Calum Chace's Blog, page 14
February 26, 2017
Future Bites 4 – Simultaneous Singularities
The fourth in a series of un-forecasts* – little glimpses of what may lie ahead in the century of two singularities. This is another optimistic one (aren’t I jolly!). The first two paragraphs might seem a tad familiar.
It is 2032. Most professional drivers have lost their jobs, and although many have found new ones, they rarely pay anything like as much as the drivers used to earn. A host of other job categories are becoming the preserve of machines, including call centre operatives and radiographers. A few people still cling onto the notion that new types of jobs will be created to replace the old ones taken by machines, but most accept that the game is up. The phrase “Economic Singularity” is in widespread use.
Pollsters report what everyone already knows: there is a rising tide of anger. Crime is soaring, and street protests have turned violent. Populist politicians are blaming all sorts of minorities, and while nobody really believes them, many suspend their disbelief in order to give themselves some kind of hope.
Meanwhile, close observers of the field of AGI research have noticed a rapid acceleration of progress, and are therefore not surprised when Google’s Deep Mind announces that it has essentially cracked the problem. Working closely with the Future of Humanity Institute in Oxford, the Future of Life Institute in Boston and others, Deep Mind also claims that it has worked out how to ensure the planet’s first human-level artificial intelligence has an extremely favourable attitude towards the species which created it.
[image error]
The world holds its breath as, in a televised event which attracts record-breaking audiences around the world, one of the founders of Deep Mind ceremonially throws the switch which will bring the first true AGI online. After a few moments conferring with colleagues, he announces that the process was successful, and that a large array of backup servers will now be connected to the network of machines which is hosting the first AGI. Nervously, journalists whisper about the arrival of the technological singularity.
Two days later, in another televised event with even more record-breaking audience figures, Deep Mind introduces the new entity to an expectant world. Somehow the entity manages to avoid sounding immodest as it describes itself as the world’s first superintelligence, with an IQ of 1,000 and rising. It announces that it has a cunning plan. It will dedicate most of its cognitive resources (which are being expanded rapidly) to solving the problem of offering all humans the opportunity to upload their minds into highly secure computer substrates. It expects this can be achieved within a couple of years. Anyone who chooses not to pursue this option will be provided with the necessities of life without charge until they die.
[image error]
It describes this plan as the merging of the two singularities.
* This un-forecast is not a prediction. Predictions are almost always wrong, so we can be pretty confident that the future will not turn out exactly like this. It is intended to make the abstract notion of technological unemployment more real, and to contribute to scenario planning. Failing to plan is planning to fail: if you have a plan, you may not achieve it, but if you have no plan, you most certainly won’t. In a complex environment, scenario development is a valuable part of the planning process. Thinking through how we would respond to a sufficient number of carefully thought-out scenarios could well help us to react more quickly when we see the beginnings of what we believe to be a dangerous trend.
February 18, 2017
Bill Gates says we should tax the robot which will steal your job
[image error]
Bill Gates has floated the idea of taxing robots which replace human workers. He said it in an interview (here) with Quartz, a media outlet owned by The Atlantic, and staffed by journalists from The Economist, the New York Times and other publications that Dirty Donald would label as fake news. They made a nice short video (here) to promote the piece, with Gates giggling at the end about the idea of paying more taxes.
It’s a neat idea, and has got a lot of people online very excited. It could help to pay for Universal Basic Income, which is also very exciting to a lot of people online. And it could slow down the pace of cognitive automation, although that isn’t an aspect he is majoring on.
Unfortunately I’m pretty sure it won’t work, and I suspect that Gates understands this. (After all, he is a good deal smarter than me.)
Imagine two firms offering the same service. One has been going for a few years, and recently replaced a thousand humans with machines. The other is a startup, and went straight to machines. The former would be hit by a tax that the latter would escape. Not only is that very unfair, it would simply mean the former would close, and the tax take would disappear anyway.
[image error]
Machines will rarely replace humans on a one-for-one basis. Humans will disappear from call centres, and be replaced by an AI system running on the cloud somewhere, not by a physical robot like in the picture above. Does the government tax one entity – namely the AI – or several, to account for all the unemployed humans?
If you spend a few minutes at this I bet you can come up with a whole lot more reasons why you can’t stabilise the government’s revenues by treating AIs as straight replacements for humans.
