0:00:01 > 0:00:04Now on BBC News, it's time for Click.
0:00:06 > 0:00:09This week, mapping the poorest parts of the world.
0:00:09 > 0:00:12Inside a shopper's mind.
0:00:12 > 0:00:16And, hands up who is not flying the plane?
0:00:39 > 0:00:40Thanks for the information.
0:00:40 > 0:00:42Pay attention.
0:00:42 > 0:00:48We are living in interesting times.
0:00:48 > 0:00:50There are things happening in artificial intelligence right
0:00:50 > 0:00:55now that will fundamentally change our world.
0:00:55 > 0:00:59Soon, machines will learn to do our jobs.
0:00:59 > 0:01:02They will learn to walk.
0:01:02 > 0:01:06They will learn to drive.
0:01:06 > 0:01:11And at that point, things get very interesting.
0:01:11 > 0:01:14We'll talk more about the consequences of an automated
0:01:14 > 0:01:20society in a few minutes but, after thinking, walking and driving,
0:01:20 > 0:01:26have you ever wondered how hard it would be for a computer to fly?
0:01:27 > 0:01:31I'm not talking about drones that can fly between points, follow
0:01:31 > 0:01:36a target or perform circus tricks.
0:01:36 > 0:01:38I'm talking about aircraft that could intelligently
0:01:38 > 0:01:43decide on a flight plan, just as a human would.
0:01:43 > 0:01:47And that is what Mark has been hunting down.
0:01:50 > 0:01:53Here at BAE Systems in Woolton in Lancashire, they are used
0:01:53 > 0:01:58to the roar of jet engines.
0:01:58 > 0:02:02They built and have tested Eurofighter Typhoons here.
0:02:02 > 0:02:06Today, however, I am going to take a flight in an aircraft that is much
0:02:06 > 0:02:10slower than a Typhoon.
0:02:10 > 0:02:13This is a Jetstream 31, a small passenger aircraft.
0:02:13 > 0:02:19It's a design from the 1980s but it's currently used by BAE
0:02:19 > 0:02:22Systems as a flying test-bed for technology which could lead
0:02:22 > 0:02:26to fully autonomous aircraft.
0:02:27 > 0:02:30Maureen McCue is head of research here.
0:02:30 > 0:02:32Tried and tested.
0:02:32 > 0:02:35Very well flown and understood aircraft from the outside,
0:02:35 > 0:02:40but on the inside, it's filled with the latest technology.
0:02:40 > 0:02:42That technology will eventually allow this aircraft to fly
0:02:42 > 0:02:45itself autonomously.
0:02:45 > 0:02:48Today, they are testing the plane's ability to detect and avoid
0:02:48 > 0:02:52clouds as well as testing its satellite communication systems.
0:02:52 > 0:02:55But take-off and landing will still be handled by human
0:02:55 > 0:02:57pilots and the plane will be remotely controlled
0:02:57 > 0:03:00at some points, too.
0:03:00 > 0:03:03How does this fit into the autonomous equation?
0:03:03 > 0:03:05At the moment, it's effectively a remotely controlled aircraft.
0:03:05 > 0:03:08It is, and really, with autonomous operations,
0:03:08 > 0:03:11you need to progressively expand the boundary.
0:03:11 > 0:03:15You can't start with such a big bang right out at the full range
0:03:15 > 0:03:18of capability of operations.
0:03:20 > 0:03:25This humble looking outbuilding houses the grand station.
0:03:25 > 0:03:28Here, a pilot will remotely fly the plane and he can ensure it
0:03:28 > 0:03:37will react to instructions from air traffic control.
0:03:37 > 0:03:43I would expect to see a joystick and images coming
0:03:43 > 0:03:46through from the cockpit but you're not going to be flying
0:03:46 > 0:03:48like that at all, are you?
0:03:48 > 0:03:51No, everything is through the numbers that you can see there.
0:03:51 > 0:03:51Altitude, speed.
0:03:51 > 0:03:52Time for take-off.
0:03:52 > 0:03:55These flights are taking place in uncongested airspace.
0:03:55 > 0:04:02Today, we will be flying over the Irish Sea.
0:04:02 > 0:04:07One minute to handover point.
