Tracking (known) delivery of the material

[davanstrien]

It would be good to try (as far as practical) to track delivery of the materials. This will be useful to:

• help validate people are using the material
• know who has delivered material and can give feedback
• learn our audience better

Suggest posting replies to this thread as a low-key way to track this. cc @MikeTrizna @mark-bell-tna @noramcgregor

[davanstrien]

#80

As part of The Congruence Engine training program, we are planning to host a course on Machine Learning for GLAM using this course - the session is now planned for the 3rd May, at the MakerSpace, SAS, London. We are excited to have @mark-bell-tna with us co-delivering the lesson. Hopefully we will be able to provide feedback both from participants and instructors.

[jt14den]

@leighphan, @kjallen, @jmjamison, and I are teaching it on May 8-9 at the UCLA Library https://ucla-data-science-center.github.io/2023-05-08-UCLA/. We are really looking forward to it and hope to gather some good feedback.

[davanstrien]

@leighphan, @kjallen, @jmjamison, and I are teaching it on May 8-9 at the UCLA Library ucla-data-science-center.github.io/2023-05-08-UCLA. We are really looking forward to it and hope to gather some good feedback.

Awesome! thanks for letting us know. Very happy to get feedback after this :)

[kjallen]

Suggestions for Chapter 01 – Welcome – Intro to AI for GLAM

A visual demo of AI would be helpful early as a way to introduce what the curriculum will cover.

Recommendation:

• Incorporate 2-3 examples of how AI is used in day-to-day life (e.g. recommendation systems, facial recognition, autonomous cars, smart phones).

• Include one interactive demo using an online tool to enforce learning topics. A demonstration of a tool that does not require registration is ideal (for example, displaCy).

[kjallen]

Suggestions for Chapter 02 – Artificial Intelligence (AI) and Machine Learning (ML) in a nutshell

• Provide clear, concise definition for Artificial Intelligence and Machine

• Split lesson into 2 sections:

  • AI/ML and What Is Machine Learning
  • Supervised vs Unsupervised Learning

• Reduce jargon use in first paragraph of ‘What is Machine Learning’

• Reduce the knowledge required from the jump from Chapter 1 to Chapter 2, the jump from basic ML definition to the types of ML (supervised, unsupervised, etc...) could require some background knowledge

[leighphan]

Suggestions for Chapter 03 – Machine Learning Modelling Concepts

Overall this lesson seemed very advanced for novice learners, the concept of a model/features/labels was difficult to convey in a short time in a clear, concise manner and would benefit from more visual graphics to help explain some of the core concepts.

Recommendations:

• Rearrange (switch) the order of Lesson 3 and Lesson 4. An audience with an introductory level of experience may benefit from first learning what AI is good at before learning how it works.
• The most valuable section of the lesson is the introduction to modeling & algorithms, the content on feature engineering and embedding seemed like it is not an essential concept for beginners to AI and could be de-emphasized or added as an addendum.

[leighphan]

Suggestions for Chapter 04 – What is Machine Learning good at? – Intro to AI for GLAM

This portion was of great interest to learners, specifically how this can be applied and where these concepts are being used in the learner's workplace.

Recommendations:

• Incorporate working example projects in a live environment; this would provide demonstrated examples for lesson’s concepts.
• It would be ideal to show several different demos of actual applications being used in GLAM space that learners can try on their own. For example, A working live demo using tools learners can play with such as Voyant, or some other graphical tool would be valuable.
• On the topic of NLP, please add the term ‘Text Analysis’ in addition to (or taking precedence over) NLP as it may be a more familiar term for beginners.

[kjallen]

Suggestions for Chapter 05 - Understanding and managing bias – Intro to AI for GLAM

Overall this is a very dense lesson and could benefit from breaking up and adding more interactive exercises or demos to reduce cognitive overload and increase understanding.

• Break up lesson, current iteration is very text dense - two sections

  • Bias in Machine Learning
  • When might bias enter a machine learning pipeline?

• The Common Bias types is a lot of information to absorb, maybe have a brief example for a few types or consolidate this into activities demonstrating the bias type

• Move Common Bias definitions to the References section Intro to AI for GLAM: Glossary

• Simplify initial example to show bias in a basic case before moving to more advanced ‘real world’ data set. The numerous text dense examples can be difficult to parse.

• Add real-world examples to start of each section to illustrate how bias occurs in ML/AI applications

[kjallen]

Suggestion to add a jargon-busting or jam session type discussion to kick off the workshop as part of the lesson rather than informal add on.

Our target audience for this workshop was library staff and IS graduate students, with the assumption of no background knowledge of Artificial Intelligence and no technical background requirements.

Specific areas we addressed in our jargon busting session and to consider for more structured discussion addition to the lessons:

• Self selected groups identify terms they have heard adjacent to the topic
• Each group shares list and definitions with the class
• Instructors add and roughly cluster terms in the Etherpad
• Instructors review terms and cover how/if they will be covered or relate to lesson

There was a lot of interest around topics not explicitly covered by the workshop including:

• ChatGPT / LLMs
• Image generative models / GANs
• Acronym demystification (AI, ML, LLM, NLP, GAN, NN, DL, etc...)
• Relationship between algorithms, artificial intelligence, deep learning, and machine learning -- how do these terms relate to each other and what makes each distinct and when is it appropriate to use one term over another

[leighphan]

Lesson Structure Observations

The lessons contain a lot of information for someone new to AI/ML to take in. This includes a significant number of acronyms, basic vocabulary and some statistical concepts to absorb.

Some suggestions to flatten the learning curve:

• Subdivide the episodes into smaller bite-size portions.

• Account time for audience conversation and extraneous asides (for example: AI in the news).

• Include group exercises, interactive if possible, with relevant data sets. Example: As part of their Leading Equitable Data Practices training, LA Tech4Good includes several interactive exercises including on in which learners create a datasheet for a dataset that is part of their work. (Reference: Datasheets for Datasets).

• Reduce prerequisite knowledge of existing AI/ML terms.

• Standardize use of vocabulary throughout the lesson, (i.e. use either Artificial Intelligence, Machine Learning, or AI/ML (as single term)) or explain the subtle differences early in the lesson so the learners are not confused. This can be thought of as differentiating between terms in other lessons (Arrays vs Lists vs Vectors).

[MikeTrizna]

I gave a 2-hour version of this workshop to an internal group of almost 50 Smithsonian employees last Tuesday (Jan 23). Slides are available via FigShare here: https://doi.org/10.25573/data.25105826.v1.

New sections I tested out are the addition of LLMs, added slides about Risk to the episode on Ethics.

Ran very low on time -- which totally makes sense since I added material to a supposedly 3-hour workshop -- but survey respondents especially enjoyed the activities.

[kjallen]

We are creating specific issues to address the comments on this issue.

Please reference this issue when creating an issue for a specific comment so we can track unfinished change requests.