At The Economist we are best known for producing a weekly newspaper, but we also publish an annual every November, looking forward to the year ahead. The latest edition, “The World in 2020” has just appeared. It features analysis and forecasts from our journalists (unusually, this is one occasion on which we get bylines), our colleagues at the Economist Intelligence Unit (who provide pithy summaries of the outlook for dozens of countries and industries) and a distinguished group of external contributors. This year this final group included Demis Hassabis, co-founder of DeepMind; Jacinda Ardern, prime minister of New Zealand; Ren Zhengfei, founder of Huawei; and Robert F. Smith, a private-equity billionaire who, while giving a commencement address in May at Morehouse College, told the class of 2019 that he would pay off their student loans when they graduate. Our aim with this annual, of which I am the deputy editor, is not so much to make precise predictions about the coming year as to give readers a wide range of stimulating ideas and perspectives to help them navigate it.

But all our contributors have one thing in common: they are human. So for this year’s edition, I thought it might be fun to ask an artificial intelligence (AI) about the future. I have been tinkering with chatbots and text-generating algorithms since I was a teenager, and I went to university to study AI in the late 1980s—only to discover, alas, that AI didn’t really work. But in recent years a specific technique called deep learning has led to extraordinary progress in the field. Most of what is called AI today, from facial recognition to voice assistants to machine translation, is in fact deep learning if you look under the hood. And in February 2019 OpenAI, a research outfit based in San Francisco, unveiled a “large-scale unsupervised language model” called GPT-2 that was created by applying a flavour of deep learning, called unsupervised learning, to 40 gigabytes of text extracted from 8m web pages on a wide range of topics. The resulting system is uncannily good at generating text in a specific style: you give it a few words as a prompt, and it then guesses what comes next, based on patterns in the text it was trained on, like a sort of supercharged autocomplete, powered by gigabytes of past examples.

OpenAI’s announcement included examples of GPT-2 writing Tolkienesque fantasy fiction and (fictitious) news-agency stories about unicorns and stolen nuclear material, complete with quotes from (entirely made-up) sources. It also turned out to be surprisingly good at comprehension, summarisation and other language tasks, despite not having been designed to do those things. Citing the risk that this program might be misused by propagandists to generate “fake news” stories, OpenAI decided not to release the full version of GPT-2 right away; instead, it was released in stages, starting with an initial, watered-down version, and slowly working up to the full-strength version over several months. This was something of a publicity stunt, and resulted in many headlines along the lines of “AI lab decides its creation is too dangerous to make public”. But it was also a way for OpenAI to emphasise the point that it takes the misuse of technology seriously (its mission is “to ensure that artificial general intelligence benefits all of humanity”), with the strong implication that others in the tech industry ought to do the same.

When I saw GPT-2’s results, though, its propaganda potential was not what most interested me. Instead I began to wonder how I could use it as a chatbot, like the ones I used to play with in the 1980s. (The most famous example from the early personal-computer era is ELIZA, but my favourite was RACTER, a text-generation system that wrote a book called “The Policeman’s Beard is Half-Constructed”.) Helpfully for me, the GPT-2 code is available on Github; even more helpfully, it can be accessed via a Jupyter notebook created by Ignacio López-Francos, a researcher at NASA. This spins up a powerful computer at Google, loads the GPT-2 model onto it, and then lets anyone play with it via a web browser. With a bit of fiddling I figured how I could use it to do an “interview” with GPT-2. I thought it would be amusing to get a deep-learning system to generate answers to questions about the impact of AI on society in the coming decades, a subject of endless speculation and debate. Human experts cannot agree on whether robots and AI will lead to mass unemployment or not, for example. Might a machine produce a more coherent answer? Probably not, but it would be an interesting experiment.

Of course, GPT-2 cannot really predict the future. For a start, it doesn’t actually understand anything: it just sometimes appears to, because it is very good at generating text in particular styles, by regurgitating words and phrases it has heard before in the same kind of context in response to a prompt. And the reams of text it was trained on were gathered in late 2017, and are thus rather out of date. As a predictive tool, then, GPT-2 is no better than a Magic 8-Ball. But I was curious about how plausible its answers might be. It turns out that simply feeding it questions as prompts does not produce very relevant answers; instead, it helps to give it a clearer idea of context. So I wrote an introductory paragraph, setting the scene and indicating to GPT-2 the style, tone and subject-matter of the text I wanted it to generate. I then added a question, prefixed with “Q:”, and began the next line with “A:” to indicate that the next words should be the answer. (GPT-2’s training set includes many Q&As, so it recognises the format. Similarly, it can be prompted to generate numbered lists, or recipes, which also appear in its training data.) The first prompt I used was:

This publication draws on a wide range of expertise to illuminate the year ahead. Even so, all our contributors have one thing in common: they are human. But advances in technology mean it is now possible to ask an artificial intelligence (AI) for its views on the coming year. We asked an AI called GPT-2, created by OpenAI, a research outfit. GPT-2 is an “unsupervised language model” trained using 40 gigabytes of text from the internet. Given a prompt, such as a question or the first sentence of a story, it generates what might plausibly come next. Here are some of its (unedited) ­answers to our questions on the big themes of 2020.
Q: Greetings, GPT-2. Have you done an interview before?
A:

I configured GPT-2 to generate five responses to each question and set the maximum output length to 75 words. (I did all this in September, so I used the 774M version of the GPT-2 model; the full-strength version had not been released at the time, though it since has been.) Here is a typical answer:

I did at the Gartner Symposium on Data-Driven Technologies in 2017. I also did an interview for Gartner's Data Analytics Summit in June 2016. Q: Why the name GPT? A:  GPT stands for Graph-Tranformation-Thin-Layer. Q: What is your role with OpenAI? A:

