Reviewing last year’s AI-related forecasts

Robodamus 3

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 2017 panned out.

A bit rubbish, to be honest – five out of 12 by my reckoning.  Must do better.

  1. 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.

A machine called Libratus beat some of the best human players of poker, but speech recognition in noisy environments is not yet at human standards. DeepMind made several breakthroughs. I’ll award myself a half-point.

  1. Unsupervised neural networks will be the source of some of the most impressive results.


  1. 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.

Not as far as I’m aware. No points.

  1. 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.

Yup: Alexa is very popular, and the competition is indeed heating up.

  1. 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.

There have been fewer case studies than I expected, but they do exist, and it is a rare CEO who is not building capability in AI. CAIOs are not yet common, but Dubai has a minister for AI, and the UK’s All-Party Parliamentary Group on AI has called for the UK to have one. (Disclosure: I’m an adviser to the APPG AI, and I support that call.) Half a point.

  1. 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.

Investment in Autos is galloping ahead, and they are clocking up huge numbers of safe miles, and generating huge amounts of data. States and countries are competing to declare themselves Auto-friendly.

  1. Some multi-national organisations will replace their translators with AIs.

Not as far as I’m aware. No points.

  1. Some economists will cling to the Reverse Luddite Fallacy, continuing to deny that cognitive automation will displace humans from employment. 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.

The Reverse Luddite Fallacy is proving tenacious. If anything, there seems to be a backlash against acceptance that widespread mass unemployment is a possibility that must be addressed. No points.

  1. 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.

Nope. No points.

  1. Machine language will greatly reduce the incidence of fake news.

Sadly not yet. No points.

  1. 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.

There was less news about the IoT this year than I expected. It was all blockchain instead, thanks to the Bitcoin bubble. But there was plenty of fourth industrial revolution nonsense.

  1. 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.



Putting your money where your mouth is

Long Bet image

Robert Atkinson and I have made the 749th Long Bet shown above (and online here). Robert is president and founder of the Information Technology and Innovation Foundation, a Washington-based think tank.

Robert’s claim

With the rise of AI and robotics many now claim that these technologies will improve exponentially and in so doing destroy tens of millions of jobs, leading to mass unemployment and the need for Universal Basic Income. I argue that these technologies are no different than past technology waves and to the extent they boost productivity that will create offsetting spending and investment, leading to offsetting job creation, with no appreciable increase in joblessness.

My response

AI and robotics are different to past technology waves. Past rounds of automation have mostly been mechanisation; now we will see cognitive automation. Machines can already drive cars better than humans, and their story is just beginning: they will increasingly do many of the tasks we do in our jobs cheaper, better and faster than we can. Unlike us, they are improving at an exponential rate, so that in ten years they will be 128 times more powerful, in 20 years 8,000 times, and in 30 years (if the exponential growth holds that long) a million times. We are unlikely to see the full impact of technological unemployment by 2035, but it should be appreciable. Our job now, of course, is to make sure that an economy which is post-jobs for many or most people is a great economy, and that everyone thrives. The way to do that may well be the Star Trek economy.

I would like to be able to credit the person who created the excellent image below. If you know who it is (or if it is you!) please do let me know.

Horses and tech unemp

Our wonderful future needs you!

The Future needs you, Uncle Sam

The media today is full of stories about artificial intelligence, and there is universal agreement that it is a very big deal. But ironically, most people are not paying close attention. This is probably because the stories are confused and confusing. Some say that robots will take all our jobs and then turn into murderous Terminators. Others say that is all hype, and there is much less going on than meets the eye.

And so most people shudder slightly, shrug their shoulders and get on with the business of living. And who can blame them?

When you pull back from the headlines, much of the difference between the two camps is about timing. No, robots will not take all our jobs by 2019, but can we be so sanguine about 2039? And while few AI researchers agree with Ray Kurzweil’s confident prediction that we will create the first artificial general intelligence (an AI with all the cognitive abilities of an adult human) by 2029, surveys indicate that most of them think it likely to happen this century.

There are many reasons to be excited about AI and what it will do for us and to us. It is already making the world more intelligible, and making our products and services more capable and more efficient. This progress will continue – at an exponential rate; if we are smart and perhaps a bit lucky, we can make our world a truly wonderful place.

There are also many reasons to be concerned about AI. People worry about privacy, transparency, security, bias, inequality, isolation, killer robots, oligopoly and algocracy. These are all important issues, but none of them is likely to throw our civilisations into reverse gear, or even destroy us completely. There are two issues which could do precisely that: the technological and the economic singularities.

