Not just a pretty face

A couple of years ago we were dazzled by the release of image recognition tools such as Google Goggles, which, for the first time, allowed regular users to perform instant image-based web searches by using the camera in their hand-held devices. The technology was touted as a Swiss Army knife of visual recognition (applicable to anything from bar codes to restaurant menus in foreign languages).

A bit of road testing soon made it clear that it wasn't great at distinguishing, say, a dalmatian from a poodle, or a Wedgewood plate from a Clarice Cliff. However, it was very good at doing one thing in particular -- identifying individual art works and providing their name and author.  This made it quite a useful app to have in those annoying situations when, flicking through a magazine, you come across a famous painting whose artist you can't quite remember. You simply point your camera at the picture, tap on the "Go" button, and hey presto: title, date and artist.

Museums like the New York Met and the Getty started a series of collaborations with Google, providing metadata for thousands of paintings. Although the app was certainly very handy (and a great talking point in the pub), its pedagogical value was still somewhat limited during those early stages of development -- the best one could hope for was a couple of basic facts about the painting and a list of Google search results (some more relevant than others).

Google Googles at work (Image © by Google).

However, that is changing fast. Educational researchers and IT developers are creating new ways of digging through the data, contextualising, meshing up, inter-linking and reshaping results. It is these developments that can turn visual search technology from a gimmicky app to a powerful educational / research tool.

The University of California Riverside, for example, is developing an ambitious facial recognition project designed to identity individual historical characters portrayed in paintings. The principle is similar to Facebook's infamous facial recognition / photo tagging technology (without the Big Brother implications).

Applied to historical portraits via the Google Goggles infrastructure, this UCR tool is expected to provide answers to questions such as: Who is this person? Where did s/he live? What stage of his/her life is s/he in in this portrait? What was happening in the world at that time? What is his/her facial expression? What other portraits exist of this individual? Who else is in the painting with him/her? What webs of relations can we tease out based on people's associations in different portraits?

UCR's project is still at a very early stage of development and there are plenty of obstacles to overcome before the tool is sufficiently stable and useful. However, it is an encouraging example of the kind of research that is helping us make the crucial transition from visual recognition to visual data mining (and ultimately visual data analysis) in mobile learning.

Mobile learning or drive-by learning?

According to Chinn and Fairlie, in 2001 there were 61 computers per hundred people in North America, but only 0.5 per 100 in South Asia. In response to this imbalance, the One Laptop per Child (OLPC) project took as its mission “to create educational opportunities for the world’s poorest children by providing each child with a rugged, low-cost, low-power, connected laptop with content and software designed for collaborative, joyful, self-empowered learning”.

OLPC’s underlying premise is that the large scale distribution of Information and Communication Technology among the less privileged will tackle problems of accessibility to computers and simultaneously improve IT literacy rates. OLPC thus links access with use and practice, relying on models of self directed learning.

OLPC love. (Image: CC by laihiu.)

The project has been welcomed by various NGOs and HE institutions. However, it has also attracted its fair share of controversy. Nicholas Negroponte, founder of the OLPC project, takes the principle of self-directed learning to its logical extreme. Much to the consternation of some teachers and educational researchers, he has boldly summed up the association between ICT access and use with the phrase “you can give kids XO laptops and just walk away”.

From Negroponte's point of view, the availability of ML resources alone is enough to encourage the development of valuable learning skills among children. These skills, in turn, will have a positive impact on interrelated socio-economic factors, reducing multiple forms of deprivation such as poverty, social exclusion and illiteracy, not just among the children but also within their broader community.

From the point of view of many teachers and educational researchers, Negroponte's model (which has been dubbed "drive-by learning") is deeply flawed. Efficient learning skills are unlikely to materialize out of thin air just because the right set of tools are placed in front of the student. A laptop is a great tool in the classroom, but a poor substitute for a teacher. Some OLPC insiders have recently joined the sceptics by putting a big question mark over the drive-by approach as a result of a damning assessment of failures in the project's implementation across Peru.

The OLPC project is an inspirational and ambitious attempt to reduce the global digital divide and bring socio-economic advantages to deprived communities. The world needs more projects like OLPC, but we also need to work harder to ensure that their underlying pedagogical principles are right.

Whitewashing the ML fence

One of the reasons behind some of the scepticism towards ML is the popular perception that this form of education leads to an over-reliance on the very resources that make it possible. Students, some argue, are reaching a stage where they aren't able to achieve anything for themselves without the use of their smartphone or tablet.


On one end of the spectrum, there are those who argue with horror that modern pupils do not bother to learn facts any more, because they can simply google anything up on the spot if and when they need the information. The implications are that ML students don't really know anything themselves -- they just know where to find someone (usually in the form of a digital resource) who does know.