In this case GPT-2 has provided an answer to my original question, and has then generated subsequent questions and answers. The “Gartner Symposium on Data-Driven Technologies” is an entirely made-up event, but is a surprisingly plausible invention (the Gartner Data Analytics Summit, by contrast, does really exist). Here is another response to the same question, which starts with a much less interesting answer (and also includes follow-on questions and answers generated by GPT-2):

No. Q:  When will you be done with all this? A:  I have not finished all my work yet. The next job will be on the next page. Q:  If you did not have an interview or the job was not offered, what would you be doing now? A:  I would be working on a website

And here’s another one:

No, I have only done a video interview for a few months. Q:  In the last few years, a lot of people have said you have become less human and more machine. What do you say to that? A:  I am not sure I consider myself machine, but I have been thinking about it. Q: So you are not really

In this case, the follow-on questions and answers show that GPT-2 is generating text in the right subject area, namely the relationship between people and machines. But that wasn’t part of the answer to my initial question. So to generate my “interview”, I selected the most coherent, interesting or amusing of the five responses in each case, chopping off any follow-on questions and answers at the end. I then added the resulting answer, and my next question, to the end of the prompt, and fed it back into GPT-2. Taking the first example above, that would mean extending the prompt as follows:

This publication draws on a wide range of expertise to illuminate the year ahead. Even so, all our contributors have one thing in common: they are human. But advances in technology mean it is now possible to ask an artificial intelligence (AI) for its views on the coming year. We asked an AI called GPT-2, created by OpenAI, a research outfit. GPT-2 is an “unsupervised language model” trained using 40 gigabytes of text from the internet. Given a prompt, such as a question or the first sentence of a story, it generates what might plausibly come next. Here are some of its (unedited) ­answers to our questions on the big themes of 2020.
Q: Greetings, GPT-2. Have you done an interview before?
A: I did at the Gartner Symposium on Data-Driven Technologies in 2017. I also did an interview for Gartner's Data Analytics Summit in June 2016.
Q: Which technologies are worth watching in 2020?
A: 

Repeatedly extending the prompt in this way ensured that the questions and answers took place in the same context. In fact, there is a limit to the size of the input prompt, so after several questions I had to remove the initial paragraph from the prompt, and just feed in the preceding set of question and answer pairs. Here is the resulting article, which was published in “The World in 2020”. It includes the following exchange about the risks of AI:

Q: How worried do you think we humans should be that ­machines will take our jobs?
A: It depends what role machine intelligence will play. Machine intelligence in some cases will be useful for solving problems, such as translation. But in other cases, such as in finance or medicine, it will replace people.

Q: Do fake news stories, generated using AI, pose a threat to democracy? Are we likely to see this tactic being used in the 2020 American presidential elections?
A: Yes, we’ll probably see them. It’s just a matter of when. Fake news stories are generally generated by political campaigns, and have the potential to have a huge impact on the outcome of an election. This is because they spread misinformation, often with no basis in fact.

Some people on Twitter have challenged my use of the word “unedited” to describe these answers. They have a point. As explained above, I selected an answer from the five responses generated in each case (and did not keep the discarded responses, a decision I now regret). I also lopped off any follow-on questions and answers. I did not tinker with the text of the resulting answer; each of the answers in the piece really was generated in that form by GPT-2. But calling them “unedited” was, in hindsight, something of a stretch. Several people also asked for more detail about how I conducted the interview, given that there was not room to explain it in the original article (the length of which was limited to a single printed page). That’s why I’ve written this post. If you want to give GPT-2 a quick try, though, there are easier ways to do it, though they allow less control: talktotransformer.com, another web-based interface to GPT-2 (which is named after the program’s so-called transformer-based architecture), makes it as easy as using a search engine.

Mine is just one contribution to a burgeoning field: the New Yorker used a specially tuned version of GPT-2 to generate paragraphs in the middle of an article by John Seabrook, and the New York Times used it, and a similar text-generation system developed by the Allen Institute, to create paragraphs of “fake news”, challenging readers to distinguish between human- and computer-generated disinformation. GPT-2 has also been used to generate recipes, fan fiction and poetry (see Janelle Shane’s hilarious website, AIweirdness, for some of the funniest examples). In nearly all of these cases human selection has played a role, picking out the most interesting examples from its output. Does that give a misleading impression of what GPT-2 and similar systems are capable of? One of the great things about this technology is that, simply by playing around with GPT-2 in a web browser, you can decide the answer to that question for yourself.

Tom Standage is deputy editor of The Economist and of “The World in 2020”.

How I (sort of) interviewed an artificial intelligence was originally published in The Economist on Medium, where people are continuing the conversation by highlighting and responding to this story.

At The Economist we have been involved in digital-audio publishing for a long time. We launched our first podcasts in 2006—back in the heyday of the iPod, ours were the podcasts that Apple loaded onto the demo devices found in its stores—and we now publish five shows a week, plus various extras.

In 2007 we launched our audio edition, which consists of every story from each week’s print edition, read aloud by professional newsreaders. Recording it every Thursday, in time for publication that evening, involves running six or seven studios in parallel, and a lot of editing, because we have different readers doing the headlines and body of each article. (The audio edition is available to subscribers only, via app or download.)

In recent years we’ve also collaborated in audio joint ventures with Mic.com (“Special Relationship”), the BBC (“On Background”), WNYC (“Indivisible”) and Slate (“The Secret History of the Future”). In other words, despite being known primarily as a newsweekly in print, we are serious about audio, and have been for a long time. Now we are taking the next step: we are adding a daily news-analysis podcast to the mix.

It’s called “The Intelligence”.

