際際滷

際際滷Share a Scribd company logo

Error or Serendipity: a small experiment with AI at Royal Pavilion & Museums

Download as PPTX, PDF
F
fauxtoegrafik

Presentation discussing an experiment with computer vision tagging of collection images. It looks at the appeal of AI for museums with limited resources, and how the frequent errors produced by this technology suggest that we may be better placed to consider AI a provocation rather than a solution. It concludes by noting that hybrid human-AI approaches to documentation could be used to offset some of the cognitive biases bring to museum cataloguing, but only if we move beyond the solutionist rhetoric through which AI is frequently promoted.

1 of 18
Download to read offline
Error or Serendipity? Kevin Bacon, Digital Manager A small experiment with AI at Royal Pavilion & Museums Digital Collecting Practices 2019
Royal Pavilion & Museums is Royal Pavilion Preston Manor Brighton Museum Hove Museum Booth Museum of Natural History
Error or Serendipity: a small experiment with AI at Royal Pavilion & Museums
Royal Pavilion & Museums is Collections include World Art, Decorative Art, Natural Sciences, Fine Art, Local History, Archaeology, Film & Media, Costume, Toys, Musical Instruments, Royal Pavilion Archive
We are a Victorian accident. Wild eclecticism of our collections is not commensurate with our resources. How can we sustainably make our collections relevant to the digital age? The problem?
Collection Management Pressures 1. There will never be enough staff 2. Accreditation requirements prioritise basic cataloguing over digitisation 3. Intellectual accessibility becomes marginalised
Is AI (computer vision) the solution? 1. Boost documentation output without threatening staff posts 2. Supports value of digitisation as an activity 3. Enables knowledge mining and improved intellectual accessibility
AI & RPM: DAMS RPM uses Asset Bank both internal access and online collections Sometimes used to harvest data about our collections that cant easily be accommodated in our Collection Management System Auto-tagging tool introduced in 2018 using Amazon Rekognition
Brighton Town Hall as Parthenon in Ancient Egypt Its only 50% accurate but is there serendipity in the error? Tags: Art, Building, Architecture, Temple, Worship, Column, Pillar, Parthenon, Shrine, Ancient Egypt, Ruins
RPM & AI: Gift EU funded project led by IT University of Copenhagen RPM one of 10 partner museums in action research module Projects have focused on playful and personalised hybrid experiences gifting.digital Gift Experience by Blast Theory
The Experiment: Question If AI provides an unreliable solution, might it provide better use as a provocation?
Test whether AI driven keyword tagging can inspire better documentation. Can an AIs mistakes prove helpful to humans who can respond to the errors? Instructions to participants: 1.Write description 2.Check and correct AI tags 3.Revisit and improve the description after correcting the tag The Experiment: What we did
Discovery 1: Errors can lead to Enrichment Tags: Animal, Mammal, Rodent, Rat Mistakes can prompt corrections: rodent not removed, but replaced by mustelid Error begets new data that may not have been present had the error not existed
Discovery 2: Refocus Attention Tags: Animal, Mammal, Rodent, Rat Town Hall is not just a Brighton landmark, but an example of an architectural style Identifies the small detail: humans are an important element of this print Tags: Human, Person, Art, Building, Architecture, Drawing
Scientists ---------- Researchers, like data Artists ---------- Browsers, looking for inspiration Discovery 3: Inspires Conversation Experiment lead to discussion about multiplicity of audiences for different digital assets Possible tool for driving audience centred description
WYSIATI What You See Is All There Is: a focus illusion that blinds people to the surrounding context of a problem Cognitive bias that means humans have a tendency to draw on very limited information when solving a problem More likely to occur when conducting repetitive tasks like documentation Thinking systems: System 1: quick, without conscious control, often relying on heuristics System 2: slow, conscious, using deliberative reasoning
Conclusions / Provocations Unwise to conclude much from a small and wilfully unacademic experiment. But: 1. The rhetoric of AI in cultural heritage is overwhelmingly solutionist, and this may be obscuring its short-term value as a prompt or correction to human practice. 2. If the limitations of AI and human cognition can be checked against each other, does this mean that museum documentation can become a meaningful act of playful dialogue? 3. How can we create workflows to encourage this hybrid approach?
kevin.bacon@brighton- hove.gov.uk @fauxtoegrafik brightonmuseums.org.uk Thank you