0:04:07 > 0:04:11To help fly itself, this aircraft uses data from satellites
0:04:11 > 0:04:14as well as identifying radio signals broadcast by other aeroplanes, so it
0:04:14 > 0:04:18knows who it's in the sky with.
0:04:18 > 0:04:22It is also fitted with a camera that can see other air
0:04:22 > 0:04:26users, even if there are not warning signals.
0:04:26 > 0:04:28So right now, the pilots aren't actually flying the aircraft.
0:04:28 > 0:04:38This gentleman right here is.
0:04:38 > 0:04:40From that 15-year-old Dell laptop that's running Windows XP.
0:04:40 > 0:04:43At this point, the aircraft is flying autonomously with a human
0:04:43 > 0:04:47watching what it's doing.
0:04:47 > 0:04:51Handing control of the computing over to the autopilot in the back
0:04:51 > 0:05:04and once established on the route, I can hand the computing
0:05:04 > 0:05:07over to Clive on the ground by the satellite.
0:05:07 > 0:05:10So that little shed-like building we were in earlier with Clive
0:05:10 > 0:05:13in front of the computer, he is now flying the aircraft.
0:05:13 > 0:05:16Over the course of the testing of this aircraft, it's going to have
0:05:16 > 0:05:18to perform a variety of different, complex tasks.
0:05:18 > 0:05:22For instance, it's going to have to recognise and avoid bad weather.
0:05:22 > 0:05:23Not just weather, but other aeroplanes too.
0:05:23 > 0:05:26It will eventually be able to select a safe landing spot
0:05:26 > 0:05:28and touchdown by itself.
0:05:28 > 0:05:31Today, we can't really test its weather detection
0:05:31 > 0:05:33abilities, though, as, unusually for the UK,
0:05:33 > 0:05:36we have a cloudless sky.
0:05:36 > 0:05:38BAE suggests that autonomous aircraft could be used
0:05:38 > 0:05:41to perform dirty, dangerous or repetitive tasks.
0:05:41 > 0:05:43But could this technology be introduced into
0:05:43 > 0:05:47commercial air travel?
0:05:47 > 0:05:52At the moment all commercial aircraft have a set number of crew.
0:05:52 > 0:05:55There are programmes in existence looking at how you reduce crew
0:05:55 > 0:05:58either planned from the outset or, in the case of an emergency,
0:05:58 > 0:06:01you've got the autonomous system as a fallback so that you can
0:06:01 > 0:06:04still have perhaps a two-crewed aircraft on a certain length
0:06:04 > 0:06:08of flight but one of the crew happens to be an autonomous helper
0:06:08 > 0:06:14as opposed to a real person.
0:06:14 > 0:06:17We're going to end up in the Hudson.
0:06:17 > 0:06:19I'm sorry, say that again.
0:06:19 > 0:06:21But what happens when things go wrong?
0:06:21 > 0:06:22This is the captain.
0:06:22 > 0:06:27Brace for impact.
0:06:30 > 0:06:33While aerospace manufacturers are exploring the possibilities
0:06:33 > 0:06:37of fewer cockpit crew, what do commercial pilots think?
0:06:37 > 0:06:42To find out, we paid the British Airline Pilots Association a visit.
0:06:42 > 0:06:45Many decades of looking at aviation has brought us to the position
0:06:45 > 0:06:48where we have come to the conclusion it's best to have two
0:06:48 > 0:06:52pilots in the cockpit, because if you reduce that to one,
0:06:52 > 0:06:55the problem you've got then is you've got no one to cross check
0:06:55 > 0:06:56each other's decisions.
0:06:56 > 0:06:58Take for example, the miracle on the Hudson.
0:06:58 > 0:06:59Captain Salinger.
0:06:59 > 0:07:02When the aircraft lost both its engines, the pilots had
0:07:02 > 0:07:04to have a discussion and they decided their only course
0:07:04 > 0:07:06of action was to land on the river.
0:07:06 > 0:07:11No computer can be programmed to do that.
0:07:11 > 0:07:13The flight testing of autonomous aircraft continues but the debate
0:07:13 > 0:07:17about regulating them and how they are going to be used has
0:07:17 > 0:07:22really only just begun.