The Intelligence: Trailer | The Intelligence on acast

We realise that we are joining a crowded field of daily podcasts from news organisations. But having taken the decision to launch our own about a year ago, we wanted to make sure it had the right formula, fitting into listeners’ lives while also capturing the essence of The Economist—global, analytical, playful and occasionally unexpected—in a daily podcast. After careful consideration, testing and piloting during 2018 we settled on a 20-minute, three-segment show, consisting of an opening news segment, a backgrounder/explainer/feature segment, and a jolly segment at the end that is often spun around an unusual fact or statistic (for example, what sales of mooncakes tell you about the state of Chinese politics and economics).

The result, we think, brings The Economist to life in audio every day, complementing the rest of our output for existing subscribers and acting as an introduction to our journalism for those not already familiar with it.

“The Intelligence” will be published every weekday at 6am ET/11am GMT. Its host is Jason Palmer, who has previously worked on our science desk and as editor of Espresso, our morning-briefing email and app, and before that was a broadcast journalist at the BBC.

Jason is currently the host of “The week ahead”, a forward-looking news-analysis podcast we publish every Friday. This weekly podcast will be discontinued and replaced by “The Intelligence”, appearing every weekday.

Our existing weekly podcasts (“Money talks”, on business and finance; “Babbage” on science and technology; “The Economist Asks”, an interview show; and “Tasting Menu”, featuring highlights from our entire output) will continue in their current form.

The main area where we feel we can differentiate ourselves is the global reach of our network of correspondents in 22 offices around the world, from Mexico City to Madrid to Moscow to Mumbai. Of course we’ll cover American politics as well, but our aim is, as ever, to provide a global picture, digging behind the headlines to get to the stories beneath — and highlighting stories that aren’t making headlines, but should be. Our aim is that by listening to “The Intelligence”, you will get more than just the news — it will also help you stay on top of wider global trends. We hope you’ll tune in.

Listen and subscribe here: Apple Podcasts / iTunes | Google Podcasts | Spotify | Stitcher | RSS

Why The Economist is launching a daily podcast was originally published in The Economist Digital on Medium, where people are continuing the conversation by highlighting and responding to this story.

What better way to celebrate the 175th birthday of a Victorian-era newspaper than by producing its first augmented-reality (AR) feature? That was one of the thoughts at the back of our minds as we embarked on an AR project earlier this year. Having seen other news organisation dabble with the technology, we wanted to take a closer look at it ourselves. Our New Product Development unit in New York duly teamed up with two digital designers with experience of AR, Michelle Hessel and Nitzan Bartov. Their mission: to evaluate the opportunities in AR, figure out how we might use it in our journalism and develop some prototypes.

Unlike virtual reality (VR), which transports you to another world, AR makes virtual, computer-generated objects appear in the real world. The best known example is Pokémon Go, a game in which players pursue virtual monsters that are drawn over a smartphone’s-eye view of the world. Smartphone-based AR is generally considered a mere stepping-stone towards a world in which smart glasses, like those being developed by Magic Leap, Microsoft and others, can seamlessly paint labels, data and objects into the wearer’s field of vision. AR headsets are still in their infancy, however. They are clunky and expensive, much like the giant mobile phones of the 1980s. So for the time being the focus is on smartphone-based AR. Publishers including the New York Times, Quartz and the BBC have already produced AR features: the first two added AR into their existing apps, and the BBC created an impressive standalone AR app to accompany its “Civilisations” series.

Having settled on smartphone-based AR, we began to explore possible topics and ideas. Being a global publisher, we wanted to create something that would work anywhere, which ruled out the possibility of making a “site specific” AR piece (such as a visualisation of what London or New York would look like if sea-levels rise, for example). We wondered about building an AR app that could identify satellites as they cross the sky, or make space junk visible. Or could we build an AR experience that shows what drone-filled skies might look like in future? Another area we thought about was data visualisation, one of The Economist’s traditional strengths. Could we use AR to present data around a particular topic? Or should we create a board game, with the board printed in the newspaper, but with virtual counters and other objects projected onto it?

After an initial exploration we ended up focusing on three topics: the rise of space junk, and what to do about it; what the Earth would be like if the Moon had never existed; and the future of food. Our aim was to create some prototypes in the hope of launching something in July alongside our annual supplement that imagines future scenarios, “The World If”.

For the first topic, we started by creating a detailed model of the Earth that could be made to hover just above the floor, when viewed through a smartphone. We used this as the basis for a 3D infographic about space junk, showing the steady build-up of orbital debris over the past 60 years with a timeline slider. By walking up to hotspots in the debris cloud, the user could trigger animations and commentary explaining various ways the junk might be cleaned up. For the “Earth with no Moon” idea, which is one of the ideas explored in “The World If” this year, we created a model of the Moon with hotspots that, when triggered, display information or play music. Had the Earth’s lunar companion not existed, for example, Frank Sinatra would never have sung “Fly Me to the Moon”. (Triggering actions simply by moving around, or by proximity to hotspots, rather than explicitly tapping things, is one of the emerging user-interface conventions for AR.)

At the same time, we began to explore the idea of food in augmented reality. One style of AR is known as “tabletop AR” because it involves projecting a scene, such as the battlefield for a video game, onto a table. By moving around the table, players can see the action from different angles, leaning in to get a closer look. But we wondered whether tabletop AR could be taken more literally: what better thing to project onto a table than the unusual foods we might find ourselves eating in 2050 in a warmer, more crowded world? The future of food is a very Economist topic: it’s global, it connects intensely personal choices with wider political, economic and environmental concerns, and it’s forward-looking. And given that we were founded in 1843 to campaign against the protectionist Corn Laws, it’s also in our wheelhouse.