On the other end of the spectrum, there are those who don't see a problem with the idea of not learning facts. We don't memorise all our friends' phone number or all the times and venues of the meetings we need to attend this month (we simply use a calendar or a diary to remind us), so why should education be any different?

Tom Sawyer whitewashing the fence.

When the problem is approached from a pedagogical point of view, most people take a middle ground. In many ways, developments in ML do not challenge modern attitudes to teaching -- in fact, they complement approaches to pedagogy that were already well established way before tablets and mobile phones were around.

In the 1960s, there was a radical change in instruction strategy: a shift from test-driven transmission teaching (conceived as a top-down, mental transfusion from the teacher to the pupil) to exploratory, hands-on and problem-centred approaches. The maxim became: learning is not the same as memorising. Encouraging students to develop useful learning skills became more important than hammering raw data into their brains. Aspects like observation, critical thinking and communication started making their way to the top of the curriculum.

This led to decades of debate among educational researchers. People like Paolo Friere and John Dewey argued against rote learning and for an active, dialectical, student-centred approach, while E.D. Hirsch, Jr. argued against what he termed their naturalist perspectives and for the core root of facts necessary for domain specific knowledge. Today, most teachers see these apparently conflicting positions as two sides of the same coin -- complementary aspects of the learning process. Developing knowledge is still an important part of education, but learning to develop that knowledge and what to do with it (i.e. the 4 Cs -- critical thinking, creative thinking, communicating, and collaborating) is also crucial. This aspect of education is what educational researchers call 'learning skills'.


As well as providing access to online repositories full of facts, ML resources allow students to develop a broad range of transferable learning skills as they classify, analyse and evaluate content (assessing the reliability of particular online resources, collating useful data from a range of platforms and learning how to adequately reference online and offline materials); they participate in problem solving activities; imagine, design and create their own resources; and collaborate with groups of people, setting goals and exercising team building skills.


Although ML is a very young field in education, its underlying principles are  not a radical departure from the teaching methods that educational researchers have been developing since the 1960s. Contrary to popular perception, in many ways ML complements and extends a long established pedagogical tradition.

Galileo and ML

Galileo is a global positioning system that is currently being deployed by the European Space Agency. It's made up of a network of satellites which will complement and integrate with the functionality of existing international positioning systems. Despite a fair amount of funding headaches (and a little tantrum thrown in by the USA), the first Galileo satellite was successfully launched in 2005. The system is not expected to be fully operative until around about 2015.

Galileo Galilei. Portrait by Ottavio Leoni, 1624.

Compared to current global positioning systems, Galileo introduces a few snazzy features. However, the feature that really stands out (in terms of its implications for ML) is its level of precision. Galileo will be accurate down to the metre range. Current general-purpose GPS receivers have an accuracy range of about 10-20 metres (or even worse at high altitudes). This means that, in practice, they only work as navigation systems for large, open areas. For example, integrated with a ML smartphone app, these systems are able to tell us where the nearest open museum is, what building we are looking at, or what part of town we are in.

With Galileo, provided a clear satellite signal is available, mobile devices will be able to detect which display case museum visitors are standing by (and provide personalised content at point of consumption); generate diagrams showing how people move around the gallery on a metre-by-metre basis and how long they spend interacting with each individual display (facilitating user data analysis for curators and events organisers); they could allow participants in educational crowdsourcing projects to geotag locations / learning resources / interesting features in libraries, archaeological sites, art galleries, etc. with a high level of precision (making it easy for other users to go back to that precise spot, find the resource and benefit from it); they could provide a new set of tools for blind students to engage with educational resources spatially in a classroom or lecture hall.

By reducing the error margin of existing geolocation systems from 20 metres to <1 metre, Galileo could open a massive range of new options for teachers and students in mobile learning.

Mobiles in numbers

A few quick stats about the current state of the mobile industry:

-The global mobile phone user base is growing exponentially

There are approximately 6 billion mobile phone subscribers in the world. It took 20 years to reach the first billion mobile phone users, but only 15 months to reach the last billion users.

-Most mobile phone users do NOT speak English

Many ML resources are created with Western English speakers in mind, but these users account for less than 10% of the global user base. The fastest growing world regions in terms of mobile phone usage are China and India (which, together, make up 30% of the global user base).

-The amount of internet resources accessed through mobile devices is also growing exponentially.

70% of all phones sold in the USA are internet-enabled smartphones (this could reach 90% by X-Mas). Less than 50% of all mobile traffic is voice. 60% of all Twitter traffic and 70% of all Pandora traffic come from mobile phones.

-The mobile phone sector is highly volatile.

Around about a year ago, the word ‘iPhone’ was synonymous with smartphone. In fact, things were looking quite rosy for Apple – the iOS market share had grown by a whopping 115% in 2011. Today, Android has more than doubled the smartphone user base of iOS. Half a million Android handsets are activated every day. Blackberry, the brand that epitomised flashy corporate smartphones merely four years ago, is now looking at a global market share of potentially less than 5% next year.

