‘‘ A
Much of our thinking about AI is coloured by science fiction visions of what AI researchers call ‘Artificial General Intelligence’ – human level capabilities.
INSIGHT
Future Technology New science from old books
RTIFICIAL Intelligence (AI) is all around us in the apps and websites we use every day, but there are a lot of areas
where AI isn’t being used at all yet, or it is still in its infancy. In this article I’ll look at the potential for AI to help us understand more about archives, rare books and manuscripts, try to demystify what AI can and can’t do, and consider the potential bear traps and tar pits that lie in wait for the unwary. Earlier this year there was excitement amongst Biblical scholars after AI based analysis of the handwriting of the Dead Sea Scrolls showed that the Great Isaiah Scroll (a copy of the Book of Isaiah found in both the Hebrew Bible and the Old Testament) was written by two scribes with very similar handwriting. This is a great example of the kind of thing AI is good at – pattern matching and classifying.
Much of our thinking about AI is coloured by science fiction visions of what AI researchers call “Artificial General Intelligence (AGI)” – human level capabilities. The reality is more mundane. Even much vaunted AI implementations like Tesla’s self- driving cars achieve their seemingly magical capabilities by plugging this pattern recognition capability into conventional computer code: “Take a picture and pass it to the AI for classification. If the AI says the picture includes a red traffic light, stop the car safely and keep taking and classifying pictures until the AI says the light has changed to green”, and so on. As the Dead Sea Scrolls example shows, this similarity checking capability can be used to achieve things that would be prohibitively slow or costly for humans to do by hand. Sometimes the AI approach doesn’t work out as planned, though, and we need to be aware of the ways
September 2021
that it can go wrong. In one case it turned out that a medical AI had learned to recognise imagery produced by a particular model of scanning device, rather than the tumours in the pictures. Likewise, Google famously discovered to their horror that an AI model for automatically categorising Google Photos images into albums had classified dark skinned people as gorillas due to deficiencies in its training data.
Choosing the right problem In the world of special collections and rare books there may be large amounts of material that could be used as training data for an AI model, such as in an author’s archive. Shakespearean scholars recently used AI to gauge the extent to which Shakespeare’s contemporaries John Fletcher and Philip Massinger might have contributed to Henry VIII. In this case the researchers used 53 training samples of what is believed to be Shakespeare’s own wording, 90 from Fletcher’s published works and 46 from Massinger’s. But before we get too excited about the potential of AI, we need to understand what the problem is that we are trying to solve. For example, the Dead Sea Scrolls researchers were interested in understanding whether the Great Isaiah Scroll was the work of one person, and the Shakespearean scholars wanted to understand whether there were any elements of Henry VIII that bore the hallmarks of Shakespeare’s contemporaries. Both of these problems lent themselves to an AI enabled approach, but not all problems will. We also need to keep in mind that it can be very difficult to understand how a neural network, the core AI technology, decided that two items were similar in the first place. And until we manage to figure out AGI, we can’t ask the AI to explain its conclusions. This is why it’s important to have a
Martin Hamilton is a strategist and innovation adviser. See
https://martinh.net for his portfolio and contact details.
human validate any conclusions or recommendations reached by an AI rather than blindly accepting them as fact. Computer scientist Toby Walsh has proposed that a “Turing Red Flag” be legally mandated to highlight truly autonomous systems – a latter day equivalent of the 19th century Locomotive Act, which required motorised vehicles to be preceded by a person waving a red flag to signal the coming danger. And finally, whilst we might not see ourselves as AI experts the truth is that nowadays we are all AI trainers already. Sometimes this is more obvious than others. For instance, when we are presented with website CAPTCHA prompts and asked to label traffic lights and fire hydrants we are actually supervising the AI’s learning process for a future self-driving car system. Perhaps it’s time we started to explore what the technology could do to help us better understand our archives and special collections? IP
This is a companion piece to my presentation at the CILIP Rare Books and Special Collections Group 2021 Study Conference earlier this month and a talk on 22 September at this year’s Internet Librarian Conference
www.internet-librarian.com/2021.
INFORMATION PROFESSIONAL 23
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60