As we started to investigate the feasibility of scanning futuristic foods in 3D, we made contact with Kabaq, a startup based in New York that specialises in combining food with AR to create detailed 3D models. (Its app, Kabaq AR Food, lets you preview foods from the menus of various New York eateries before ordering.) Its founders turned out to be Economist fans, and were keen to collaborate. They also suggested a way we could deploy our AR foods without requiring readers to download a special app: in the form of Snapchat “lenses”. Given the cost and complexity of modifying our existing apps, or creating and maintaining a new one, this approach had obvious appeal. And Snapchat lenses turn out to be capable of doing more than just plonking a 3D model into the world: they also allow for audio and limited interaction. That meant we could add audio commentary and pop-up infographics for each food, explaining its pros and cons.

By the beginning of April we had decided what we wanted to build: a handful of Snapchat lenses depicting various futuristic foods (edible insects, artificial meat, algae, meal-replacement foods and 3D-printed food) and showing information about their pros and cons. All this would allow readers to examine the foods up close and decide which of them they find most palatable. Together with Kabaq, we created two lenses during April (artificial meat and edible insects) and showed them at the Augmented World Expo, a big AR/VR fair in San Francisco. AR can’t capture taste or smell, so we gave away insect-protein bars for people interested in actually eating them.

We’ve now completed all five lenses, which meant working with Kabaq to get hold of the various foods, scan them in 3D and build the associated infographics together with our data team. Natural Machines, maker of the Foodini 3D printer, agreed to print something for one of our photo-shoots.

You can see the results, and find out more about the various foodstuffs, on this page. (If you’re reading this on a smartphone and have the Snapchat app, tap here to see the “Edible Insects” lens.) Check out all five, and share an image with the hashtag #econfood once you’ve decided which of these options for the future of food you find most appetising. Tuck in!

Would you eat that? The Economist’s first foray into augmented reality was originally published in The Economist Digital on Medium, where people are continuing the conversation by highlighting and responding to this story.

Autonomous vehicles (AVs) are one of the most talked-about technologies of the moment. And little wonder: they promise to revolutionise the transport of people, and physical goods, just as dramatically as the internet transformed the delivery of information. But they raise many questions. When will they be available? Will they be safe? Will they make car-ownership obsolete? And most of all: what is it like to ride in a car that drives itself?

I’ve spent the past few months working on a 10,000-word special report on AVs for The Economist, which was published in this week’s issue. The focus of my report is mostly on the long-term implications of AVs, based on the assumption (a reasonable one, I think) that the technology can be made to work reliably in the next few years. Rather than focusing on the minutiae of things like the ever-changing industry alliances, or who is suing who, I concentrated instead on the impact on urban planning, the transformation of retailing and the broader social and political implications of cars that can drive themselves. I spoke to as many urban planners and social historians as machine-learning experts or car-industry executives. All this horizon-scanning and future-gazing was fun. But to kick off the report, I had to actually go in a self-driving car. Which is how I found myself, on a snowy morning a few weeks ago, standing in a car park in Pittsburgh, waiting for an automated ride.

Three years ago I went in a self-driving car in Shanghai. It was quite a basic example: what is known in the field as a “Level 2” vehicle. This means it can steer itself, and maintain a safe distance from the car in front, while driving in highway traffic. I rode in an Audi A7, and a few cars now on the market (notably those made by Tesla) are capable of Level 2 automation. But the driver is required to keep hands on or near the wheel, and to pay attention to the surroundings, in case anything unexpected happens. With the next level up, Level 3, the car takes more responsibility for monitoring its surroundings, allowing the driver to relax a bit more. But if the car encounters a situation it cannot deal with, it sounds a warning telling the driver to resume control. The first Level 3 vehicle, the Audi A8, goes on sale this year.

Level 2 and Level 3 are really just glorified forms of cruise control. A truly self-driving vehicle doesn’t just follow the road ahead of it. It does route-planning and knows where it’s going. It handles junctions, crossings, traffic lights and road signs, and interacts smoothly with other vehicles, pedestrians and cyclists. This is, to say the least, a big leap. A Level 4 vehicle is defined as one that can do all of this, without any input from a human driver, within a limited area: in practice, a city neighbourhood that has been mapped in very fine detail, to give the car a big head start in understanding its surroundings. A Level 5 vehicle (something that does not yet exist) is one that can in theory drive anywhere, like a human driver. That may be an unattainable goal: some people dislike driving at night, or in snow, and I would not volunteer to drive a car in Delhi unless I really had to. The upshot is that the most advanced AVs on the roads today are generally Level 4 vehicles that operate within specific regions of particular cities.

Different cities offer different testing environments. Phoenix, Arizona is popular because it has a regular grid system and reliably good weather (snow can confuse the LIDAR sensors that AVs use to scan their surroundings). AVs can also be seen on the roads in and around Mountain View, for similar reasons, and because so many technology firms are based in the Bay Area. The early history of self-driving vehicles was shaped by the rivalry between Stanford University in California and Carnegie Mellon University in Pittsburgh, which have produced many of the engineers now leading AV projects around the world. This has made Pittsburgh another hub of AV research and testing. The city is considered a more challenging environment than Phoenix, because its road layout is more complex, and the weather is worse. San Francisco’s urban environment is particularly complex, which is why Kyle Vogt, the boss of Cruise, an AV startup acquired by General Motors, says it is the best place for testing (check out his blog post for some very impressive video footage). If you can make it there, you might say, you can make it anywhere.

Anyway, back to Pittsburgh. Uber has hired a lot of engineers from Carnegie Mellon, and Uber’s Advanced Technologies Group, which is developing its self-driving cars, is based in the city. The vehicle I climbed into was a modified Volvo XC90, with a bundle of extra sensors, including cameras and a spinning LIDAR unit, on its roof. Ryan, the vehicle’s safety driver, manually drove the vehicle out of the car park and onto the public roads, before pressing a button to engage the self-driving system. And then the car started driving itself.