2 sing

The technological singularity is the moment when (and if) we create an artificial general intelligence which continues to improve its cognitive performance and becomes a superintelligence. If we succeed in ensuring that the first superintelligence really, really likes humanity – and understands us better than we understand ourselves – then the future of humanity is glorious almost beyond imagination. The solutions to all our major problems should be within our grasp, including poverty, illness, war and even death. If we don’t manage that … well, the outcome could be a lot less cheerful. Ensuring that we do manage it is probably the single most important task facing us this century – and perhaps ever, along with not blowing ourselves up with nuclear weapons, or unleashing a pathogen which kills everyone.

Before we reach the technological singularity we will probably experience the economic singularity – the point when we have to accept that most people can no longer get jobs, and we need a new type of economy. The stakes here are not so high. If we mis-manage the transition, it is unlikely that every human will die. (Not impossible, though, as in the turmoil, someone might initiate a catastrophic nuclear war.) Civilisation would presumably regress, perhaps drastically, but our species would survive to try again. Trying again is something we are good at.

On the other hand, assuming it is coming at all, the economic singularity is coming sooner than the technological singularity. The technological singularity is more important but less urgent, while the economic singularity is less important but more urgent.

The economic singularity is not here yet. The impact of cognitive automation is being felt in modest ways here and there, but the US, the UK, and many other leading economies are close to full employment because there are still plenty of jobs that humans can do. (Some of it doesn’t pay very well, but there are jobs.) This will not last.

Self-driving cars will be ready for prime time in five years or so. When they arrive, inexorable economic logic dictates that professional drivers will start to be laid off rather quickly. At the same time, most other sectors of the economy will be seeing the effects of advanced AI. The outcome can be wonderful – a world where machines do the boring jobs could be one where humans get on the important parts of life: exploring, learning, playing, socialising, having fun. But it is not obvious how to get from here to there: we need a plan, and we need to communicate that plan to avoid a dangerous panic.

It will probably take at least five years to develop that plan and generate a consensus around it. So we have to start now. We need to set up think tanks and research institutes all over the world, properly funded and staffed full-time by smart people with diverse backgrounds and diverse intellectual training. In the context of the importance of the challenge, the resources required are trivial – probably a few tens of millions of dollars – but they are sufficient to require significant political support.

At the moment, our politicians and policy makers are distracted. The US is understandably mesmerised by the antics of the 45th President, and in the UK, Brexit has swallowed the political class whole. Other countries have their own distractions, and the pain of the recession which started in 2008 endures. Artificial intelligence is poised to create the biggest changes humanity has ever been through, and yet it hardly featured at all in recent elections.

But the race is far from run. Politicians do respond to the public mood. (The most talented ones anticipate it slightly, although they are careful not to get too far ahead of us, or we sack them.) If we demand they pay attention to the coming impact of AI, they will. It is time to make that demand, and you can help. Talk to your friends and colleagues about this: get the conversation going. Insist that your political representatives pay attention.

A wonderful world can be ours if we rise to the challenges posed by the exponential growth of our most powerful technology, and navigate the two singularities successfully. Let’s grasp that wonderful future!


Putting the AI in retail: How cognitive reasoning systems could make life easier for consumers

Another guest post by Matt Buskell, head of customer engagement at Rainbird.

There was a time when booking a holiday meant a single trip to the high street travel agent. Nowadays, the process of online research seems to take longer than the holiday itself. The difference, back then, was the travel agent – a human being who could look you up and down, talk to you about your preferences, and make a recommendation based on their judgement.

In the world of AI, we like to call this ‘inference’. Travel agents never asked any questions like the filters and features you find on travel websites today – location, price range, number of stars. Nothing. Instead, they inferred what we would like, basing their judgement on factors such as how we were dressed, what we liked to do, and how we spoke to them.

Where does the time go?

The average time spent researching holidays in 1997 was just 45 minutes. Now, it’s over eight hours.

The pattern is the same with other retail sectors that have moved online – from books, to groceries, music, clothing, and even cars. Hours are whittled away on websites like ASOS, TripAdvisor and Amazon. Imagine walking into a real-life store, asking the assistant for advice, and being handed a mountain of products and reviews to spend the next few hours scouring through. You’d probably just walk straight back out. So why do we settle for it online?