0:07:22 > 0:07:24That was Mark and this is Tim Harford, columnist
0:07:24 > 0:07:28for the Financial Times and Tim, you've written a book about how
0:07:28 > 0:07:35the systems that we now rely on can sometimes backfire.
0:07:35 > 0:07:39What do you make of the idea of planes that might only need one
0:07:39 > 0:07:41pilot or even no human pilots?
0:07:41 > 0:07:44It's exciting, isn't it?
0:07:44 > 0:07:47And of course, auto pilots have made planes safer
0:07:47 > 0:07:49but what worries me is, what happens when
0:07:49 > 0:07:50something goes wrong?
0:07:50 > 0:07:52No system is perfect, including a system where
0:07:52 > 0:07:55the humans fly the plane.
0:07:55 > 0:07:58I guess when it goes wrong, it has two hand back
0:07:58 > 0:08:00to the human in the cockpit.
0:08:00 > 0:08:03Autopilot hands back to the human in the cockpit but then what?
0:08:03 > 0:08:06The human is out of practice, the human is not used to flying
0:08:06 > 0:08:09the plane and because the autopilot has failed, it's probably
0:08:09 > 0:08:11in an unusual situation.
0:08:11 > 0:08:14There was a worrying example of this a few years ago.
0:08:14 > 0:08:16An Air France crash over the Atlantic Ocean.
0:08:16 > 0:08:20The plane was flying quite high above a storm.
0:08:20 > 0:08:22The autopilot disconnected and the pilots just weren't used
0:08:22 > 0:08:29to flying at high altitude.
0:08:29 > 0:08:33They were only used to operating the plane on take-off and landing
0:08:33 > 0:08:36and they flew a perfectly good plane into the Atlantic Ocean
0:08:36 > 0:08:39because they were confused about what was happening.
0:08:39 > 0:08:42They killed everybody on board, an absolute tragedy, and this
0:08:42 > 0:08:43is the paradox of automation.
0:08:43 > 0:08:46The autopilots are normally so safe, so reliable,
0:08:46 > 0:08:47that when they fail, the pilots find
0:08:48 > 0:08:51themselves so confused.
0:08:51 > 0:08:55I guess the next question is, what about autonomous cars?
0:08:55 > 0:08:58We have been talking about how they will blissfully drive us around
0:08:58 > 0:09:03and we won't have to take control.
0:09:03 > 0:09:06I suppose for the foreseeable future, they won't be good enough
0:09:06 > 0:09:09to be 100% reliable?
0:09:09 > 0:09:14I guess they'll never be 100% reliable but the model
0:09:14 > 0:09:17where if it's confused, it hands back to the human,
0:09:17 > 0:09:19it's not going to work.
0:09:19 > 0:09:21You are there with your bagel, your coffee, your newspaper.
0:09:21 > 0:09:25You look up, there's a bus coming towards you, and the car goes,
0:09:25 > 0:09:27autopilot disengaged, human take control, it clearly not
0:09:27 > 0:09:30going to work in that way.
0:09:30 > 0:09:33What makes more sense is for the human to be driving
0:09:33 > 0:09:36and for the computer to be watching out for a dangerous situation,
0:09:36 > 0:09:39for the computer to take over if there's a problem.
0:09:39 > 0:09:42Humans get bored, get distracted, lose their skills.
0:09:42 > 0:09:48None of these things happened to computers.
0:09:48 > 0:09:52But I guess we are in an extended period of time before the far future
0:09:52 > 0:09:54happens and computers drive and fly us.
0:09:54 > 0:09:58We've got possibly decades if not a century of being in this
0:09:58 > 0:10:00interim period where, if there is a problem,
0:10:00 > 0:10:03we are going to end up blaming them for this really unusual,
0:10:03 > 0:10:09weird crash that a human wouldn't have made.
0:10:09 > 0:10:12Yes, and I think a glimpse of that is where we are asking
0:10:12 > 0:10:16the computers to make a decision not about planes or cars but about,
0:10:16 > 0:10:20for example, who'll get a promotion or who gets a special deal in a shop
0:10:20 > 0:10:22or who gets arrested for shoplifting because the computer
0:10:22 > 0:10:24recognises your face.
