Hidden Workers powering AI
Many people are aware of various AI tools and technologies but most of us aren’t aware of invisible workers involved in the production of AI. This blog post focuses on hidden labour involved in AI production. We aim to raise awareness and educate educational institutions about this important issue. Although research in this area is limited, the post highlights the role of hidden workers, who they are, what can be done to improve the situation and make informed decisions about using AI products and services.
However, any AI system requires a significant amount of human effort that is sometimes disregarded. This “hidden work” is crucial to the development and functioning of AI systems but is often unnoticed. The production of AI involves a significant amount of human labour, often called as “Ghost workers “who empower AI, behind-the-scenes. This hidden labour is often overlooked in discussions of AI, but it is essential to understanding the implications of AI on society.
Kate Crawford, in her book “Atlas of AI,” highlights dark side of AI production and suggests AI is neither “artificial” nor “intelligent.” The creation of AI involves a large consumption of resources, such as energy and minerals, and the workers who contribute to its production often have their rights disregarded. Crawford’s book sheds light on the exploitation of workers in the background of AI’s “automation.”
Mary Gray’s book “Ghost Work” focuses on invisible workers behind AI and challenges faced i.e., lack of job security, benefits, fair pay, and recognition. The book highlights need for better treatment, fair work conditions and raises important questions about the ethics of AI and its impact on hidden workers.
How Hidden workers are empowering AI:
To shape and train these applications, digital giants like Amazon, Google, and Facebook, employ an army of invisible labour. But there are others such as Appen, Upwork, CrowdFlower and Microwork. They hire workers for remote work from anywhere in the world to train machine learning models by perform tasks like data labelling, data annotations, transcription etc.
It is the process of marking or classifying data used to train AI systems. This includes classifying text, transcribing audio, or recognizing objects in photos. For example, in image recognition, data labelling could involve manually labelling each image with information about what’s in it (e.g., “dog”, “car”, “person”). This labelled data is then used to train artificial intelligence algorithms to accurately recognize and categorize images on their own. In essence, data labelling is a crucial step in the development of AI systems as it provides the foundation for the algorithms to learn from.
Autonomous cars , a rapidly growing sector that is expected to be worth $556 billion by 2026. To navigate its driverless vehicles, companies like Tesla require clean and tagged data. This information is obtained via onboard cameras and must be classified and labelled for the automobile to detect items such as people, traffic signs, and other cars.
Data labelling is labour-intensive, time-consuming, and repetitive procedure that needs to be done with great precision. People in low-wage countries who work long hours for low and non-negotiable wages are frequently hired to perform tasks like data labelling. They neither receive contracts nor incentives and are unaware of the usage of their work.
Somehow, we all have likely been participating in data labelling by completing captcha codes, as the responses we provide are used to train machine learning models. It’s possible that you have encountered instances where a website prompts you to enter a series of characters displayed in a distorted image. These are known as captcha codes, which are designed to differentiate human users from automated bots. Your responses, along with others, are utilized as a form of data labelling for machine learning models. This data labelling is important in enhancing the precision of these models across a range of applications, such as image classification and speech recognition.
Who are Hidden Workers?
According to the study these are the individuals who desire employment but are unable to secure suitable job opportunities due to multiple factors such as discrimination, lack of education/ skills, discouragement due to their repeated unsuccessful job search. This includes refugees, people with disabilities, veterans, prisoners, care givers and relocated partners and spouses. The study estimated 27 million hidden workers in US alone by interviewing 8000 hidden workers and over 2250 executives in US, UK, and Germany. These are invisible workers, paid as little as 10 cents- $2 an hour to feed information into computer systems.
Refugees empowering machine learning advances by working for Silicon Valley corporations like Google, Amazon, Facebook from a tent with computers in Dadaab, Kenya, one of the world’s largest refugee camps. They conduct click work, such as video tagging and audio transcription, as one of their limited legitimate job possibilities. However, the task is difficult, paid per piece, and performed in confined, airless environments. The refugees in Lebanon’s Shatila camp are compelled to labour at night labelling footage of large cities for the benefit of foreign capitalists, although the specific goal or beneficiaries of their work are unknown.
M2Work, coordinated by infoDev, is an online program aimed at encouraging poor countries to participate in the production side of digital economy and create jobs. Similarly, Sama, data annotation and labelling platform, teach and train refugees in Kenya, India, and Uganda to accomplish brief data tasks. Sama works with clients such as Walmart, Google, Microsoft and General Motors. Some of its employees in Kenya contributed to filtering toxic content for ChatGPT, which led to wide spread criticism due to the work negatively impacting their mental health, poor working conditions and little to no pay.
Traditionally, prison labour involved physical work but Prisoners in Finland are recruited by Vainu to conduct data labour for struggling startups, with the government getting paid for each assignment done. However, no information is available on what percentage of the cash goes to the convicts. The company claims that the collaboration is a sort of jail rehabilitation that teaches essential skills, but other experts feel that the job training claims are simply hype around AI’s promises.
The Online Labour Observatory platform serve as a centralized hub for researchers and decision-makers. It tracks the real-time number of tasks and projects completed by freelancers on different websites across various countries and job categories. The platform offers valuable information on the gig economy and assists those in decision-making positions to make informed decisions. The Online Labour Observatory provides a comprehensive view of the developments and patterns in online work, and its effect on both workers and the economy.
Despite these known issues this area remains under research. EU- Funded project conduct in-dept study about the working condition of hidden work force involved in the production of AI and its impact on their wellbeing.
How can we make it better?
- Offer fair pay, remove inequalities.
- Regulate and compensate hidden labour.
- Provide proper training and guidelines to labellers to avoid bias in the labelling work.
- Diversify labellers to include people from different cultures, backgrounds, and perspectives.
- Implement algorithmic fairness techniques, review and audit labelled data to identify and remove bias.
Universities and colleges must be aware of the labour behind AI and consider its impact when using AI services, including examining biases in data used to train AI algorithms and potential consequences in outcomes produced. By approaching AI responsibly, universities and colleges can ensure its benefits are shared and workers are treated fairly. Institutions should thoughtfully consider the ethical implications of AI and ask questions about the workers powering it before making decisions on its use.
- Who is developing and creating AI technology?
- How is data pre-processed and does it affect accuracy and fairness of AI systems?
- Are workers involved in AI production empowered and supported professionally?
- Are there concerns about exploitation or human rights violations in AI production process?
- How are AI algorithms tested for fairness and bias, and who determines test rigor?
- What steps are taken to ensure AI benefits are widely shared and technology is developed responsibly?
- Potential impact of AI systems on various demographic groups and how these impacts are monitored and addressed.
Understanding human involvement in AI systems enable individuals to make informed decisions about products and services they use and understand the consequences on society. By promoting critical thinking about the role of humans in AI, we can work towards creating a more equitable and ethical AI industry.
Find out more by visiting our National centre for AI page to view publications and resources, join us for events and discover what AI has to offer through our range of interactive online demos.
For regular updates from the NCAI sign up to our mailing list.
Get in touch with the team directly at NCAI@jisc.ac.uk