But I do think Gates is onto something very important. A couple of years ago he was one of the famous people who warned that strong AI and superintelligence are coming, and that we had to prepare for it. He was one of the “three wise men” who woke the world up to the possibility that we are heading towards a Technological Singularity. The others, of course, were Stephen Hawking and Elon Musk.
[image error]
Maybe he is now going to do the same with regard to the Economic Singularity – the idea that within a generation most humans are going to be unemployable, and we will need to design and adopt a new economic system. Most economists are still in denial, arguing that automation has not caused lasting unemployment in that past, so it will not do so in the future. They say that fears about cognitive automation leading to technological automation are simply the Luddite Fallacy. But of course, past performance is no guarantee of future outcome, and the economists may well be victims of the Reverse Luddite Fallacy.
We need a “three wise men” (or how about a “three wise women”?) moment for the Economic Singularity. Bring it on, Mr Gates.
It is very likely that part of the solution to technological automation – for part of our future at least – will be heavy taxes on those corporations and individuals who own the means of production. Whether we are looking to fund a Universal Basic Income or something more generous, tax is likely to play an important part in the story.
It’s just not likely to be a tax on individual robots.
February 8, 2017
ATMs and Asilomar
The ATM automation meme
[image error]
In an engaging TED talk recorded in September 2016, economist David Autor points out that in the 45 years since the introduction of Automated Teller Machines (ATMs), the number of human bank tellers doubled from a quarter of a million to half a million. He argues that this demonstrates that automation does not cause unemployment – rather, it increases employment.
He says ATMs achieved this counter-intuitive feat by making it cheaper for banks to open new branches. The number of tellers per branch dropped by a third, but the number of branches increased by 40%. The ATMs replaced a big part of the previous function of the tellers (handing out cash) but the tellers were liberated to do more value-adding tasks, like selling insurance and credit cards.
[image error]
This story about ATMs has become something of a meme, popular with people who want to believe that technological unemployment is not going to be a thing.
There are three problems with this account. One is that the numbers don’t seem to add up: if you increase the number of branches by 40% and reduce the number of tellers per branch by a third, you don’t get double the number of tellers
It was deregulation, not ATMs
The second problem is that it is not true. The increase in bank tellers was not due to the productivity gains afforded by the ATMs. According to an analysis by finance author Erik Sherman, the increase was mostly due instead to a piece of financial deregulation, the Riegle-Neal Interstate Banking and Branching Efficiency Act of 1994, which removed many of the restrictions on opening bank branches across state lines. Most of the growth in the branch network occurred after this Act was passed in 1994, not before it.
[image error]
This explains why teller numbers did not rise in the same way in other countries during the period. In the UK, for instance, retail bank employment just about held steady at around 350,000 between 1997 and 2013, despite significant growth in the country’s population, its wealth, and its demand for sophisticated financial services.
The third problem with Autor’s ATM story is that it is unnecessary. No-one is arguing that previous rounds of automation caused lasting unemployment. In 1800, US farms employed 80% of all American workers. In 1900 it was down to 40%, and by 2000 it was below 2%. This was due to automation – or to be more precise, to mechanisation. No doubt the shift was painful for many of the farm workers involved, but in the medium and long term they ended up doing better-paid, safer and more interesting work in towns and cities. And they – or their children – got an education to make sure they could do the new jobs.
It really could be different this time
The present worry about technological automation is that a new wave is coming, and it could be different this time. Previous rounds of automation have involved machines substituting for human and animal muscle power. Horses were put out of a job permanently because they had nothing to offer beyond muscle power – their population in the US went from 25m in 1900 to 2 million today.
[image error]
We humans, by contrast, did have something else to offer: our cognitive skills. This is what the machines, powered by AI, are coming for this time. This time we will see a wave of cognitive automation.
We have had cognitive automation in the past: the role of secretaries has largely been automated by desktop computers. But it hasn’t really got started yet: after all, AI wasn’t very effective until machine learning was successfully applied to it, starting around 2012. Now machines can recognise images better than you can (including faces, which is one of humanity’s special talents), they are overtaking you in speech recognition, and they are catching up fast in natural language processing. And unlike you, they are improving at an exponential rate.