0:10:24 > 0:10:27We are already asking computers to make this sort of decision
0:10:27 > 0:10:31and the lesson of the paradox of automation is that we need to be
0:10:31 > 0:10:34much more savvy about the fact that computers do make mistakes
0:10:34 > 0:10:37and we need to take this seriously.
0:10:37 > 0:10:39Thank you very much, Tim.
0:10:39 > 0:10:39Thank you.
0:10:39 > 0:10:41Don't trust the machines.
0:10:44 > 0:10:46Hello and welcome to The Week In Tech.
0:10:46 > 0:10:49It was the week that inventor Haiyan Zhang developed a smart
0:10:50 > 0:10:51wristband to help people with Parkinson's disease
0:10:51 > 0:10:55draw in a straight line.
0:10:55 > 0:10:57The device's in-built motors vibrate to distract
0:10:58 > 0:11:02the user and reduce tremors.
0:11:02 > 0:11:04It was also the week that we discovered queueing
0:11:04 > 0:11:07at the shops and using those beyond infuriating self-service
0:11:07 > 0:11:09checkouts could soon be a thing of the past.
0:11:09 > 0:11:12Amazon has unveiled a sci-fi store in Seattle that uses your smartphone
0:11:12 > 0:11:15and advanced technologies like deep learning, computer vision and sensor
0:11:15 > 0:11:17fusion to automatically detect when products are taken
0:11:17 > 0:11:23off its shelves.
0:11:23 > 0:11:26When you're done, you can simply trot off and then wait
0:11:26 > 0:11:31for that gargantuan virtual receipt to follow.
0:11:31 > 0:11:34And if you feel like you're forever stuck in traffic,
0:11:34 > 0:11:38Audi has rolled out an update to make every second count.
0:11:38 > 0:11:41Its new traffic light information feature tells drivers exactly how
0:11:41 > 0:11:44long they'll have to wait behind a red signal before it turns
0:11:44 > 0:11:47green and the length of time it will stay green.
0:11:47 > 0:11:49It works by connecting directly to the city's
0:11:49 > 0:11:53traffic management centre.
0:11:53 > 0:11:58And finally, robotic research has reached new heights, literally.
0:11:58 > 0:12:02This hopping mad bot developed at UC Berkley cannot only jump a meter off
0:12:02 > 0:12:05the ground but can then again jump off objects to reach
0:12:05 > 0:12:09even greater heights.
0:12:09 > 0:12:11Inspired by the agility of bushbabies, researchers hope
0:12:11 > 0:12:17it can one day identify jumping spots for itself.
0:12:23 > 0:12:26Also in the news, the news.
0:12:26 > 0:12:32Which these days isn't guaranteed to be true.
0:12:32 > 0:12:34In the run-up to the US election, for example, the Speaker
0:12:35 > 0:12:39of the House of Representatives did not get naked, the Pope did not
0:12:39 > 0:12:44endorse Donald Trump and he did not win the popular vote,
0:12:44 > 0:12:46but these stories, from websites posing as real news sites,
0:12:47 > 0:12:50were heavily shared on Facebook.
0:12:50 > 0:12:53Of course, it doesn't help that in 2016, the real news sounds
0:12:53 > 0:12:55sounds just as made up.
0:12:55 > 0:12:58But anyway, it's made events like the Trust Hack here in London
0:12:58 > 0:13:03even more timely and urgent.
0:13:03 > 0:13:06Here, journalists and technologists from large news organisations
0:13:06 > 0:13:09are workshopping ways to help readers tell the difference
0:13:09 > 0:13:10from well-researched journalism, propaganda, advertising,
0:13:10 > 0:13:17satire and complete hogwash.
0:13:17 > 0:13:22The thought is to provide images like icons back-up materials
0:13:22 > 0:13:26that the public could see connected to a piece of news and then it
0:13:26 > 0:13:29would send a signal back to the news distribution platform like Google
0:13:29 > 0:13:33or Twitter so that they can identify quality news out of the fake news
0:13:33 > 0:13:37and elevate it in search and social.
0:13:37 > 0:13:41There are already projects afoot to try to flag up stories on sites
0:13:41 > 0:13:45known to generate fake news like this plug-in
0:13:45 > 0:13:48but the ideas here are not about blacklisting sites or producing
0:13:48 > 0:13:52automated fact-checking systems.
