ALIA Online 2019 – Day 2

Day 2 – Wednesday 13th February 2019

Connecting with users and enriching the library experience in the digital age – Carla Hayden (Librarian of Congress)

The importance of reading can not be underplayed. In USA history, African Americans who learnt to read were severely punished, as were the people who taught them to read.

Palaces for the people” – Eric Klinenberg (writer for New York Times). He writes that there needs to be a third place, a safe place for people to go to. To restore civil society, he says that they need to start with the library. Libraries need defending in the digital age – but they act as bedrocks in civil society. Libraries are universal equalisers.

Pew Internet reported that more than half of over 16s have visited a library in the last twelve months. The biggest growth audience is Millennials. There are more branch libraries than there are McDonalds.

When we are asked why we need libraries in the digital age? – point to examples around the world. Libraries represent accessibility, respect for people, provide the best that a community can offer, they lessen inequality and are essential any democratic society. Not too many places that offer services and don’t’ expect anything in return.

In the digital age, we are even more providers of additional digital content, provide expertise in the use and access of technology and we bridge the gap between technology and users.

Our goal now is to bring the library to the users, virtually. Need to give our users access to our services and resources in the most convenient way for our users.

Libraries are now lending materials that are non-traditional, such as briefcases, fishing poles, cake pans and more. Libraries are all about meeting community needs.

Still need to ensure that we are a trusted and authoritative source of information in this digital age. People need to know that what they find is not only true, but also the best source for the content they need. A book, a laptop or tablet is a rich gift for those who are seeking information.

Its most important to be neutral – but it is our job to present different points of view – “let the books battle it out themselves on the shelf”. We must withstand calls from censorship, regardless of how uncomfortable some people may be as a result. We are not neutral about inaccurate information though, we must provide the best information we can.

Motors of sustainable development: libraries and the UN 2030 Agenda – Gloria Perez-Salmeron – IFLA President

The United Nations Millennium Goals came out with no reference to the libraries and how they can impact communities and cultures, even with their IFLA Trend Reports. The next trend report will be in 2020.

IFLA WLIC 2014, Lyon –Lyon Declaration about Access to Information. It resulted in being added to Goal 16 (section 10) in the UN Agenda 2030. It meant that each library association could work with their governments to incorporate it into their national goals.

Libraries and the Sustainable Development Goals – a storytelling manual.


Vision for a globally united library field – Gerald Leitner – IFLA Secretary General

Headquarters in The Hague – team from 18 countries. 60 standing committees, 1200 experts, standards, guidelines and support – all available in 7 key languages – Russian, Spanish, Chinese, French, German, English and Arabic.

Change produces infinite possibilities, but it can also produce winners and losers, so the more we work together, the less chance that anyone loses.

From Jeff Bezos at Amazon: “The internet is disrupting all businesses” and “Complaining is not a strategy.”

Our future is bright, but whatever it is, it needs to be sustainable and it needs to be together. “The best way to predict your future is to create it together.” Abraham Lincoln. “Build the future you never do it alone.” Genevieve Bell.

We can’t leave it to others to decide our future – libraries will be the best they can be when we work together.

Global Vision Ideas Store- biggest ideas store for libraries. Already contains nearly 1000 submissions, 170 reports and so much more. Already have 8500 ideas from all over the world. The goal is to inspire librarians to act, to help support and achieve the IFLA Strategy and Actions 2019-2024.

Feel free to take ideas, but also contribute for others to benefit from.


Machine Vision, human gaze: deep learning and cultural heritage – Peter Leonard

How do we negotiate this moment when cultural heritage connects with artificial intelligence?

Researchers have a good handle on close analysis, but machines can really help us with collection level analysis.

Experiment 1 – Can part of an image stand in for the whole? Inspired by the real face of White Australia, they used thumbnails of people to represent collections of photos taken during the 1940s.

Issues were that it supported out own biases and that there were issues with racial classification.

Convolutional neural networks are what is used by our phone photo search functions. It looks for curves, straight lines, that it has been taught represents objects and uses that to make an assessment about the image.

Why can’t we use these on cultural heritage material? What could go wrong? If this is used for the modern world, it will be unable to correctly interpret historic material. They are optimized for modern materials, modern devices and monetizable use cases. The training data and categories makes the difference.

Experiment 2 – Can pre-trained captioning networks be useful without their categories? Tried it out on a Civil War photographic collection – the Meserve-Kunhardt Collection. They used Google’s Inception, discarding the last layer, taking the penultimate layer – which is far more abstract, it sees in 2048 ways. The aim was to approximate its nearest neighbours in the high dimensional space. A lot more successful.

It surfaces pictorial tropes in large archives but is a purely visual recommendation engine. But is it a product or a feature and the most unique images show the worst similarity results and it needs a large collection?

Experiment 3 – Can large visual collections organise themselves? The result built on Experiment 2 and was called PixPlot. It was used initially with Yale’s 31,000 British artworks, but has since be used for photographs,

It produced a truly unique collection view (through your browser), breaks the tyranny of the next button and is fun to demonstrate. However, future versions will allow you to make the final step and label groups of images that the computer has collected them.

Experiment 4: What happens when neural networks go to sleep? To explore that they started playing with Generative Models. They played with recipes and ST: TNG scripts, before moving to Generative Adversarial Networks. Eg. Forger looks at thousands of photos and tries to dream up new ones. Detective looks at them and tries to determine the fakes. Each system learns from the accuracy or not of the other. Showed a demo of the two systems working live.

Interesting that an algorithm could hallucinate, without knowing anything of the human face. It is also part of an emerging conversation about fakes. Ethical considerations including data problems (relationship between photographer and subject), identity and its presentations, stereotype and caricature.

Our responsibility: thoughtfulness about what we choose to experiment on.

Promising areas: large scale datasets from combined collections, placing human knowledge at same level as machine, producing and finding the captioning models that reflect the complexity of images.

PixPlot is under the MIT Licence – Yale DH Lab on GitHub. (Python script)