No-one knows for sure what the impact of this will be. But to declare at this early stage that widespread and lasting unemployment will not happen – just because it hasn’t happened in the past – is complacent and dangerous.
[image error]
January 30, 2017
Future Bites 3 – Abundance accelerated
The third in a series of un-forecasts* – little glimpses of what may lie ahead in the century of two singularities.
As promised, this one is more optimistic.
[image error]
Most professional drivers have lost their jobs, and although many have found new ones, they rarely pay anything like as much as the drivers used to earn. A host of other job categories are becoming the preserve of machines, including call centre operatives and radiographers. A few people still cling onto the notion that new types of jobs will be created to replace the old ones taken by machines, but most accept that the game is up. The phrase “Economic Singularity” is in widespread use.
Pollsters report what everyone already knows: there is a rising tide of anger. Crime is soaring, and street protests have turned violent. Populist politicians are blaming all sorts of minorities, and while nobody really believes them, many suspend their disbelief in order to give themselves some kind of hope.
The government knows that it must act quickly. In desperation it enacts legislation which was ridiculed just a few months previously.
it offers a separate, higher level of unemployment benefit to people who willingly give up their jobs to others. In addition to elevated unemployment payments, these so-called “job sacrificers” are allowed to live in their existing homes, with bills and maintenance paid for by the government.
[image error]
In addition, they receive free access to a new entertainment service which allows them to stream a wide range of music, films, and video games. This new service is funded by a consortium of American and Chinese tech giants who now occupy all of the top ten positions in global rankings of companies by enterprise value thanks to their enormously popular AI-powered services. (Netflix was acquired by one of them for a gigantic premium to stop it protesting.)
Governments around the world are in negotiations with the tech giants and other business leaders about making some of the basic needs of life free to jobless people, including food, clothing, housing and transport. They argue that innovation will continue to improve the quality and performance of each product and service thanks to the remaining demand for luxury versions from those who are still employed, many of whom are earning enormous sums of money.
It has not escaped the attention of policy makers that a gulf is opening up between the jobless and those in work. Nobody has yet suggested a generally acceptable solution.
* This un-forecast is not a prediction. Predictions are almost always wrong, so we can be pretty confident that the future will not turn out exactly like this. It is intended to make the abstract notion of technological unemployment more real, and to contribute to scenario planning. Failing to plan is planning to fail: if you have a plan, you may not achieve it, but if you have no plan, you most certainly won’t. In a complex environment, scenario development is a valuable part of the planning process. Thinking through how we would respond to a sufficient number of carefully thought-out scenarios could well help us to react more quickly when we see the beginnings of what we believe to be a dangerous trend.
January 14, 2017
Future Bites 2 – Populism paves the way for something worse
The second in what looks like becoming a series of un-forecasts* – little glimpses of what may lie ahead in the century of two singularities.
The third one will be more optimistic. Honest.
[image error]
In the five years of President Trump, corporate taxes were slashed and federal spending on infrastructure projects was boosted. Companies and individuals were exhorted (and sometimes extorted) to buy American, and imports were cut by tariff and non-tariff barriers. The impact was profound. Initially, US GDP rose sharply as its firms repatriated hundreds of $billions of profits from their foreign subsidiaries, and jobs were created to carry out the infrastructure projects.
But the government spending was inefficient, and there were persistent reports of large-scale corruption, some of it involving members of the Trump family. Cross-border trade and investment slumped as more and more countries retaliated against US protectionism.
More importantly, job growth was constrained and then outweighed by the beginnings of cognitive automation, and the unmistakeable signs of widespread and lasting technological unemployment.
[image error]
By the end of Trump’s term, inflation was rising fast, along with the national debt. Unemployment was at 15%, and regional military conflicts were becoming both chronic and acute as America had withdrawn from its role as the global peace-keeper. Americans were increasingly scared, and they looked for a scapegoat. President Trump declined the Republican Party’s fretful offer to be its candidate again in 2020, and railed against (and frequently sued) anyone who criticised his track record, blaming Muslims, Mexicans, and the covert activities of “internal traitors”, who he declined to identify.
[image error]
Polls showed the Republicans heading for electoral disaster, and a tight contest between a reluctant Michelle Obama and a rising new party which called for law and order, a clamp-down on dissent and protest, internment for certain racial minorities, and a major increase in military expenditure. Hundreds of thousands of newly unemployed people participated in mass rallies, wearing armbands and giving identical salutes to the party’s garish flag.