0:13:52 > 0:13:55Both would be massive undertakings and would themselves provoke cries
0:13:55 > 0:13:58of bias and censorship.
0:13:59 > 0:14:02This is more about letting news organisations prove to their readers
0:14:02 > 0:14:04that they are trustworthy.
0:14:04 > 0:14:08The journals of the Washington Post, I work with some amazing people.
0:14:08 > 0:14:10We produce really great stuff and they remain really
0:14:10 > 0:14:12committed to continuing that.
0:14:12 > 0:14:15The tools that we are building here are just a way for us
0:14:15 > 0:14:18to communicate that we are putting in the effort,
0:14:18 > 0:14:21where our stuff is coming from, who we are talking to.
0:14:21 > 0:14:22We have access to great sources.
0:14:22 > 0:14:26We are trying to create something that would easily allow audiences
0:14:26 > 0:14:28to verify for themselves what sources we have used.
0:14:28 > 0:14:31You'd be able to click and see, who did we talk to,
0:14:31 > 0:14:34and you'd click on that.
0:14:34 > 0:14:37Readers want to feel like journals are being held accountable to them
0:14:37 > 0:14:41and this is a way of doing that.
0:14:41 > 0:14:42We are showing our work.
0:14:42 > 0:14:45Other ideas here include ways to fight information bias
0:14:45 > 0:14:53by surfacing articles that support the opposite side of an argument
0:14:53 > 0:14:56or to check the likelihood of truth by finding similar articles
0:14:56 > 0:14:57by other outlets.
0:14:57 > 0:14:59Many think the reputation of the journalists themselves plays
0:14:59 > 0:15:01a big part in the trustworthiness of reporting.
0:15:01 > 0:15:04Italian newspaper La Stampa is suggesting a system
0:15:04 > 0:15:07where an author is assigned a unique identifier that shows
0:15:07 > 0:15:12their piece can be trusted.
0:15:12 > 0:15:12Come on.
0:15:13 > 0:15:14Best platform goes to...
0:15:14 > 0:15:18La Stampa.
0:15:18 > 0:15:22The best ideas won a small prize at the end of the day but it's
0:15:22 > 0:15:24the taking part that counts here.
0:15:24 > 0:15:26The hope is that this part-Google funded initiative may lead
0:15:27 > 0:15:30to a system that helps news outlet stories rank high up on search
0:15:30 > 0:15:33engines and social media.
0:15:33 > 0:15:36That said, I can't help wondering if that is actually something that
0:15:36 > 0:15:40platforms like Google and Facebook really want.
0:15:40 > 0:15:44Do you think they care what it is they serve to us or do
0:15:44 > 0:15:47you think, really, as long as you click on it, that's
0:15:47 > 0:15:56what they're interested in?
0:15:56 > 0:15:58How do we know the motivations of any company?
0:15:58 > 0:16:00Money.
0:16:00 > 0:16:02These companies make their money through people clicking
0:16:02 > 0:16:05on the adverts, so do you think any of the large companies care
0:16:05 > 0:16:07what it is they present to you?
0:16:07 > 0:16:10Based on my conversations with them, I think they do.
0:16:10 > 0:16:13The argument would be if they start being perceived as not caring
0:16:13 > 0:16:15about the information that's sent out there,
0:16:15 > 0:16:27that damages their reputation.
0:16:27 > 0:16:31Yes, it may be that the truth will out, not because of a desire
0:16:31 > 0:16:34for the facts, but because everyone, readers and news aggregators,
0:16:34 > 0:16:47ultimately don't want to lose face.
0:16:47 > 0:16:50More and more people are shopping online but still at this time
0:16:50 > 0:16:53of year, the high street seems pretty chaotic and the retailers
0:16:53 > 0:17:05want to make sure it stays that way.
0:17:05 > 0:17:07So they are trying to create some more engaging experiences.
0:17:07 > 0:17:10But do they help us or are they just a distraction?
0:17:10 > 0:17:13Here in London's Covent Garden, 140 shops and restaurants are taking
0:17:14 > 0:17:16part in creating one huge augmented reality experience.
0:17:16 > 0:17:19With the help of AR app Blipper, things come to life.