 

Image: CC by Irargerich

 What does this tell us? 

• Most internet resources / services that are not adequately accessible through mobile devices could risk disappearing in the medium / long term. 

• The global mobile phone user base is enormous, but cultural and linguistic differences make it difficult to create resources that have a global appeal.

• The demand for mobile resources, including ML, is likely to grow exponentially in the short and medium term.

• The high level of volatility in the mobile industry means that platform-specific resources (e.g. apps) could have a relatively short shelf-life and limited scalability. Web-based mobile resources might be a more attractive prospect for long-term projects.

The battle of the ages

The Nintendo 3DS console was introduced to the Louvre last March with much fanfare. You might think that this games console is more likely to be associated with mustachioed Italian plumbers than the Venus of Milo. Well, on this occasion it is conceived as an audio guide on steroids. It provides museum visitors with interactive maps of the museum, suggests themed itineraries for different visitor profiles, complements the displays with hundreds of commentaries recorded in seven different languages, provides high resolution images of paintings that may be difficult to look at closely in the galleries (like the Mona Lisa) and even generates 3D reconstructions of statues. According to Agnes Alfandri, Louvre’s head of multimedia, less than 4% of visitors were using the old audio guide, so it was imperative to galvanise the system in order to keep up with the times.

Interestingly, user feedback suggests that the Nintendo 3DS guide may have received mixed reactions. In fact, it seems to have drawn a wedge between two group categories: the under-thirties (approximately 1/3 of all visitors), who have responded quite positively to the new system, and the over-thirties (approximately 2/3 of all visitors), who appear to be somewhat less enthralled by the entire thing. 

Image: Left CC by bixentro. Right CC by Matthew Miller (modified)

This battle of the ages is being fought on three fronts:

1. Games consoles in a museum

• Under-thirties: many in this category see museums as intimidating, alien environments. Games consoles make museums appear less serious, more approachable and engaging.
• Over-thirties: the idea of using games consoles in a museum is about as appropriate as bringing a vuvuzela to a funeral.

2. Ease of use

• Under-thirties: The console and its interface are very intuitive and easy to use.
• Over-thirties: The console and its interface are very alien and difficult to use.

3. Eye candy

• Under-thirties: The 3D reconstructions of artefacts are cool.
• Over-thirties: I’m standing in front of the actual statue. What is the point?

Of course, these are gross generalisations, concealing an even broader and more complex range of reactions to the system. However, the initial user feedback does highlight two important aspects about ML and some of the challenges faced in its implementation:

1. ML (and digital learning resources in general) have the capacity to reach a broad user base, but these users have very different needs, tastes and priorities.

2. The digital divide is not just about being able to access ICT, it’s also about developing the skills and knowledge required to use it efficiently.

ML in museums

Mobile Learning is an efficient way of introducing educational material to non-traditional learning environments (e.g. bus stops, lunch breaks and that unbearable speech that drags on for hours). However, some of the most exciting applications of ML (particularly in the Arts) are taking place in very traditional environments – museums. Why?

-Museums are highly mobile environments.

Unlike classrooms or conference rooms, most museum displays require people to spend a lot of time walking around.

-Museums are great for learning...

They contain a wealth of artefacts and information that are directly relevant to most academic disciplines.

-...but learning doesn’t happen by itself.

Visitors need to be engaged with these artefacts and information, which is one of the biggest challenges of museums. ML has the potential to add an interactive element to the museum visit, attracting people to displays, presenting information in a new light, providing new tools for the interpretation of material and contextualising data to encourage further exploration. This is the principle of engagement through interaction.

-ML could help solve an old problem in museums.

Traditional information panels struggle to please all types of museum visitors (divers, swimmers and skimmers). ML can provide contextualised and personalised information – content at the point of interest. In principle, visitors can consume as little or as much of this content as they wish. Since the material is accessed through their mobile device, it does not take up precious space in display cases or the walls and it is easier and faster for the museum to update or replace the data.

-Most museums are keen on new technologies

Curators are generally interested in attracting lots of visitors and shedding the image of museums as stuffy, snooty temples of crystallised knowledge. Thanks to aggressive marketing campaigns, mobile devices are seen as fun and edgy. They appeal to young people and help stimulate the public’s imagination.

-ML makes sense financially

Visitors bring their own mobile device to the display (which reduces the upfront cost for the museum). In addition, mobile phone operative systems provide a ready-to-use infrastructure for the sale of content (e.g. Google Play, iPhone App Store, etc.).

-ML can be implemented on various levels

ML solutions can range from merely printing a few QR codes (which are free to produce and can be ready to use within minutes) to more sophisticated solutions (such as augmented reality apps, involving a team of coders, digital artists and months of preparation).

Image: CC by Conxa Roda