Partly thanks to Hollywood, artificial intelligence (AI) is, for many, synonymous with robots, mass surveillance, and even an existential threat to human civilisation. In the real world, it’s far less explosive – although it has tremendous practical potential.
In libraries, classification of images, articles, books and collections is labour-intensive, time-consuming work. The potential of AI to reduce that human effort is significant, not only for librarians, but for all of us who benefit from rich information discovery.
Calling a spade a spade
In machine learning – which is at the heart of AI – a classifier algorithm can ‘learn’ to recognise patterns that are common to a particular class. In simple terms, if we provide enough examples of what a cat looks like — or a book or a sunflower — then, when presented with a new image of a cat, a book, or a sunflower, the machine can put it in the class.
“The potential of AI to reduce that human effort is significant, not only for librarians, but for all of us who benefit from rich information discovery”
On the surface, that seems easy enough – but images can be complex. One person might look at an image of a child holding a book and label it ‘book’. Another might ignore the book completely and focus on the child’s ethnicity, or the fact that they are in a park. Classification forces us to make decisions, and those decisions reflect our world view and context. Expert training and contextual reference is missing from the creation of data that informs so many of the AI systems in use today.
A sporting chance of success
AI systems are also limited by the boundaries of the data sets they are trained on. For example, an image classifier returned the label ‘basketball’ for a photograph of President Barack Obama holding a football. Why? What pattern has the machine ‘learned’ from the example images in the class ‘basketball’?
The researchers looking critically at this misclassification assumed the problem was due to an overabundance of images in their data set showing black people with a basketball. In fact, the set had an equal number of white basketball players. The problem was actually that black individuals were under-represented overall. What this shows is that, when data sets are gathered, there are limits to what we can learn from them based on their composition. That’s why collection development, ensuring broad representation, is an essential part of the library system – and we need to bring that expertise to AI.
A time to rethink and re-see
While AI technology is an incredibly valuable tool to automatically cluster and classify information, we need to take that capability and apply it with human expertise to use it in a refined and effective way. Having humans designing and verifying those processes is therefore extraordinarily important. As I highlighted in my session at Jisc’s Digifest event last month, this is a moment for libraries to rethink and re-see. How can we make the best use of the technologies and the people that we have? I can’t see without my glasses, but they don’t see for me, they just help me see better than I otherwise could. That’s the role of AI in libraries.
A safe space
A library is a trusted place of reliable information, and the role of the human beings in that environment is increasingly recognised as crucial. I’ve seen people from industry come to libraries, archives and museums cap-in-hand asking for help because, until very recently, when it came to training AI models, the more data we had, the better. Now, there’s a realisation that it’s not just about the quantity of data – it’s about the qualities of data, and consideration for how it has been shaped and curated.
There’s also the physical experience. Libraries provide a place to interact and access expertise. Technology should reflect that ethos. AI isn’t something to be suspicious of in this context. Rather, it has huge potential to support librarians’ work. Once we embrace that fact, we can build nuanced and sophisticated systems, making this technology our own.
Jisc’s CNI leaders conference, 7–9 July 2021, will further explore ‘Frontiers in research practice – the university library as a catalyst’.
You might also like: VR doesn’t just belong in the classroom – it is the classroom of the future
Advertisement / Google
Advertisement / Google
‘Libraries need to rethink the role of technology – here’s why’
Catherine Coleman
Partly thanks to Hollywood, artificial intelligence (AI) is, for many, synonymous with robots, mass surveillance, and even an existential threat to human civilisation. In the real world, it’s far less explosive – although it has tremendous practical potential.
In libraries, classification of images, articles, books and collections is labour-intensive, time-consuming work. The potential of AI to reduce that human effort is significant, not only for librarians, but for all of us who benefit from rich information discovery.
Calling a spade a spade
In machine learning – which is at the heart of AI – a classifier algorithm can ‘learn’ to recognise patterns that are common to a particular class. In simple terms, if we provide enough examples of what a cat looks like — or a book or a sunflower — then, when presented with a new image of a cat, a book, or a sunflower, the machine can put it in the class.
On the surface, that seems easy enough – but images can be complex. One person might look at an image of a child holding a book and label it ‘book’. Another might ignore the book completely and focus on the child’s ethnicity, or the fact that they are in a park. Classification forces us to make decisions, and those decisions reflect our world view and context. Expert training and contextual reference is missing from the creation of data that informs so many of the AI systems in use today.
A sporting chance of success
AI systems are also limited by the boundaries of the data sets they are trained on. For example, an image classifier returned the label ‘basketball’ for a photograph of President Barack Obama holding a football. Why? What pattern has the machine ‘learned’ from the example images in the class ‘basketball’?
The researchers looking critically at this misclassification assumed the problem was due to an overabundance of images in their data set showing black people with a basketball. In fact, the set had an equal number of white basketball players. The problem was actually that black individuals were under-represented overall. What this shows is that, when data sets are gathered, there are limits to what we can learn from them based on their composition. That’s why collection development, ensuring broad representation, is an essential part of the library system – and we need to bring that expertise to AI.
A time to rethink and re-see
While AI technology is an incredibly valuable tool to automatically cluster and classify information, we need to take that capability and apply it with human expertise to use it in a refined and effective way. Having humans designing and verifying those processes is therefore extraordinarily important. As I highlighted in my session at Jisc’s Digifest event last month, this is a moment for libraries to rethink and re-see. How can we make the best use of the technologies and the people that we have? I can’t see without my glasses, but they don’t see for me, they just help me see better than I otherwise could. That’s the role of AI in libraries.
A safe space
A library is a trusted place of reliable information, and the role of the human beings in that environment is increasingly recognised as crucial. I’ve seen people from industry come to libraries, archives and museums cap-in-hand asking for help because, until very recently, when it came to training AI models, the more data we had, the better. Now, there’s a realisation that it’s not just about the quantity of data – it’s about the qualities of data, and consideration for how it has been shaped and curated.
There’s also the physical experience. Libraries provide a place to interact and access expertise. Technology should reflect that ethos. AI isn’t something to be suspicious of in this context. Rather, it has huge potential to support librarians’ work. Once we embrace that fact, we can build nuanced and sophisticated systems, making this technology our own.
Jisc’s CNI leaders conference, 7–9 July 2021, will further explore ‘Frontiers in research practice – the university library as a catalyst’.
You might also like: VR doesn’t just belong in the classroom – it is the classroom of the future
Advertisement / Google
2021: Get students excited about STEM
2021 was an overwhelming year during which we faced endless unprecedented challenges. Yet, it was…
HRH Princess Anne debuts Covid Unsung Heroes awards for women in STEM
The Princess Royal has announced six winners of the Women in Science and Engineering (WISE)…
Global bodies sign letter calling for governments and businesses to improve access to edtech
The power of edtech partnership
English Mastery receives strong backing from teachers
10 Ways to Become a Greener School this Earth Day