At first, the experience is thrilling. It seems like magic when the steering wheel turns by itself, or the car gently slows to a halt at a traffic light. The autonomous Uber drove carefully but confidently in downtown traffic and light snow, slowing down when passing a school or approaching the brow of a hill, and putting its foot down (as it were) when faced with an open, straight road with no other traffic. The most noticeable difference from a human driver was that the vehicle made no attempt to avoid Pittsburgh’s notorious potholes, making the ride slightly bumpy at times. Sitting in the back seat, I could see a digital representation, displayed on an iPad mounted between the front seats, of how the car perceived the world, with other vehicles, pedestrians and cyclists highlighted in clusters of blue dots. I felt as though I was living in the future. But then, after a minute or two, the novelty wore off. When technology works as expected, it’s boring.

This is, in fact, exactly the reaction that engineers are hoping for. Noah Zych, Uber’s head of system safety for autonomous cars, told me that after working on AVs for ten years, he was finally able to offer his parents a ride in a self-driving car last year when they came to visit. After their ride ended, he asked them what they thought of it. “And my mom said, ‘actually, it was kind of boring’. And that’s the response that we really want,” he says. Uber has offered some riders in Pittsburgh and Phoenix the option to travel in its self-driving vehicles, provided the start and end points of their ride fall within their area of operation. (Riders can say no if they want to.) Around 50,000 people have travelled in Uber’s self-driving cars in the past couple of years. Uber wants to understand how to design the in-car experience (such as what information should be shown on the screen), and it also wants to reassure both riders and other road-users about the safety of autonomous vehicles. “The best way to convince people that a self-driving vehicle is going to be safe and capable of driving them around in the future is to give them that first experience,” says Mr Zych.

Ryan, the safety driver in my self-driving Uber, had to take over occasionally, for example to steer the car around a delivery truck that had blocked the road — the car was programmed to play things safe and wait, rather than cross the double-yellow lines in the middle of the road — and to guide the car through roadworks where the lane markings had been recently changed. A couple of times he also took over when the car looked as though it might be passing a bit too close to another vehicle. In each case a collision was unlikely, Ryan explained, but if people think a collision is imminent, they will not feel safe. So part of his job is to flag up instances where the car’s driving style could be tweaked to provide a better experience for passengers. At the end of each day, the contents of the car’s on-board computers are downloaded for analysis. Each time the safety driver had to take over — an event known as a “disengagement” — the corresponding data can be analysed to see how the car’s software could be improved. It is then possible to simulate how the car would have responded with various modifications to its algorithms. “We can play it back again and again, vary the scenario and see the distribution of outcomes,” says Mr Zych. After being tested in simulation, the improved software is then rolled out in real vehicles. It is first tested on a small subset of routes, called “canonical” routes, which test different aspects of its behaviour. If it works as expected, the software is then rolled out for general use.

I spent most of my hour in a self-driving Uber discussing disengagements, algorithm design and user interfaces, which (to me, at least) are just as exciting as being in a futuristic robocar. And even when the driving algorithms are working perfectly, there are several practical questions that still have to be addressed. For example, how will people actually hail driverless vehicles? They can’t just stop anywhere, or they will block traffic and annoy people. Human drivers can pick a good place to stop, but machines will need help. Uber has already started identifying good pick-up and drop-off points in some cities, and suggesting them to riders of human-driven vehicles. But it may be that in future, streets will have designated pick-up and drop-off areas; already, some university campuses and apartment blocks are being built with ride-hailing in mind. And how will a self-driving vehicle be able to tell that everyone is on board and ready to go? Some kind of “start” button will be needed — and a “stop” button, too, in case a rider suddenly wants to get out of the vehicle. (The driverless Uber has a “pull over” button for this.)

Waymo, the self-driving unit of Google’s parent company, hopes to launch a robotaxi service in Phoenix later this year. Waymo has the lowest disengagement rate in the industry, and is generally considered the leader in the field; its autonomous vehicles can now operate in Phoenix without the need for safety drivers. GM’s Cruise, which is fast catching up with Waymo, hopes to launch a robotaxi service in 2019, using autonomous Chevy Bolts that do not have a steering wheel, pedals or any other kind of manual controls. So they, too, will have to be able to operate entirely autonomously without a safety driver. Dozens of other firms are also working on self-driving vehicles. Over the coming months and years more AVs will take to the roads in more cities, and the areas in which they operate will gradually expand. Probably sometime in the 2020s, you will take your first ride in a self-driving car. It will be exciting at first — but then, if all goes well, it should quickly become reassuringly boring.

Further reading:

• Policymakers must apply the lessons of the horseless carriage to the driverless car
• Special report: The implications of autonomous vehicles

What’s it like to ride in a self-driving car? was originally published in The Economist on Medium, where people are continuing the conversation by highlighting and responding to this story.

And why tintypes are the opposite
of Instagram

The 19th-century technique of wet-plate collodion photography, and the tintype process in particular, is undergoing something of a renaissance at the moment as digital photography prompts people to explore older analogue methods. Check out Joni Sternback’s tintypes images of surfers, for example, or Victoria Will’s tintype portraits of movie stars, taken at the Sundance Festival.

I’m interested in Victorian technology, so I wanted to give it a try. So did my wife Kirstin, for a rather different reason: she has been shooting a lot of film lately, and enjoys the way its physical limitations force her to put much more thought into each image.

So the chance to spend two days doing photography using 150-year-old techniques appealed to us both, because it’s pretty much the oldest and most fiddly kind of photography there is. In an age when photography has never been easier or quicker, we wanted to see what it was like when it was difficult and slow.