0:17:19 > 0:17:21It may not have created the personalised shopping experience
0:17:21 > 0:17:24I dreamt up, but there were some promotional offers presented
0:17:24 > 0:17:26as virtual Christmas presents almost around the tree.
0:17:26 > 0:17:30A reindeer hunt and a giant reindeer you can take a selfie with,
0:17:30 > 0:17:41if the mood takes you.
0:17:42 > 0:17:45I have to say, it wasn't quite as cutting-edge as I'd hoped
0:17:45 > 0:17:47but I suppose it's a bit of light-hearted fun.
0:17:47 > 0:17:50Rather more purposefully, the signs in windows can be scanned
0:17:50 > 0:17:52using image recognition, taking you to online content,
0:17:52 > 0:17:56partly the sort of stuff you'd be able to look at from your sofa
0:17:56 > 0:17:58but with a few extras to boot.
0:17:58 > 0:18:00Then came our trip to London's Westfield where augmented
0:18:00 > 0:18:06reality has had a makeover.
0:18:06 > 0:18:09We've seen technology like this before but now it's actually
0:18:09 > 0:18:11on the shop floor here at Charlotte Tilbury.
0:18:11 > 0:18:14This is what's known as the Magic Mirror and this
0:18:14 > 0:18:33is what it does.
0:18:33 > 0:18:36You choose a lipstick and in real-time, you will see
0:18:36 > 0:18:37your face transformed.
0:18:37 > 0:18:40Bright red lips, although it doesn't seem to have any around the edges,
0:18:40 > 0:18:42I think that looks all right.
0:18:42 > 0:18:45I've tried the Rimmel app that does something similar on your phone
0:18:45 > 0:18:49and you can buy things through it, but here you can do it in the store
0:18:49 > 0:18:52with assistants all around and a whole shop of products that
0:18:52 > 0:18:55you can test, smell, and after you see what your face
0:18:55 > 0:18:59looks like on here, you might want to have a go to check that
0:18:59 > 0:19:01you can do it yourself that well.
0:19:01 > 0:19:04Meanwhile, here at this eBay event, they are taking things
0:19:04 > 0:19:04a step further.
0:19:05 > 0:19:06They want to get inside your mind.
0:19:06 > 0:19:09The data on this screen represents what is apparently visitors
0:19:09 > 0:19:10emotional responses to products.
0:19:10 > 0:19:13Using what they call facial coding, the camera looks for reactions
0:19:13 > 0:19:16which these guys reckon you have when you do online shopping.
0:19:16 > 0:19:19Nice but don't know who I'd give it to.
0:19:19 > 0:19:24Interesting.
0:19:24 > 0:19:24That's ridiculous.
0:19:25 > 0:19:27And I've been told to overact my reactions.
0:19:27 > 0:19:28Oh, that's cute.
0:19:28 > 0:19:31Whilst my results bore absolutely no correlation to what I'd liked,
0:19:31 > 0:19:34maybe they were the ones I contorted my face to the most.
0:19:34 > 0:19:37Maybe it would have worked better if I'd reacted more naturally,
0:19:37 > 0:20:09although I struggle to imagine that my face would have shown anything.
0:20:09 > 0:20:12Now, in 2015, members of the United Nations adopted a set
0:20:12 > 0:20:13of sustainable development goals.
0:20:13 > 0:20:17Number one on that list is to end poverty and to achieve that goal,
0:20:17 > 0:20:20you first need to work out where poverty exists and how
0:20:20 > 0:20:33to measure it.
0:20:33 > 0:20:36We met up with some scientists at Stamford who have that
0:20:36 > 0:20:36task in hand.
0:20:36 > 0:20:39Marshall Burke is a professor of earth systems science
0:20:39 > 0:20:42at Stanford University but he spends much of his time in Africa
0:20:42 > 0:20:43gathering poverty data.
0:20:43 > 0:20:46The way this is done is to elicit from the household a listing
0:20:46 > 0:20:49of everything they've consumed in the last week,
0:20:49 > 0:20:49the last month.
0:20:49 > 0:20:51So literally everything they've consumed.
0:20:51 > 0:20:55Every single thing and the value of that item and then you add up
0:20:55 > 0:20:57all these items for every single person in the household.
