Libraries of machine learning models
If you wanted to explore one of these more specialist models, how might you find one? One of the significant advan- tages of systems like ChatGPT is that they provide a single interface for a broad range of tasks. If we want to use a more specific type of machine learning system for our task, we need to find one that might fit our needs. This is where platforms like the Hugging Face Hub can help. The Hugging Face Hub is a platform which hosts hundreds of thousands of models, datasets and machine-learning applications that the community can utilise. These platforms often help you find a model that will fit your use case, for example, a model that performs a task similar to the one you want to do or trained on data similar to yours. Librar- ians understand their collections better than anyone, so they are ideally placed to determine which models might be most appropriate for their organisation, collection, and specific use case.
Information literacy: librarians can evaluate models
One of the main challenges of using machine learning is ensuring that you have a machine learning system that works for your particular use case. A
machine learning system that works well in one context might not be helpful in another. For example, one could train a machine learning model to assign incom- ing chat messages from an enquiry system into one of three categories: positive, negative or neutral. However, a system trained on the chat messages from a uni- versity library system might not perform well when used in a law library where the types of messages and the language used may differ quite drastically. We should be wary, then, of any promises of ‘off the shelf’ solutions that will work ‘out of the box’ for all use cases.
There are various ways in which it is possible to evaluate how well a machine learning system is performing. Librarians already have skills in critically assessing information; these can be expanded to learn how to evaluate whether a machine learning system performs properly. Whilst the power to make decisions about what technology to adopt might not always be fully in the hands of librarians, they should feel more empowered to ask point- ed questions to vendors who seem like they are promising too much.
DIY: librarians can create their own models
Librarians already have many of the skills needed to train their own models for their
exact use case. Whilst there is a percep- tion that AI can only be done by huge tech companies, this is not the case for many applications of AI. There is a major benefit in libraries and librarians becom- ing more involved in machine learning. Libraries can benefit from AI, but they will be best able to do this if they are deeply involved.
Librarians can impact AI
Beyond this, there is a broader potential role for librarians to play in contributing to and shaping the machine learning ecosystem as society more broadly adopts AI. Returning to my starting question: how will libraries impact AI? Libraries are ideally placed to ensure that AI hype is accompanied by scrutiny; they can ensure their patrons know what an AI system can and can’t do. The library sector can also play a key role in ensuring that the artefacts of machine learning (datasets, models, documentation) are carefully managed using the lessons learned by librarians over many years of collecting, curating and preserving information. IP
l Daniel will discuss these issues in more detail at the forthcoming Libraries Rewired conference on the 10 of November, 2023 –
https://librariesrewired.org.uk/.
October-November 2023
INFORMATION PROFESSIONAL 19
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