Fortunately, there are people out there who will show you how to do it, even if you’ve never been near a darkroom before. Our tutor for the weekend was John Brewer, who holds regular workshops in Manchester and London. The course was hosted by the Double Negative Darkroom, based in East London, not far from where we live.

The seven people on the course (including the two of us) all had different reasons for wanting to explore this old technique: for personal projects, to teach it to others, to try something new or to experience photography in its original, primitive form.

John began by telling us the history of the wet-plate collodion process, which was the dominant process from the 1850s to the 1880s. Along the way, John repeatedly emphasised just how dangerous it was to mix the chemicals yourself. It turns out that you can quite easily create explosives by accident, and deadly poisonous cyanide gas. Fortunately John had mixed everything for us in advance, which greatly reduced the risk and complexity. (We wanted it to be hard, but not that hard.) He also explained how to operate the old cameras and lenses we were using, many of which were themselves antiques. Then he did a glass-plate test shot (of Kirstin and me, as it happens) to show us how the process worked in practice.

The wet-plate collodion process involves a huge number of manual steps: cutting the glass or metal plate; wiping egg-white along its edges; coating it evenly with a syrupy substance called collodion; making it light-sensitive by dunking it in silver nitrate for a few minutes; loading the wet plate carefully into a “dark slide” which is inserted into the camera; taking the picture; then developing it, which is rather like processing a black-and-white print.

You have about 15 minutes to expose and process the wet plate before it dries out. This is why photographers in the 19th century had to take chemistry labs with them everywhere, in black tents. There are a lot of steps, in short, and a lot of things to go wrong. And it was even harder and more dangerous without modern chemicals. We quickly came to appreciate just how amazing it was that people managed to photograph anything at all.

For each image, we set up the shot we wanted first, so we wouldn’t waste valuable time composing and focusing. Then we went into the darkroom to make the wet plate.

On the first day we all made images on glass, and on the second day we used aluminium (images on metal plates are called tintypes, though nobody actually uses tin). Next we went outside to expose the plate. This process has an effective ISO/ASA of less than 1, so it requires long exposures, typically of between four and 11 seconds. You can’t use a light meter, because wet plates are sensitive to a different range of the visible spectrum. So John had to guess the initial exposures for us as a starting point, and we all adjusted as we went along. The exposures are so long that cameras of the period do not need shutters. Instead, we could just put a bowler hat over the end of the lens, remove it for the exposure, and then put it back again.

Then we rushed back inside to develop the image under a red light in the darkroom. This involves pouring a tiny cup of developer over the plate so that it is entirely covered, which is more difficult than you might think. Once the image has appeared in negative, water is used to stop the process, and the plate is then dropped into a fix bath, which causes it to turn, magically, into a positive image. It’s the fix bath that can produce cyanide fumes if you get things wrong.

Finally, the image goes into a water bath to be washed. When the plate is dry, there’s another opportunity to mess things up: coating the image with lavender oil, which again must be poured by hand to make an even coat, and only when the plate has been heated up just the right amount, from below, with a hair dryer. It smells lovely, though.

Each of us made several images. A few of the plates came out perfectly, but most of them went wrong in one way or another, and we were all able to learn from each other’s mistakes. It didn’t matter that things went wrong, and some of the mistakes looked really arty, like my first glass-plate image of Kirstin. You can tell that I didn’t fully coat the plate with collodion, which is why there’s a blank area at the bottom right. Balancing the plate on your fingertips, pouring liquid over it and tilting it carefully to ensure an even coat is a skill that takes much longer than a weekend to master.

We all came away from the course impressed by just how difficult photography used to be. At the same time, back then it must have seemed like magic, so you can see why people were prepared to make all that effort. And when you see a picture that has been taken outdoors, it’s amazing to think how much equipment had to be carried around.

Given the immediacy of the process, we were tempted to liken it to a kind of Victorian Instagram. But we had the darkroom nearby and everything mixed and ready to go, which made it relatively quick and easy. For practitioners at the time, wet-plate photography, with its inconvenience and lack of portability, would have been far more laborious. There is however another analogy between the wet-plate era and modern photography — one that holds a lesson for us today.

The Google Glass of the 1870s

Wet-plate collodion was displaced almost overnight by the advent of dry-plate photography in the late 1870s. Suddenly there was no need to carry chemicals around: companies sprang up to make dry plates, which were much more convenient to handle and far more sensitive than wet plates, allowing for faster exposures. For the first time it was possible to take a camera anywhere and snap an image without a tripod, and without asking permission. The combination of dry plates and affordable cameras from Kodak and other manufacturers democratised photography, an activity which was quickly embraced by an army of enthusiastic amateurs.

The snapshot craze caused an immediate backlash as people objected to being “Kodaked” by “camera fiends.” They were worried that photography constituted an invasion of personal privacy, and that it might be used to capture images of people in compromising situations without their permission (for example, when wearing swimwear).

In 1884, under the headline “The Camera Epidemic”, the New York Times likened the spread of cameras to an outbreak of cholera. “Photography affords a very nice pastime to certain people, but the multiplication of instantaneous cameras has become a perfect nuisance to the general public,” one man complained to his local newspaper. (This episode is brilliantly described by Bill Jay, a historian of photography, in his 1986 essay “The Camera Fiend”.)

The parallel with modern worries about smartphones and wearable cameras like Google Glass, which allow photographs to be taken anywhere, even surreptitiously, is striking. Efforts to ban snapshot cameras failed, except in Germany, which introduced a law banning photography without permission in 1907. Instead, new social norms arose to govern the use of snapshot cameras.

Today we are still in the process of negotiating the finer points of the etiquette of smartphone photography — is it acceptable for the president of the United States to shoot a selfie at a memorial service? — but it is now generally accepted. Arguments over the use of wearable cameras are at an earlier stage. But the history of dry-plate photography suggests that initial outrage will give way to eventual acceptance, provided the broader benefits (and not merely the drawbacks) of the new technology become apparent.