More Related Content

Error or Serendipity: a small experiment with AI at Royal Pavilion & Museums

  • 1. Error or Serendipity? Kevin Bacon, Digital Manager A small experiment with AI at Royal Pavilion & Museums Digital Collecting Practices 2019
  • 2. Royal Pavilion & Museums is Royal Pavilion Preston Manor Brighton Museum Hove Museum Booth Museum of Natural History
  • 4. Royal Pavilion & Museums is Collections include World Art, Decorative Art, Natural Sciences, Fine Art, Local History, Archaeology, Film & Media, Costume, Toys, Musical Instruments, Royal Pavilion Archive
  • 5. We are a Victorian accident. Wild eclecticism of our collections is not commensurate with our resources. How can we sustainably make our collections relevant to the digital age? The problem?
  • 6. Collection Management Pressures 1. There will never be enough staff 2. Accreditation requirements prioritise basic cataloguing over digitisation 3. Intellectual accessibility becomes marginalised
  • 7. Is AI (computer vision) the solution? 1. Boost documentation output without threatening staff posts 2. Supports value of digitisation as an activity 3. Enables knowledge mining and improved intellectual accessibility
  • 8. AI & RPM: DAMS RPM uses Asset Bank both internal access and online collections Sometimes used to harvest data about our collections that cant easily be accommodated in our Collection Management System Auto-tagging tool introduced in 2018 using Amazon Rekognition
  • 9. Brighton Town Hall as Parthenon in Ancient Egypt Its only 50% accurate but is there serendipity in the error? Tags: Art, Building, Architecture, Temple, Worship, Column, Pillar, Parthenon, Shrine, Ancient Egypt, Ruins
  • 10. RPM & AI: Gift EU funded project led by IT University of Copenhagen RPM one of 10 partner museums in action research module Projects have focused on playful and personalised hybrid experiences gifting.digital Gift Experience by Blast Theory
  • 11. The Experiment: Question If AI provides an unreliable solution, might it provide better use as a provocation?
  • 12. Test whether AI driven keyword tagging can inspire better documentation. Can an AIs mistakes prove helpful to humans who can respond to the errors? Instructions to participants: 1.Write description 2.Check and correct AI tags 3.Revisit and improve the description after correcting the tag The Experiment: What we did
  • 13. Discovery 1: Errors can lead to Enrichment Tags: Animal, Mammal, Rodent, Rat Mistakes can prompt corrections: rodent not removed, but replaced by mustelid Error begets new data that may not have been present had the error not existed
  • 14. Discovery 2: Refocus Attention Tags: Animal, Mammal, Rodent, Rat Town Hall is not just a Brighton landmark, but an example of an architectural style Identifies the small detail: humans are an important element of this print Tags: Human, Person, Art, Building, Architecture, Drawing
  • 15. Scientists ---------- Researchers, like data Artists ---------- Browsers, looking for inspiration Discovery 3: Inspires Conversation Experiment lead to discussion about multiplicity of audiences for different digital assets Possible tool for driving audience centred description
  • 16. WYSIATI What You See Is All There Is: a focus illusion that blinds people to the surrounding context of a problem Cognitive bias that means humans have a tendency to draw on very limited information when solving a problem More likely to occur when conducting repetitive tasks like documentation Thinking systems: System 1: quick, without conscious control, often relying on heuristics System 2: slow, conscious, using deliberative reasoning
  • 17. Conclusions / Provocations Unwise to conclude much from a small and wilfully unacademic experiment. But: 1. The rhetoric of AI in cultural heritage is overwhelmingly solutionist, and this may be obscuring its short-term value as a prompt or correction to human practice. 2. If the limitations of AI and human cognition can be checked against each other, does this mean that museum documentation can become a meaningful act of playful dialogue? 3. How can we create workflows to encourage this hybrid approach?

Editor's Notes

  • #4: Wild eclecticism