Convenience is one thing: for most of us, the ability to browse during the morning commute or on a lunch break is more appealing than a trip to the high street.

Many of us have also convinced ourselves that spending time looking at different online retailers and social media sites is the best way to ensure we have all the facts we could possibly need to get a ‘good deal’.

Online research
But when the choice of online stores and the availability of information was limited, it was a much simpler task. Now, we’re faced with an overload of choice, and the process of doing thorough research can feel laborious.

So why is it that targeted personal advice is lacking in online stores, whilst it is universally expected in the physical stores of our best retailers?

There are three main problems with online retailers today that limit their ability to provide the most suitable recommendations for individual customers:

1) You search for a product using narrow features, e.g. price, size, or category.

2) The system does not explain or offer a rationale for any recommendations it makes.

3) It’s a one-way interaction. You click, the computer displays.

Back to the future

The good news is that ‘conversational commerce’ and cognitive reasoning are going to bring the human element to online retailers. Ironically, the latest trends in AI are actually sending us back in time to the days of personalised shopping.

Imagine an online store in which an artificially intelligent assistant has been trained by your best human retail assistant, your favorite DJ, an experienced travel agent, or a stylist. You ask for advice or a product recommendation, and the system conducts a conversation with you, just like a real-life shop assistant.

Let’s take holiday booking as an example. The cognitive reasoning system, channeled via a chatbot – let’s call it Travel Bot – asks you a range of questions to gauge your priorities and their order of importance. During this interaction, you say that you like the beach, enjoy city breaks, hate long journeys, and indicate that price isn’t your deciding factor. Travel Bot recommends a five-star beach resort in Cannes. You baulk at the price and ask for an explanation, and Travel Bot explains that beach property is fifty percent more expensive than inland. You decide that the beach isn’t that important – to which the Travel Bot responds with an altered recommendation.

You end up with the perfect compromise – a reasonably priced hotel stay in the centre of Nice, ten minutes from the beach.

In this instance, Travel Bot mirrors a human travel agent. It makes inferences, explains its recommendations, and continuously alters its advice to cope with uncertain customer responses.

A computer cannot completely replace a good human adviser- yet. We are just too complex to model. But by bringing back the human element of customer service and combining it with the retail arena of the future, we can take a lot of the stress out of online shopping.

Rainbird is a leading Cognitive Reasoning engine that is re-defining decision-making with AI in Financial Services and Insurance.





Don’t get complacent about Amazon’s Robots: be optimistic instead!

In an article for the Technology Liberation Front, Adam Thierer of George Mason University becomes the latest academic to reassure us that AI and robots won’t steal our jobs.i His article relies on three observations: First, Amazon is keen to automate its warehouses, but it is still hiring more humans. Second, ATMs didn’t destroy the jobs of human tellers. Third, automation has not caused widespread lasting unemployment in the past.

TLF banner

Unfortunately, the first of these claims is true but irrelevant, the second is almost certainly false, and the third is both irrelevant and false.

Amazon has automated much of what humans previously did in warehouses, which has undoubtedly reduced the number of humans per dollar of value added, but the penetration of retail by e-commerce is rising very fast, and Amazon is taking share from other retailers, so it is not surprising that it is still hiring. Amazon may never get to the legendary “dark” warehouse staffed only by a human and a dog (where the dog’s job is to keep the human away from the expensive-looking machines), but it will keep pushing as far in that direction as it can. One of the major hurdles is in picking, and that looks like falling fairly soon – in years not decades.ii

Picking arm

ATMs did destroy bank tellers’ jobs, but some of the peak years of their introduction to the US market coincided with 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. 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.iii

As for the third observation, automation certainly has caused widespread lasting unemployment in the past – of horses. Almost all the automation we have seen so far has been mechanisation, the replacement of human and animal muscle power by steam and then electric power. In 1900 around 25 million horses were working on American farms and now there are none. The lives of humans, by contrast, were only temporarily disrupted – albeit severely and painfully – by this mechanisation. The grandchidren of US farm labourers from 1900 now work in cities in offices, shops and factories.

The question is whether it is different this time round, because the new wave of automation – which has hardly begun as yet – is cognitive automation, not mechanisation. My book “The Economic Singularity” presents a carefully-argued case that it is.

Of course no-one knows for certain what will happen in the next few decades. Mr Thierer and others may be right, and the people (like me) that he excoriates as “automation alarmists” and “techno-pessimists” who “suffer from a lack of imagination” may be wrong. Time will tell.