Going through the laborious process of wet-plate photography provides an obvious contrast with digital photography, but it also offers some sense of the earthquake that the advent of dry-plate photography must have caused. We’re used to thinking of photography as being divided into analogue and digital eras, but the transition from long exposures with giant cameras to snapshots with pocket cameras was arguably just as significant. That’s why the alchemical process of wet-plate photography is as far from digital photography as it is possible to get: in both respects, wet-plate photography is the opposite of Instagram. And that contrast explains why it is attracting so much renewed interest today.

Tom Standage is digital editor of The Economist and author of “Writing on the Wall: Social Media — The First 2,000 Years” (Bloomsbury). He co-wrote this post with Kirstin Mckee. Please ask before reproducing/republishing any of the images.

Twitter | Facebook | Instagram

A First-Timer’s Foray Into Wet-Plate Photography was originally published in Vantage on Medium, where people are continuing the conversation by highlighting and responding to this story.

If you like camping or you’re into archaeology—and Egyptian archaeology in particular—then it doesn’t get much more exciting than camping in the British Museum, in the Egyptian sculpture galleries, surrounded by statues and sarcophagi. Having done several such sleepovers with our daughter, we recently got to do our first with our younger son. This is what it’s like.

You arrive at the museum on a Saturday evening at 6pm, just as it’s closing. Tourists and other visitors are filing out, and excited families, wheeling suitcases and carrying bedding rolls, are filing in. To be able to apply for tickets for one of the sleepovers, one of your children needs to be a “Young Friend” of the museum.

Once inside, everyone lines up in the Great Court, and we inch forward to the administration desk, where we’ll be told which of four groups we’ve been assigned to, and gallery we’ll be sleeping in.

The administration desk for the sleepover, which steers us towards either the Assyrian or Egyptian sleeping areas, is set up in front of the Rosetta Stone. That’s the Rosetta Stone, which was discovered in 1799 during Napoleon’s expedition to Egypt, and which provided the key to deciphering Egyptian hieroglyphics, because it has the same text written in three different scripts. There it is on the left.

We are told to make camp in one of the Egyptian sculpture galleries, much to our relief. The Egyptian galleries are quite spacious, are filled with epic sculptures and have a few warm spots on the floor. We set up our bedding rolls and sleeping bags as best we can.

The Assyrian galleries have spectacular carvings, but we know from previous sleepovers that they can be quite cold. They are also much smaller and more resonant, which can be a problem if someone nearby is snoring loudly.

Once we’ve made camp, the evening’s activities can begin. There are four different activites that the four groups each do in a different order. For our group, the first activity was making canopic jars out of long sausages of clay. During the mummification process, internal organs were removed and stored in these jars, which were placed in the tomb with the mummy.

The jars traditionally had lids that represented various gods (the four sons of Horus). Some of our jars look more authentically Egyptian than others.

The next activity for our group is role-playing and Egyptian-style dancing, which wears the children out quite effectively. There’s then a brief break during which the adults are served coffee.

Then it’s on to the next activity, which is drawing amulets and writing hieroglyphics on pieces of papyrus, which are then arranged to decorate this papier-maché sarcophagus. Drawing on papyrus with felt-tip pens is tricky.

The final activity is a storytelling session in front of an Ancient Greek temple. We listen to creepy tales of mummies and mummification, both ancient and modern. By this stage it’s nearly midnight, and many of the children (and some of the adults) are falling asleep. We retreat to our sleeping areas and get ready for bed. Walking through the darkened museum is both spooky and strangely exciting.

The nightwatchman makes his rounds with his torch, but nearly everyone is asleep. The children are too exhausted to worry about statues coming to life, marauding mummies or scary sarcophagi.

Morning! The lights come on at 7am and after a quick breakfast of fruit, croissants and coffee we pack up our sleeping bags and bedding rolls.

We’re moving out, but some of the sleepers in the Egyptian gallery are permanent residents, like this intricately carved sarcophagus.

Having packed up our gear, we head for the Egyptian galleries. The museum hasn’t opened to the public yet, so we campers have them all to ourselves. Sleepovers are usually linked to a temporary exhibit as well (in this case, of eight mummies which have been “unwrapped” using medical CT scanners). We went to that exhibit too, but we weren’t allowed to take pictures.

Finally, the bleary-eyed campers head home. But we’ll be back for another sleepover before long. We thoroughly recommend it. Similar sleepovers are also possible at the Science Museum in London, in the American Museum of Natural History in New York and two of the Smithsonian museums in Washington, DC. If you get a chance to spend a night in the museum, take it!

All photos © Tom Standage. Please ask if you want to reproduce/republish them.

YES, the Romans really did have their own version of the iPad. Instead of notebooks they jotted things down on wax tablets of various shapes and sizes, from small ones (the size of an iPhone) to big ones (the size of a big iPad, before the iPad Air showed up). The image at the top of this post is a particularly fine example, from the Roman-Germanic Museum in Cologne (to which I am indebted for the picture). It is the size, shape and aspect ratio of a modern iPad. (Note to Google: it’s a 4:3 aspect ratio rather than 16:9, so the Romans are with me when I say that I find 16:9 a bit odd.)

These tablets consisted of a layer of wax inside a wooden frame. You’d use a stylus to scratch things into the wax, and flip the stylus over to use its flat end to smooth the wax and erase things when necessary. That’s why I like to say that the woman in this image (a mural from Pompeii) is using the Roman equivalent of a Samsung Galaxy Note 3. As you can see, several tablets were often bound together, like the pages in a book, to increase storage capacity. You can buy a modern replica, if you’re interested in trying this, or make your own.

