And most poor countries don’t have the resources — financial or professional — to deal with the virus.
When I think about how COVID-19’s security measures are affecting organizational networks, I see several interrelated problems:
One, employees are working from their home networks and sometimes from their home computers. These systems are more likely to be out of date, unpatched, and unprotected. They are more vulnerable to attack simply because they are less secure.
Two, sensitive organizational data will likely migrate outside of the network. Employees working from home are going to save data on their own computers, where they aren’t protected by the organization’s security systems. This makes the data more likely to be hacked and stolen.
Three, employees are more likely to access their organizational networks insecurely. If the organization is lucky, they will have already set up a VPN for remote access. If not, they’re either trying to get one quickly or not bothering at all. Handing people VPN software to install and use with zero training is a recipe for security mistakes, but not using a VPN is even worse.
Four, employees are being asked to use new and unfamiliar tools like Zoom to replace face-to-face meetings. Again, these hastily set-up systems are likely to be insecure.
Five, the general chaos of “doing things differently” is an opening for attack. Tricks like business email compromise, where an employee gets a fake email from a senior executive asking him to transfer money to some account, will be more successful when the employee can’t walk down the hall to confirm the email’s validity — and when everyone is distracted and so many other things are being done differently.
Worrying about network security seems almost quaint in the face of the massive health risks from COVID-19, but attacks on infrastructure can have effects far greater than the infrastructure itself.
After the analogue hammer, comes the data-driven dance.
“Coronavirus has reminded even the most conservative among us that there is a role for the state after all. No government can outsource their way through this test. Suddenly, the absence of data skills at the centre of government is a life and death issue. The hammer blows will decrease. As the dance begins, states must respond with agility, using public and private data. An era of central data units may emerge. Regulation for data registries and more powerful registrars seems certain as public trust in government data and a new locus for privacy and surveillance are all being tried and tested on a daily basis. This is one big A/B test for governments, whether democratic or autocratic. This may not be the internet founders’ much longed-for government 2.0 moment, but we are all in beta now.
The “hammer and the dance” metaphor is becoming a meme.
Why content moderators should be designated as key workers
Important paper from the Turing Institute arguing that, just now, the people who try to keep mis- and disinformation off social media should be regarded as part of the world’s critical infrastructure.
The current crisis surrounding COVID-19 has scaled up the challenge of content moderation, severely reducing supply and massively increasing demand. On the “supply side”, content moderators have, like other workers around the world, been told not to come into work. YouTube has already warned that, as a result, it will conduct fewer human reviews and openly admits it may make poor content takedown decisions.
On the “demand side”, the growth of the pandemic has seen an upsurge in the amount of time spent online. BT recently noted an increase in UK daytime traffic of 35-60%, and social networks report similar increases, particularly in their use for education, entertainment and even exercise. Sadly, harmful activity has increased too: Europol reports “increased online activity by those seeking child abuse material” and the World Health Organisation has warned of an emerging “infodemic” of pernicious health-related disinformation. Recently, concerns have been raised that false claims are circulating online about the role of 5G.
At a time when social media is desperately needed for social interaction, a widening gap is emerging between how much content moderation we need and how much can be delivered. As a result, AI is being asked do tasks for which it is not ready, with profound consequences for the health of online spaces. How should platforms, governments, and civil society respond to this challenge? Following Rahm Emmanuel’s exhortation to “never let a crisis go to waste,” we argue that, now that the challenges in content moderation have been exposed by the pandemic, it is time for a reset.
his blog is now also available as a once-a-day email. If you think this might work better for you why not subscribe here? (It’s free and there’s a 1-click unsubscribe if you subsequently decide you need to prune your inbox!) One email a day, in your inbox at 07:00 every morning.
…As a consequence, a powerful technology with great potential for good is at the moment deployed mainly for privatised gain. In the process, it has been characterised by unregulated premature deployment, algorithmic bias, reinforcing inequality, undermining democratic processes and boosting covert surveillance to toxic levels. That it doesn’t have to be like this was vividly demonstrated last week with a report in the leading biological journal Cell of an extraordinary project, which harnessed machine learning in the public (as compared to the private) interest. The researchers used the technology to tackle the problem of bacterial resistance to conventional antibiotics – a problem that is rising dramatically worldwide, with predictions that, without a solution, resistant infections could kill 10 million people a year by 2050.
Another notion of bias, one that is highly relevant to my work, are cases in which an algorithm is latching onto something that is meaningless and could potentially give you very poor results. For example, imagine that you’re trying to predict fractures from X-ray images in data from multiple hospitals. If you’re not careful, the algorithm will learn to recognize which hospital generated the image. Some X-ray machines have different characteristics in the image they produce than other machines, and some hospitals have a much larger percentage of fractures than others. And so, you could actually learn to predict fractures pretty well on the data set that you were given simply by recognizing which hospital did the scan, without actually ever looking at the bone. The algorithm is doing something that appears to be good but is actually doing it for the wrong reasons. The causes are the same in the sense that these are all about how the algorithm latches onto things that it shouldn’t latch onto in making its prediction.
Addressing bias in algorithms is crucial, especially in domains like healthcare where accurate predictions are vital. One effective approach to recognizing and mitigating biases is to rigorously test the algorithm in scenarios similar to its real-world applications. Suppose a machine-learning algorithm is trained on data from specific hospitals to predict fractures from X-ray images. In this case, it may appropriately incorporate prior knowledge about patient populations in those hospitals, resulting in reliable predictions within that context. However, the challenge arises when the algorithm is intended to be used in different hospitals not present in the initial training data set. To avoid unintended biases, a robust evaluation process is essential, and the use of a mobile learning management system can prove beneficial. Such a system enables continuous monitoring and assessment of the algorithm’s performance across various hospital settings, ensuring it doesn’t latch onto irrelevant factors and provides accurate predictions based on genuine medical insights.
To recognize and address these situations, you have to make sure that you test the algorithm in a regime that is similar to how it will be used in the real world. So, if your machine-learning algorithm is one that is trained on the data from a given set of hospitals, and you will only use it in those same set of hospitals, then latching onto which hospital did the scan could well be a reasonable approach. It’s effectively letting the algorithm incorporate prior knowledge about the patient population in different hospitals. The problem really arises if you’re going to use that algorithm in the context of another hospital that wasn’t in your data set to begin with. Then, you’re asking the algorithm to use these biases that it learned on the hospitals that it trained on, on a hospital where the biases might be completely wrong.
There is, alas, no such thing as a free lunch. This simple and obvious truth is invariably forgotten whenever irrational exuberance teams up with digital technology in the latest quest to “change the world”. A case in point was the bitcoin frenzy, where one could apparently become insanely rich by “mining” for the elusive coins. All you needed was to get a computer to solve a complicated mathematical puzzle and – lo! – you could earn one bitcoin, which at the height of the frenzy was worth $19,783.06. All you had to do was buy a mining kit (or three) from Amazon, plug it in and become part of the crypto future.
The only problem was that mining became progressively more difficult the closer we got to the maximum number of bitcoins set by the scheme and so more and more computing power was required. Which meant that increasing amounts of electrical power were needed to drive the kit. Exactly how much is difficult to calculate, but one estimate published in July by the Judge Business School at the University of Cambridge suggested that the global bitcoin network was then consuming more than seven gigwatts of electricity. Over a year, that’s equal to around 64 terawatt-hours (TWh), which is 8 TWh more than Switzerland uses annually. So each of those magical virtual coins turns out to have a heavy environmental footprint.
At the moment, much of the tech world is caught up in a new bout of irrational exuberance. This time, it’s about machine learning, another one of those magical technologies that “change the world”…
