Scale AI, a San Francisco-based company that provides data labeling services for artificial intelligence (AI) models, is facing criticism from its Filipino contractors who work on its Remotasks platform. Remotasks is an online platform that allows workers from developing countries to perform tasks such as annotating images, transcribing audio, and moderating content for AI companies. Scale AI claims that Remotasks offers workers a flexible and rewarding way to earn money online, but many workers say that they are paid unfairly and treated poorly by the company.
According to a report by The Washington Post, some Remotasks workers in the Philippines earn as little as $2 a day for working up to 10 hours on tedious and repetitive tasks. They also complain about frequent technical glitches, arbitrary rejections, and delayed payments. Some workers say that they have been banned from the platform without any explanation or recourse. The report also reveals that Scale AI has been using Remotasks to label data for controversial projects such as Clearview AI, a facial recognition company that has been sued for violating privacy laws.
The dark side of the AI boom
The report by The Washington Post exposes the dark side of the AI boom, which relies on a large and cheap workforce of data labelers in developing countries. Data labeling is a crucial step in building and training AI models, as it involves adding tags or labels to raw data such as images, videos, text, or audio. These labels help the AI models learn how to recognize patterns, make predictions, and perform tasks. However, data labeling is also a labor-intensive and low-skill process that can be outsourced to online platforms such as Remotasks.
According to a study by MIT Technology Review, the global market for data labeling was worth $1.7 billion in 2020 and is expected to grow to $4.6 billion by 2027. The study also estimates that there are more than 2 million data labelers worldwide, with the majority of them located in countries such as India, China, Kenya, and the Philippines. These workers are often paid below minimum wage and face poor working conditions, such as long hours, high pressure, low quality control, and lack of benefits. The study also warns that data labeling platforms may exploit the economic and political vulnerabilities of their workers, especially in countries with weak labor laws and human rights protections.
The need for ethical and sustainable data labeling practices
The report by The Washington Post and the study by MIT Technology Review highlight the need for ethical and sustainable data labeling practices in the AI industry. They call for more transparency, accountability, and regulation from both the data labeling platforms and their clients. They also urge the AI companies to respect the dignity, rights, and welfare of their data labelers, and to ensure that they are paid fairly and treated humanely. They also suggest that data labelers should be given more opportunities for education, training, and career development, as well as more involvement and feedback in the data labeling process.
Some initiatives have already been launched to address these issues, such as the Data Nutrition Project, which aims to create standards and tools for assessing the quality and ethics of data sets. Another example is Weights & Biases, which offers tools for tracking and managing data labeling workflows. However, more efforts are needed from all stakeholders in the AI ecosystem to ensure that data labeling is not only a profitable but also a responsible and respectful activity.