But if we are right, and society fails to prepare for the massively disruptive impact of technological unemployment, the outcome could be grave. If we are wrong, and some modest effort is spent on analysing a problem that never occurs, almost nothing would be lost. The equation is not too hard to figure out.


Finally, I reject the claim that the people who take the prospect of technological unemployment seriously are necessarily pessimistic. It is optimistic, not pessimistic, to believe that our most plausible and most positive route through the economic singularity is to work out how to build or evolve the post-scarcity Star Trek economy, in which the goods and services that we all need for a flourishing life are virtually free. We should aim for a world in which machines do all the boring stuff and humans get on with the important things in life, like playing, exploring, learning, socialising, discovering, and having fun. I refuse to believe that being an Amazon warehouse worker or an actuary is the pinnacle of human fulfillment.

Surely it is those who do think that, and who insist that we all have to stay in jobs forever, who are the true pessimists.

In the future, education may be vacational, not vocational

This post is co-written with Julia Begbie, who develops cutting-edge online courses as a director of a design college in London.

Five classrooms

Some people (including us) think that within a generation or two, many or most people will be unemployable because machines will perform every task that we can do for money better, faster and cheaper than we can.

Other people think that humans will always remain in paid employment because we will have skills to offer which machines never will. These people usually go on to argue that humans will need further education and training to remain in work – and lots of it: we will have to re-train many times over the course of a normal career as the machines keep taking over some of the tasks which comprise our jobs. “Turning truckers into coders” could be a slogan for these people, despite its apparent implausibility.

There are several problems with this policy prescription. First, we do not know what skills to train for. One school of thought says that we will work ever more closely with computers, and uses the metaphor of centaurs, the half-man, half-horse creatures from Greek mythology. This school argues that we should focus education and training on STEM subjects (scientific, technology, engineering and maths) and downgrade the resources allocated to the humanities and the social sciences. But a rival school of thought argues that the abilities which only humans can offer are our creativity and our empathy, and therefore the opposite approach should be adopted.

Science vs liberal arts
Secondly, the churn in the job market is accelerating, and within a few years, the education process will simply be too slow. It takes years to train a lawyer, or a coder, and if people are to stay ahead of the constantly-improving machines in the job market, we are likely to have to undergo numerous periods of re-training. How long will it be before each period of re-training takes longer than the career it equips us for?  And is that sustainable?

Third, reforming education systems is notoriously difficult. Over the years, educational reform has been proposed as the solution to many social and economic problems, and it rarely gets very far. Education has evolved over the last 100 years, and teachers are more professional and better trained than they used to be. But as the pictures above illustrate, most classrooms around the world today look much the same as they did 100 years ago, with serried ranks of children listening to mini-lectures from teachers. The fundamental educational processes and norms developed to build up the labour force required by the industrial revolution have survived numerous attempts to reform them, partly because reforming a vast social enterprise which looks after our children is hard, and partly because the educational establishment, like any establishment, tends to resist change.

It therefore seems unlikely that educational reform will be much assistance in tackling the wave of technological unemployment which may be heading our way.

And oddly, this may not be a problem. If, as we believe, many or most people will be unemployable within a generation or so, the kind of education we will benefit from most is one which will equip us to benefit from a life of leisure: education that is vacational rather than vocational. This means a broad combination of sciences, humanities and social sciences, which will teach us both how the world works (the business of science), and also how we work as humans – from the inside (the business of novelists, artists and philosophers). This is pretty much what our current educational systems attempt to do, and although they come in for a lot of criticism (some of it justified), by and large they don’t do a bad job of it in most places in most countries.

Although educational systems probably won’t be reformed by government diktat in order to help us stay in jobs, they will be reformed in due course anyway, because new technologies and approaches are becoming available which will make it more personalised, more effective and more enjoyable. Some of this will be enabled by artificial intelligence.


New-ish techniques like flipped learning, distance learning, and competency-based learning have been around for years. They have demonstrated their effectiveness in trials, and they have been adopted by some of the more forward-thinking institutions, but they have been slow to replace the older approaches more generally. More recently, massive open online courses (MOOCs) were heralded as the death-knell for traditional tertiary education in 2013, but they have gone quiet, because the support technologies they required (such as automated marking) were not ready for prime-time.