0:20:57 > 0:21:00This can take hours and hours just for one single household.
0:21:00 > 0:21:04Then you have to do this for thousands of households to get
0:21:04 > 0:21:05a representative sample of the area.
0:21:05 > 0:21:08It's painstaking work but Burke has teamed with computer science
0:21:08 > 0:21:18students to test a new method.
0:21:18 > 0:21:20Using machine learning to predict poverty data
0:21:20 > 0:21:21by analysing satellite images.
0:21:21 > 0:21:24But to find out whether the people living in those areas are rich
0:21:24 > 0:21:27or poor, the researchers used a process called transferred
0:21:27 > 0:21:32learning and this image of the Earth at night.
0:21:32 > 0:21:36The parts of the world that are lit up are typically the wealthier parts
0:21:36 > 0:21:42of the world.
0:21:42 > 0:21:45So basically we use the lower resolution night-time images to help
0:21:46 > 0:21:49us figure out what in the really high resolution daytime images
0:21:49 > 0:21:52we should should be using and then we use that to predict poverty
0:21:52 > 0:21:58on the ground.
0:21:58 > 0:22:01Between 300 and 400,000 images were used to train the algorithm.
0:22:01 > 0:22:03The algorithm will figure out what's important,
0:22:03 > 0:22:05what it should be looking for.
0:22:05 > 0:22:07So some of the things it finds are things that
0:22:07 > 0:22:09you or I would recognise, things like roads,
0:22:10 > 0:22:10urban areas, farmland.
0:22:10 > 0:22:13Based on those features, the algorithm can predict
0:22:13 > 0:22:14what a family owns.
0:22:14 > 0:22:16Things like refrigerators, cars, the sum of all those assets.
0:22:16 > 0:22:19It can also be used to predict incomes.
0:22:19 > 0:22:20These poverty maps show the team's findings.
0:22:20 > 0:22:23In areas marked red, people spend as little
0:22:23 > 0:22:24as $1.5 a day.
0:22:24 > 0:22:27In green regions like Uganda's capital, Kampala, they spent
0:22:27 > 0:22:29closer to $8.
0:22:29 > 0:22:32We are providing a very cheap and scalable alternatives
0:22:32 > 0:22:34to traditional means of data collection.
0:22:34 > 0:22:36Traditionally, you have to send people out into the field
0:22:36 > 0:22:38with clipboards, the surveys aren't always accurate,
0:22:38 > 0:22:47there can be corruption.
0:22:47 > 0:22:50Like for example lots of governments where they are underperforming,
0:22:50 > 0:22:51they don't want true surveys.
0:22:51 > 0:22:54All we need to make our predictions are satellite images.
0:22:54 > 0:22:57But can you draw conclusions about the economic well-being
0:22:57 > 0:22:59of communities in Africa when you're thousands of miles away,
0:22:59 > 0:23:04sitting at a laptop in an office at Stamford?
0:23:04 > 0:23:07We actually have really good survey information in a few locations.
0:23:07 > 0:23:10We can use the satellite imagery to make a prediction about poverty
0:23:10 > 0:23:14and then we can compare that to what the survey says was actually
0:23:14 > 0:23:15going on on the ground there.
0:23:15 > 0:23:19So we used a couple of the really good surveys we had to validate
0:23:19 > 0:23:20our satellite input.
0:23:20 > 0:23:23To be a truly useful tool though, the algorithm needs an upgrade.
0:23:23 > 0:23:27We would also like to use historical imagery so maybe we can figure out
0:23:27 > 0:23:30how poverty dynamics work overtime and even give us the chance
0:23:30 > 0:23:34of predicting what's going to happen in the future.
0:23:34 > 0:23:38If you can pinpoint poverty on a map, aid could be distributed
0:23:38 > 0:23:39more evenly, policies could be more effective.
0:23:39 > 0:23:41People would get help faster.
0:23:41 > 0:23:44A picture may be worth a thousand words but combining that picture
0:23:44 > 0:23:47with artificial intelligence could make a world of difference.
0:23:47 > 0:23:49That was Sumi and that's it for this week.
0:23:49 > 0:23:52You can follow us on Twitter @BBCClick for backstage fun
0:23:52 > 0:23:58and photos and extra technology news throughout the week.