Mike Ridley, Visiting Scholar gave a talk on AI at Ryerson Library, and I was lucky enough to be invited to attend.
Visiting Scholar
Help advance the things interested in in a collaborative way and exploring.
Interested in areas of AI. Investigation and opportunities on collaboration. Look at what these technologies can do.
AI
It’s a horrible term because it’s overused and used to mean different things:
- machine learning (most common)
- deep learning
- neural networks
- autonomous vehicles
- expert systems
When we talk about AI, we often are talking about machine learning and algorithms.
- ubiquitous (using a lot of the resources we already have)
- powerful (having a deep impact)
- opaque (less likely to trust it, because not transparent)
- invisible (a lot of uses, we don’t recognize that it’s happening)
- consequential (critical e.g. screening for loans, jobs, etc.)
Exploring what we (the audience) think
Terms we think of when we hear AI: robots, algorithms, machine (learning).
MIT doing research on how “smart” the smart devices are. Children think of these things as individuals. If one Alexa can’t answer, they’ll ask the other one.
Most of us said that AI is important to libraries.
Importance to libraries
The danger is not so much in delegating cognitive tasks, but in distancing ourselves from – or in not knowing about – the nature and precise mechanisms of that delegation.
Important to libraries because it focuses on two of the most important part of libraries: knowledge organization and discovery.
And yet, very little work going on in libraries and information science. Very few are exploring the possibilities, especially in a deeper way.
I believe that artificial intelligence will become a major human rights issue in the twenty-first century. Safiya Noble, Algorithms of Oppression (2018)
What AI is doing
AI is meant to predict (not correlate or recommend). Humans often mix causation and correlation.
Prediction machines: if we give a machine of a lot of dog and cat pictures, it can predict whether we classify a picture as cat or dog.
Doesn’t have to be images, it can be:
- text
- audio
- data
- video
- tactile
- embodiment
Cost of prediction is going down as they are developed, while the value of judgment has gone up.
Components and our role
Core:
- computation
- data
- algorithms
Phones have chip specific to doing AI vs. deep learning with big data. Spanning the range, so tools are available to anyone.
Spend 80% of the time is spent. Have to make it available. Data preparation is that most time consuming. Need structured data. What libraries work with: metadata, structured data. Have to think about the validity of data. Typically, when say AI is unfair, it’s the data because of where or how the data is collected. Need to at least document if we cannot eliminate the bias.
Private companies are taking open data and making it private. So need to think about the relationship and how it’s changing. Unless we get the algorithms.
Hundreds of algorithms available.
Social, political, and economic context. They are embedded in these contexts. Need to understand the impacts and interactions. This is especially important in light of misinformation, which AI is very good at doing.
AI in libraries
Where will AI have the most impact in the Library:
- Search & Discovery in particular but can have impact in all areas including
- cataloguing/metadata,
- reference,
- acquisitions.
What’s the library’s role?
- Answers were pretty close on user awareness, critique & assessing,
- with slightly lower user skills, and design & create AI,
- low on serve AI
Unfortunately, doing a poor job of critiquing and assessing. We have skills and perspectives that are unique in many ways. Often see things in a non-competitive way, and see it as part of our responsibility.
Opportunity for us to play around with AI, and move towards design and create AI especially for our own purposes.
I think we would be wise to start thinking now about machines and algorithms as a new kind of patron. Chris Bourg
Want AI to work as effectively as possible. Sharing our data for them to get what they want. Having the data in a way that’s easy for it to ingest.
Examples
Hamlet
- out of MIT, by Andromeda Yelton
- operates on open access database of theses at MIT
- simple recommendation algorithm
- can also upload file of text to give back theses that are conceptionally similar
- can upload file and create a bibliography on conceptually similar theses
Gone from search to find. Some of it is relationship searching, some semantic searching. On GitHub if you want to replicate it.
Project Aiur
- hypothesis extraction,
- validation: will test hypothesis
- generation: maybe something that we can extract and create a new hypothesis
- working on all sorts of articles
ChatBots
- early examples were very simplistic and shallow
- but powerful because can marry other technologies:
- natural language processing,
- NL generation,
- knowledge base,
- data preparation,
- ontologies,
- training,
- learning
- elementary way of providing information, alternate ways
- commonly text driven, but audio driven is increasing
- what we can do is spend time in the data preparation and ontologies that many others skim on
Observations
New digital divide: people who can use algorithms vs. used by algorithms
The Real World of Technology (1999) by Ursula Franklin
- technology should be an interaction: reciprocity
- redemptive technologies: should design for a particular, positive purpose. Should otherwise get rid of it. Those affected should be part of design, which has not happened in AI.