MOOCs will return, and the revolution which they and other new approaches promised will happen. We will have AI education assistants which know exactly which lessons and skills we have mastered, and which ones we need to acquire next. These assistants will understand which approach to learning suits us best, which times of day we are most receptive, and which times we are best left to relax or rest. Education will be less regimented, more flexible, and much more closely tailored to our individual preferences and needs. Above all, it will be more fun.

Making grammar lessons fun

The main contribution of education to technological unemployment will probably be to make it enjoyable rather than to prevent it.

Future Bites 8 – Reputation management and algocracy

The eighth in a series of un-forecasts* – little glimpses of what may lie ahead in the century of two singularities.

This article first appeared on the excellent blog run by Circus Street (here), a digital training provider for marketers.


In the old days, before artificial intelligence started to really work in the mid-2010s, the clients for reputation management services were rich and powerful: companies, government departments, environmental lobbying groups and other non-government organisations, and of course celebrities. The aims were simple: accentuate the good, minimise the bad. Sometimes the task was to squash a potentially damaging story that could grow into a scandal. Sometimes it was to promote a film, a book, or a policy initiative.

Practitioners needed privileged access to journalists in the mainstream media, to politicians and policy makers, and to the senior business people who shaped the critical buying decisions of large companies. They were formidable networkers with enviable contacts in the media and business elite. They usually had very blue-chip educational and early career backgrounds; offering patronage in the form of juicy stories and un-attributable briefings to compliant journalists.

Digital democratisation

The information revolution democratised reputation management along with everything else. It made the service available to a vastly wider range of people. If you were a serious candidate for a senior job in business, government, or the third sector, you needed to ensure that no skeletons came tumbling out of your closet at the wrong moment. Successful people needed to be seen as thought leaders and formidable networkers, and this did not happen by accident.

The aims of reputation management were the same as before, but just as the client base was now much wider, so too was the arena in which the service was provided. The mainstream media had lost its exclusive stranglehold on public attention and public opinion. Facebook and Twitter could often be more influential than a national newspaper. The blogosphere, YouTube, Pinterest, and Reddit were now crucial environments, along with many more, and the players were changing almost daily.

informal working

The practitioners were different too. No longer just Oxbridge-educated, Saville Row tailored types, they included T-shirt-clad young men and women whose main skill was being up-to-date with the latest pecking order between online platforms. People with no deep understanding of public policy, but a knack for predicting which memes would go viral on YouTube. Technically adept people who knew how to disseminate an idea economically across hundreds of different digital platforms. Most of all, they included people who knew how to wrangle AI bots.

Reputation bots

Bots scoured the web for good news and bad. They reviewed vast hinterlands of information, looking for subtle seeds of potential scandal sown by jealous rivals. Their remit was the entire internet, an impossibly broad arena for un-augmented humans to cover. Every mention of a client’s name, industry sector, or professional area of interest was tracked and assessed. Reputations were quantified. Indices were established where the reputations of brands and personalities could be tracked – and even traded.

All this meant lots of work for less traditionally qualified people. Clients who weren’t rich couldn’t afford the established consultants’ exorbitant fees, and they didn’t need them anyway. Less mainstream practitioners deploying clever bots could achieve impressive results for far less money. As the number of actual and potential clients for reputation management services grew exponentially, so did the number of practitioners. The same phenomenon was observed in many areas of professional services, and become known as the “iceberg effect”: a previous, restricted client base revealed to be just the tip of a previously unknown and inaccessible demand.

algocracy bot

But pretty soon, the bots started to learn from the judgement of practitioners and clients, and needed less and less input from humans to weave their magic. And as the bots became more adept, their services became more sophisticated. Practising offence as well as defence: placing stories about their clients’ competitors, and duelling with bots employed by those rivals: twisting each other’s messages into racist, sexist or otherwise offensive versions, tactics that many of their operators were happy to run with and help refine.


Of course, as the bots became increasingly autonomous, the number of real humans doing the job started to shrink again. Clients started to in-source the service. Personal AIs – descendants of Siri and Alexa, evolved by Moore’s Law, – offered the service. Users began relying on these AIs to the point where the machines had free access to censor their owners’ emails and other communications. People realised that the AIs’ judgement was better than their own, and surrendered willingly to this oversight. Social commentators railed against the phenomenon, clamouring that humans were diminishing themselves, and warning of the rise of a so called “algocracy”.

Their warnings were ignored. AI works: how could any sane person choose to make stupid decisions when their AI could make smart ones instead?

* 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.