If you were a wealthy Roman who wanted to read the news over breakfast, you might send a scribe down to the forum to jot down excerpts from the daily gazette, or acta diurna, onto one of these tablets. (It was arguably the world’s first newspaper, founded by Julius Caesar in 59BC, but only one copy was produced each day, posted on notice boards in the forum.) Having read the news yourself you could then copy selected highlights to your friends: the text would be transcribed onto papyrus rolls and taken to them by messenger. They might then copy those news reports, in turn, to their own friends, adding their own comments or analysis. It was common for wealthy Romans to rely on their friends to filter and distribute the news for them in this way when they were away from the city. This is how the Roman social-media system worked. It’s just one example (and, in fact, the earliest example) of how social-media environments predate the Internet.

Because the amount of space on a wax tablet or single sheet of papyrus was limited, the Romans used abbreviations in their letters. SPD, short for “SALUTEM PLURIMAM DICIT” meant “says many greetings”, for example; if I was writing to someone called Mark I (or my scribe) might write TOM MARCO SPD (Tom, to Mark, says many greetings) at the top. Another abbreviation was used as a sign-off: SVBEEV, short for “SI VALES BENE EST EGO VALEO”, which meant “If you are well, that’s good; I’m well”. It was equivalent to TTFN or GTG today. Alas, there seems not to have been a Roman equivalent of LOL. But the Romans certainly had their own version of the iPad. SVBEEV.

Tom Standage is digital editor of The Economist and author of “Writing on the Wall: Social Media—The First 2,000 Years” (Bloomsbury)

TWITTER may define the Zeitgeist, but social media is not as new as it looks. Social-media ecosystems, in which information is passed horizontally from person to person along social connections, date back to Roman times. And more than two centuries before Twitter, Facebook, Pinterest and the rest, another social-media environment arose in colonial America. It consisted of an interlinked network of local newspapers—and its animating figure was Benjamin Franklin.

The young Franklin got his start in the media business working for his older brother James, the editor of the New-England Courant, a Boston paper. Benjamin’s first foray into newspaper writing took the form of a series of 14 letters, written under the pseudonym of a widow named Silence Dogood, which he submitted to the New-England Courant’s office, and which were enthusiastically published by his unwitting brother. (The letters proved popular with readers, some of whom even wrote in to propose marriage.) James was furious when 16-year-old Benjamin admitted to having written the letters. This tale does not simply illustrate Benjamin’s ingenuity and writing prowess; it also shows how newspapers at the time were open to submissions from anyone, provided they expressed an interesting opinion. Small and local, with circulations of a few hundred copies at best, such newspapers consisted in large part of letters from readers, and reprinted speeches, pamphlets and items from other papers. They provided an open platform through which people could share and discuss their views with others. They were, in short, social media.

Benjamin Franklin went on to run a newspaper himself, launching the Pennsylvania Gazette in Philadelphia in 1729. A few years later he became Philadelphia’s postmaster, in part because control of the postal service “facilitated the correspondence that improved my newspaper”. He ran the system so efficiently that in 1753 he was made deputy postmaster general for the American colonies. In this role he increased the reliability and frequency of the postal service, reorganising routes and streamlining procedures. The number of deliveries from New York to Philadelphia went from one a week to three a week, and the postal service became profitable for the first time.

Just as important, Franklin allowed free exchange by post of newspapers both within and between colonies, formalising the reprinting of noteworthy reports and letters by papers in different towns. As the publisher and printer of the Pennsylvania Gazette, encouraging the circulation of news in these various ways was in Franklin’s own commercial interest. But it also contributed to the dynamism, vitality and unity of the American colonies’ emerging information ecosystem. It allowed significant letters and pamphlets to reach a wide audience as they were printed in one newspaper and then copied and reprinted by others. By the 1760s the colonial newspaper network had developed into a powerful, open and social platform for rehearsing arguments, propagating ideas and exchanging opinions.

As tensions grew with the government in Britain, many notable letters and pamphlets lit up this network, including John Dickinson’s anonymous “Letters from a Farmer”, which originally appeared in the Pennsylvania Chronicle and were then widely reprinted, and John Adams’s writings under the pen name “Novanglus”, which appeared in Boston newspapers in 1775, arguing with “Massachusettensis” (Daniel Leonard, a Loyalist lawyer who defended the British government’s position). But most successful of all was Thomas Paine, a recent immigrant to the colonies who in January 1776 articulated the case for independence more clearly and forcefully than anyone had done before. His pamphlet, “Common Sense”, quickly rippled through the colonies, shared at first among the political elite, who excitedly recommended it to each other, and then widely reprinted and excerpted in local papers. Unquestionably the most popular and influential pamphlet of the American Revolution, it eventually sold more than 250,000 copies and made Paine the world’s bestselling author.

The popularity of “Common Sense” revealed to the colonists the breadth of support for independence. Many years later, John Adams wrote disapprovingly to Thomas Jefferson that “history is to ascribe the American Revolution to Thomas Paine.” That is an exaggeration, but not much of one. The revolution was certainly helped along by America’s unusually free and open media ecosystem. The circulation of letters, pamphlets and newspapers and the resulting interchange of ideas helped bind together the separate colonies and united them behind a common cause.

America’s contemporary enthusiasm for social media echoes the popularity of its original social-media platform, created by one of the Founding Fathers. Sorry, Facebook and Twitter—but Benjamin Franklin got there first.

Tom Standage is digital editor of The Economist and author of “Writing on the Wall: Social Media—The First 2,000 Years” (Bloomsbury). For more examples of historical precursors of modern social media, visit his blog.

Benjamin Franklin, Social Media Pioneer was originally published in New Media on Medium, where people are continuing the conversation by highlighting and responding to this story.