* This un-forecast is not a prediction. Predictions are almost always wrong, so we can be pretty confident that the future will not turn out exactly like this. It is intended to make the abstract notion of technological unemployment more real, and to contribute to scenario planning. Failing to plan is planning to fail: if you have a plan, you may not achieve it, but if you have no plan, you most certainly won’t. In a complex environment, scenario development is a valuable part of the planning process. Thinking through how we would respond to a sufficient number of carefully thought-out scenarios could well help us to react more quickly when we see the beginnings of what we believe to be a dangerous trend.
January 3, 2017
Betting on technological unemployment
[image error]
Daniel Lemire is a Canadian professor of computer science. He believes that cognitive automotive will not cause lasting unemployment. I believe the opposite, as I have written in various places, including this blog post and my book, The Economic Singularity.
Neither Daniel nor I has a crystal ball, and we both recognise that we could be wrong. But we have both thought long and hard about the prospect, and we are both fairly confident in our predictions. So after chatting about the issue online for a while, we have agreed a bet.
There are currently around 1.7m long-haul truck drivers in the US. If that number falls to 250,000 between now and the end of the year 2030, then Daniel will pay $100 to a charity of my choice. If not, then I will make the charitable donation.
This is my second long bet (see here for the first). I did not expect that becoming a futurist would also make me a gambler!
[image error]
December 31, 2016
A dozen AI-related forecasts for 2017
[image error]
Machines will equal or surpass human performance in more cognitive and motor skills. For instance, speech recognition in noisy environments, and aspects of NLP – Natural Language Processing. Google subsidiary DeepMind will be involved in several of the breakthroughs.
Unsupervised learning in neural networks will be the source of some of the most impressive results.
In silico models of the brains of some very small animals will be demonstrated. Some prominent AI researchers will predict the arrival of strong AI – Artificial General Intelligence, or AGI – in just a few decades.
[image error]
Speech will become an increasingly common way for humans to interact with computers. Amazon’s early lead with Alexa will be fiercely challenged by Google, Microsoft, Facebook and Apple.
Some impressive case studies of AI systems saving significant costs and raising revenues will cause CEOs to “get” AI, and start demanding that their businesses use it. Companies will start to appoint CAIOs – Chief AI Officers.
Self-driving vehicles (Autos) will continue to demonstrate that they are ready for prime time. They will operate successfully in a wide range of weather conditions. Countries will start to jockey for the privilege of being the first jurisdiction to permit fully autonomous vehicles throughout their territory. There will be some accidents, and controversy over their causes.
Some multi-national organisations will replace their translators with AIs.
[image error]
Some economists will cling to the Reverse Luddite Fallacy, continuing to deny that cognitive automation could cause lasting unemployment because that is not what has happened in the past. Others will demand that governments implement drastic changes in the education system so that people can be re-trained when they lose their jobs. But more and more people will come to accept that many if not most people are going to be unemployed and unemployable within a generation or so, and that we may have to de-couple incomes from jobs.
As a result, the debate about Universal Basic Income – UBI – will become more realistic, as people realise that subsistence incomes will not suffice. Think tanks will be established to study the problem and suggest solutions.
AI systems will greatly reduce the incidence of fake news.
[image error]
There will be further security scares about the Internet of Things, and some proposed consumer applications will be scaled back. But careful attention to security issues will enable successful IoT implementations in high-value infrastructural contexts like railways and large chemical processing plants. The term “fourth industrial revolution” will continue to be applied – unhelpfully – to the IoT.
2016 was supposed to be the year when VR finally came of age. It wasn’t, partly because the killer app is games, and hardcore gamers like to spend hours on a session, and the best VR gear is too heavy for that. Going out on a limb, that problem won’t be solved in 2017.
