1 of 28

Machine Media

Week 3

2 of 28

Week 3 - Class Overview

Themes & Timeline:

Week 1: Introductions, Chance & Protocol

Week 2: Chatbots and Generative text

Week 3: �Data Labor

Week 4:

Classification, Taxonomies, Computer Vision

Week 5:

Generative Adversarial Networks, Handmade Datasets

Week 6:

GAN review,�Photo tutorial

Week 7:

Facial Recognition,

Identity, Surveillance

Week 8:

Deepfakes

Week 9: The Digital is Physical: Environmental impact

Week 10:

Handmade Dataset mid-way presentations

Week 11:

Data augmentation�Workshop (python)

Week 12:

Writing Images: Text-to

-image

models

Week 13:

Data Augmentation workshop part 2

Week 14:

Training Demo�In-class work day

Week 15: Final Presentations

Thanksgiving Break

Handmade Dataset Project

3 of 28

Week 3 - Agenda

  • 8:45 - 9:30 Review homework
  • 9:30 - 9:50 Sorting Activity
  • 9:50 - 10:00 Break
  • 10:00 - 10:20 Lecture: Data Labor
  • 10:20 - 10:30 Break
  • 10:30 - 10:55 Technical Demo - Teachable Machine
  • 10:55 - 11:00 Homework

4 of 28

Homework review: �

Questions:

What chatbots did you talk to? Can you tell us about the experience you had chatting with them? What did you expect before chatting and how did or didn’t your experience match your expectations?

Tell us more about the text you chose to work with for your generator. Why did you choose to work with this text? Did the results surprise you?

5 of 28

Data Labor

6 of 28

Week 5 - Excavating.ai

“ To build a computer vision system that can, for example, recognize the difference between pictures of apples and oranges, a developer has to collect, label, and train a neural network on thousands of labeled images of apples and oranges. On the software side, the algorithms conduct a statistical survey of the images, and develop a model to recognize the difference between the two “classes.” If all goes according to plan, the trained model will be able to distinguish the difference between images of apples and oranges that it has never encountered before.”

7 of 28

Week 5 - Excavating.ai

IMAGENET:�“briefly became the world's largest academic user of Amazon’s Mechanical Turk, using an army of piecemeal workers to sort an average of 50 images per minute into thousands of categories.[11] When it was finished, ImageNet consisted of over 14 million labeled images organized into more than 20 thousand categories.”

8 of 28

LAION 5B:��Over 5 billion image-text pairs

Week 5 - GAN

9 of 28

The Mechanical Turk was a fraudulent chess-playing machine constructed in 1770, which appeared to be able to play a strong game of chess against a human opponent. For 84 years, it was exhibited on tours by various owners as an automaton.

Amazon Mechanical Turk (MTurk) is a crowdsourcing website with which businesses can hire remotely located "crowdworkers" to perform discrete on-demand tasks that computers are currently unable to do as economically.

10 of 28

Week 6 - Data Labor

“Unlike in 2009, when the main crowdworking platform was Amazon’s Mechanical Turk, there is currently an explosion of data labeling companies. These companies are raising tens to hundreds of millions in venture capital funding while the data labelers have been estimated to make an average of $1.77 per task. Data labeling interfaces have evolved to treat crowdworkers like machines, often prescribing them highly repetitive tasks, surveilling their movements and punishing deviation through automated tools. Today, far from an academic challenge, large corporations claiming to be “AI first” are fueled by this army of underpaid gig workers, such as data laborers, content moderators, warehouse workers and delivery drivers.”��- The Exploited Labor Behind Artificial Intelligence

11 of 28

Week 5

Icebreaker: Please sort the images on the table

12 of 28

Week 6 - Data Labor

13 of 28

The Cleaners

A documentary from 5 years ago about

Content Moderators working in the Philippines

Week 6 - Data Labor

14 of 28

Mimi Onuoha - The Future Is Here

Week 6 - Data Labor

15 of 28

Week 6 - Data Labor

Vainu

Contracted with two prisons in Finland to have prisoners work as annotators

16 of 28

Week 6 - Data Labor

M2Work (Nokia and the World Bank) Company that employs Palestinian refugees.

Frames it as providing valuable job-training

Microwork comes with no rights, security, or routine and pays a pittance — just enough to keep a person alive yet socially paralyzed. Stuck in camps, slums, or under colonial occupation, workers are compelled to work simply to subsist under conditions of bare life. This unequivocally racialized aspect to the programs follows the logic of the prison-industrial complex, whereby surplus — primarily black — populations [in the United States] are incarcerated and legally compelled as part of their sentence to labor for little to no payment. Similarly exploiting those confined to the economic shadows, microwork programs represent the creep of something like a refugee-industrial complex.

17 of 28

“Artificial intelligence may be making some jobs obsolete but it has given a new lease of life to one group of people who play an unglamorous but critical role in the machine learning pipeline: first generation women workers in Indian towns and villages.”

“One day,” Sruthi told me, “I realised this is not for gaming. We are teaching machines to see like a human. We teach a robot how to understand things on their own.”

Week 6 - Data Labor

18 of 28

1. What issues may arise when we have workers annotating data without the understanding of how it is being used?

2. What issues may arise when we have companies in the Global North outsourcing low-wage data work to the Global Majority? Discuss the potential benefits and drawbacks of “click-work” for workers.

3. How does the knowledge of the labor behind Machine Learning change the way you relate to the technology as a user?

Week 3 - Data Labor

💡

19 of 28

Teachable Machine Demo!

A web-based tool (by Google) for training a machine learning model. It’s important to note that you are not training a model from scratch with teachable machine. You are taking a model that has already been trained and training it more.

New terms* 💡 - Pre-trained, transfer-learning

Week 3 - Data Labor

💡

20 of 28

Week 1 - Class Overview

Model:

  • A machine learning model is an instance of an algorithm that has or will be 'trained' on data.

Training

  • The "teaching"
  • The machine learning model "looks" at existing data. This data influences its predictions or productions.

Training Data

  • The data that the model "looks" at.

21 of 28

Week 1 - Class Overview

Pre-trained Model:

  • A machine learning model that has already been trained on something

Transfer-learning

  • Training a model that has already been trained on something else first.
    • Speed up training time
    • Improves performance
    • Might effect bias*

22 of 28

Week 3 - Teachable machine

23 of 28

Week 3 - Teachable machine

24 of 28

Week 3 - Teachable machine - Connecting to p5

25 of 28

Week 3 - Teachable machine - Connecting to p5

Use this p5 sketch as a template instead of this one!

26 of 28

Week 3 - Teachable machine - Connecting to p5

27 of 28

Week 3 - Teachable machine - Connecting to p5

28 of 28

Week 3 - Homework

Teachable Machine:

  • Choose a “reading”: Excavating AI or What does a dataset want?
  • Continue training your teachable machine and be able to present your progress next week.