In dusty factories, cramped internet cafes and makeshift home offices around the world, millions of people sit at computers tediously labeling data.

These workers are the lifeblood of the burgeoning artificial intelligence (AI) industry. Without them, products such as ChatGPT simply would not exist. That’s because the data they label helps AI systems “learn.”

But despite the vital contribution this workforce makes to an industry which is expected to be worth US$407 billion by 2027, the people who comprise it are largely invisible and frequently exploited. Earlier this year, nearly 100 data labelers and AI workers from Kenya who do work for companies like Facebook, Scale AI and OpenAI published an open letter to United States President Joe Biden in which they said, “Our working conditions amount to modern day slavery.”

To ensure AI supply chains are ethical, industry and governments must urgently address this problem. But the key question is: how?

What is data labeling?

Data labeling is the process of annotating raw data—such as images, video or text—so that AI systems can recognize patterns and make predictions.

Self-driving cars, for example, rely on labeled video footage to distinguish pedestrians from road signs. Large language models such as ChatGPT rely on labeled text to understand human language.

These labeled datasets are the lifeblood of AI models. Without them, AI systems would be unable to function effectively.

Tech giants like Meta, Google, OpenAI and Microsoft outsource much of this work to data labeling factories in countries such as the Philippines, Kenya, India, Pakistan, Venezuela and Colombia.

China is also becoming another global hub for data labeling.

Outsourcing companies that facilitate this work include Scale AI, iMerit, and Samasource. These are very large companies in their own right. For example, Scale AI, which is headquartered in California, is now worth US$14 billion.

Cutting corners

Major tech firms like Alphabet (the parent company of Google), Amazon, Microsoft, Nvidia and Meta have poured billions into AI infrastructure, from computational power and data storage to emerging computational technologies.

Large-scale AI models can cost tens of millions of dollars to train. Once deployed, maintaining these models requires continuous investment in data labeling, refinement and real-world testing.

But while AI investment is significant, revenues have not always met expectations. Many industries continue to view AI projects as experimental with unclear profitability paths.

In response, many companies are cutting costs which affect those at the very bottom of the AI supply chain who are often highly vulnerable: data labellers.

Low wages, dangerous working conditions

One way companies involved in the AI supply chain try to reduce costs is by employing large numbers of data labellers in countries in the Global South such as the Philippines, Venezuela, Kenya and India. Workers in these countries face stagnating or shrinking wages.

For example, an hourly rate for AI data labellers in Venezuela ranges from between 90 cents and US$2. In comparison, in the United States, this rate is between US$10 to US$25 per hour.

In the Philippines, workers labeling data for multi-billion dollar companies such as Scale AI often earn far below the minimum wage.

Some labeling providers even resort to child labor for labeling purposes.

But there are many other labor issues within the AI supply chain.

Many data labelers work in overcrowded and dusty environments which pose a serious risk to their health. They also often work as independent contractors, lacking access to protections such as health care or compensation.

The mental toll of data labeling work is also significant, with repetitive tasks, strict deadlines and rigid quality controls. Data labelers are also sometimes asked to read and label hate speech or other abusive language or material, which has been proven to have negative psychological effects.

Errors can lead to pay cuts or job losses. But labelers often experience lack of transparency on how their work is evaluated. They are often denied access to performance data, hindering their ability to improve or contest decisions.

Making AI supply chains ethical

As AI development becomes more complex and companies strive to maximize profits, the need for ethical AI supply chains is urgent.

One way companies can help ensure this is by applying a human right-centered design, deliberation and oversight approach to the entire AI supply chain. They must adopt fair wage policies, ensuring data labelers receive living wages that reflect the value of their contributions.

By embedding human rights into the supply chain, AI companies can foster a more ethical, sustainable industry, ensuring that both workers’ rights and corporate responsibility align with long-term success.

Governments should also create new regulations that mandate these practices, encouraging fairness and transparency. This includes transparency in performance evaluation and personal data processing, allowing workers to understand how they are assessed and to contest any inaccuracies.

Clear payment systems and recourse mechanisms will ensure workers are treated fairly. Instead of busting unions, as Scale AI did in Kenya in 2024, companies should also support the formation of digital labor unions or cooperatives. This will give workers a voice to advocate for better working conditions.

As users of AI products, we can all advocate for ethical practices by supporting companies that are transparent about their AI supply chains and commit to fair treatment of workers. Just as we reward green and fair trade producers of physical goods, we can push for change by choosing digital services or apps on our smartphones that adhere to human rights standards, promoting ethical brands through social media, and voting with our dollars for accountability from tech giants on a daily basis.