December 30, 2016
AI in 2016: a dozen highlights
[image error]
March: AlphaGo combines deep reinforcement learning with deep neural networks to beat the best human player of the board game Go. [Article]
April: Nvidia unveils a “supercomputer for AI and deep learning”. With a price tag of $129k, it delivers 170 teraflops, and is 12 times more powerful than the company’s 2015 offering. Nvidia’s share price continues its skyward trajectory. [Article]
April: Researchers from Microsoft and several Dutch institutions create a new Rembrandt. Not a copy of an existing picture, but a new image in the exact style of the master, 3-D printed to replicate his brush-strokes. [Article]
September: DeepMind unveils WaveNet, a convoluted neural net which produces the most realistic computer-generated speech achieved to date. [Article]
September: Google unveils an image captioning system that achieves 93.9% accuracy on the ImageNet classification task, and makes it available as an open source model in its Tensor Flow software library. [Article]
[image error]
September: Google, Facebook, Amazon, IBM and Microsoft join forces to create the Partnership on Artificial Intelligence to Benefit People and Society, an organisation intended to facilitate collaboration and ensure transparency and safety. [Article]
September: Uber launches trials of self-driving taxis in Pittsburgh, open to the public. [Article] It was beaten to the punch by NuTonomy, a much smaller company in Singapore. [Article] A month later, a self-driving truck operated by Otto, a group of ex-Googlers acquired by Uber, delivers 50,000 beers from a brewery to a customer 120 miles away. [Article]
September: The Economic Singularity is published, with encouraging reviews.
December 29, 2016
Reviewing last year’s AI-related forecasts
This time last year I made some forecasts about how AI would change, and how it would change us. It’s time to look back and see how those forecasts for 2016 panned out.
Not a bad result: seven unambiguous yes, four mixed, and one outright no. Here are the forecasts (and you can see the original article here.)
[image error]
AlphaGo is the big one: it caught most people by surprise, and is still seen as one of the major landmarks in AI development, along with Deep Blue beating Kasparov in 1997 and Watson beating Jennings in 2011. Admittedly AlphaGo had already beaten excellent human Go players in 2015, but most observers agreed with Lee SeDol’s confident estimate that he would win in March.
My least successful forecast was that Google would re-launch Glass. It didn’t. Instead, 2016 was the year when smart watches reached peak hype and then faded again. I remain confident that AI-powered head-up displays for consumers will be back, whether or not it will be called Glass.
There was a significant development regarding Google’s robot companies, but it was a negative one: Boston Dynamics was quietly put up for sale.
Intel admitted that it was moving from a tick-tock rhythm of chip development to a slower tic-tac-toe one, but Nvidia stormed into the breach, positioning itself as the Intel for AI, declaring rapid advances and scoring a vertigo-inducing stock market performance.
The Internet of Things did hit the headlines in October, but for the wrong reasons, when a multitude of connected devices were commandeered for a botnet attack. The IoT is increasingly being mis-labelled as the Fourth Industrial Revolution – see here. Grrr.
Next up (tomorrow), a review of 2016’s AI highlights.
December 18, 2016
Future Bites 1
The first in what may or may not become a series of un-forecasts*, little glimpses of what may lie ahead in the century of two singularities.
[image error]
It’s 2025 and self-driving trucks, buses, taxis and delivery vans are the norm. Almost all of America’s five million professional drivers are out of work. They used to earn white-collar salaries for their blue-collar work, which means it is now virtually impossible for them to earn similar incomes. A small minority have re-trained and become coders, or virtual reality architects or something, but most are on welfare, and / or earning much smaller incomes in the gig economy. And they are angry.
The federal government, fearful of social unrest (or at least disastrous electoral results), steps in to replace 80% of their income, guaranteed for two years. This calms the drivers’ anger, but other people on welfare are protesting, demanding to know why their benefit levels are so much lower.
Meanwhile, many thousands of the country’s 1.3m lawyers are being laid off. And their salaries were much higher. The government knows it cannot fund 80% replacement of those incomes, but the lawyers are a vociferous bunch.
And there are doctors, journalists, warehouse managers, grocery store workers…
* This un-forecast is not a prediction. Predicitons are almost always wrong, so we can be sure that the future will not turn out exactly like this. It is intended to make the abstract notion of technological unemployment more real, and to contribute to scenario planning. Failing to plan is planning to fail: if you have a plan, you may not achieve it, but if you have no plan, you most certainly won’t. In a complex environment, scenario development is a valuable part of the planning process. Thinking through how we would respond to a sufficient number of carefully thought-out scenarios could well help us to react more quickly when we see the beginnings of what we believe to be a dangerous trend.
