Gray and Suri’s 2019 book Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass draws on a five-year study to investigate the booming yet still largely hidden workforce, ghost work, referring to the broad groups who are captioning photos, designing/coding a project, flagging/removing inappropriate content, and so forth. This new working style will continue to displace the traditional 9-to-5 schedule with on-demand tasks. By examining four different ghost work platforms, including MTurk, UHRS, LeadGenius, and Amara.org, it shows that ghost work is in high demand at the current stage of AI development and the data from the crowdsourcing is important to achieve a higher level of automation.
To be honest, I have heard about the ghost work platforms before but never have any experience with them. Through the real-life stories shared by a bunch of ghost workers in the book, I see both sides of the job shifting. From the worker perspective, the advantages of the new work style provide them with the flexibility of the working schedule. One of the ghost workers mentioned that she needed to take care of her mom so she quit her job and chose to remotely work in flexible hours for MTurk. The downsides are the workers are not compensated well nor have any insurance or other benefits. As more and more people shift to the work style, competition for tasks cannot be avoided. I wonder if the competition would intensity the issue of unfair compensation because those who most need the work may be willing to lower the price they ask to grasp the opportunity. As the ghost work industry growing to a certain level, rules and regulations are indeed necessary to ensure the healthy operation of different platforms.
From the ghost work itself perspective, I feel the majority of the current tasks on the platforms such as manually labeling are due to the fact that we are still at the baby stage of AI. Sometimes we are over-optimistic about the current AI achievement and expect AI can solve any problem in our daily life. For one thing, many of the tasks tackled by AI are still naive. For example, on image classification, it takes an individual less than 1 second to get the correct answer, however, machine learning models may suffer from insufficient data or unbalanced distribution between training and testing data and perform poorly in the generalization between different data sets. For another thing, merely relying on the results generated by AI, especially black-box models, to make crucial decisions may put our society at great risk. The trust and reliable AI can hardly be achieved without an understanding of the features learned in a model or the rationale beyond merely producing results. As AI moved to a more advanced stage, the major tasks of crowdsourcing may witness a big change as well. As the tasks completed through human-computer interaction are utilized to push the limit of AI, here are some questions that may be worth further discussion.
- What is the long term goal of the “AI revolution”, is leveraging human source to make AI more AI or utilizing AI to free humans to make humans more human?
- As AI replaces more and more work of comparative low-requirements, would the future increasing unbalance need for micro-tasks and macro-tasks cause societal issues?
- Would any bias be introduced in the tasks of the ghost work because of workers’ demographics, like age, gender, educational level, have an impact on the final model/system performance? If so, is there any way to mitigate the bias?
- As more and more jobs and people shift to ghost work, it raises certain issues: How do ghost workers find quality jobs? How do the platforms pay ghost workers fairly?