By making informed choices, we can all contribute to more ethical practices across the AI industry.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Around two-thirds of consumers have used a food-ordering app at least once for takeout, delivery or both, according to the September 2024 Consumer Food Insights Report (CFI). Over half have used an app for a delivery order. Of those who say they have used an app to order food, nearly half report using one for either delivery or takeout at least once a week.

The survey-based report out of Purdue University’s Center for Food Demand Analysis and Sustainability (CFDAS) assesses food spending, consumer satisfaction and values, support of agricultural and food policies, and trust in information sources. Purdue experts conducted and evaluated the survey, which included 1,200 consumers across the U.S.

“The COVID-19 pandemic changed the economy in many ways, particularly in the service economy,” said the report’s lead author, Joseph Balagtas, professor of agricultural economics at Purdue and director of CFDAS.

Earlier this year, the U.S. Department of Agriculture (USDA) reported that spending on food-ordering apps for deliveries from full-service restaurants quadrupled between prepandemic months and 2022. The trend prompted the CFDAS team to partner with Valerie Kilders, assistant professor of agribusiness marketing at Purdue, to measure and evaluate consumer usage of the apps.

When ordering food online, 68% of consumers say they “sometimes,” “often” or “always” use discounts or promo codes.

Food purchased away from home is typically more costly than food prepared at home with groceries. Understandably, many consumers seek cost reductions when paying for the convenience of a prepared meal, Balagtas said. This is particularly true for consumers who spend the least on food. Half of them used discounts and promo codes “often” or “always” when ordering food online.

The report breaks down per-person weekly food expenditure into three groups: thrifty (less than $50 a week), moderate ($50 to $85 a week) and liberal (more than $85 a week) spenders. “Consumers who spend the most on food tend to seek out discounts less frequently,” Balagtas said.

The CFI survey also asked consumers about the additional fees associated with many food-ordering apps. Many attribute the fee to operating expenses of the service, whether it’s to cover fuel and time for delivery services or administration and maintenance of the app itself.

The survey further revealed that on average, consumers say they tip between 10% and 19% for a food delivery order.

“Interestingly, 15% say they tip less than 10% of the total order, and 14% say they do not tip at all for this service,” Balagtas said. “We see little difference in the tipping percentages when disaggregating the responses by per-person weekly food spending.”

The sustainable food purchasing index remained unchanged from the CFI survey’s last assessment in June 2024.

“Consumers continue to purchase food that they feel is safe and fits their tastes, budgets and nutritional needs,” said Elijah Bryant, a survey research analyst at CFDAS and a co-author of the report. Fewer consumers currently buy or plan to buy foods with environmental and social sustainability in mind.

“Even though consumers may value the environmental impact and social responsibility of their food, when it comes to purchasing factors, more immediate priorities like food security, taste, economic factors and nutrition drive their decisions,” he said.

Since its inception in January 2022, the CFI survey has documented a gradual positive trend in per-person weekly food expenditures. In January 2022, the figure was around $72. Last month, consumers reported an average per-person weekly spending total of $83, a 15% increase.

“Consumers are having to adjust their budgets to accommodate higher food prices to purchase the same groceries,” Bryant said. “Wage growth will be a key determinant in food purchasing behavior changes as food prices remain higher after inflation spiked in 2022.”

Based on the USDA’s questionnaire for measuring food insecurity, the CFDAS researchers estimate the national food insecurity rate to be 13%, unchanged from last month. The rate of food insecurity is highest among households that spend less than $50 on food per person per week.

“We have seen a clear correlation between income and food security in the past and see that many households that spend less on food are likely doing so due to income constraints,” Bryant said. Around 29% said they use free food resources, such as food banks, to supplement their diets. This shows the importance of these resources for people who struggle with food insecurity due to a lower food budget, he said.

Around 14% of thrifty food spenders adhere to either a vegetarian or vegan diet, relative to just 6% of moderate and liberal food spenders. Thrifty spenders also report growing their own food in either a home or community garden at a higher rate (32%) than moderate (24%) and liberal (21%) food spenders.

“We do not observe many substantial differences in the frequency of a variety of surveyed food behaviors between the spending groups,” Bryant said. “However, we do observe thrifty food spenders choosing generic foods over brand-name foods more frequently than moderate and liberal spenders.”

In line with the larger share of vegans and vegetarians in the thrifty group, they are also more likely to choose plant-based proteins over animal proteins.