Generative artificial intelligence (genAI) has commanded the attention of enterprises across industries — and at this point there’s really no turning back or falling behind.

To remain competitive, enterprises are amassing more data than they can handle to fuel increasingly sophisticated models.

Behind all this: A workforce of silent, unappreciated workers. As industries are poised to make billions on AI, the workers are making mere dollars under unfair working conditions performing the meticulous, exhaustive, relentless job of data annotation — the very building block of AI.

Cogito Tech aims to pull back the curtain on this issue and demand accountability and transparency in AI development. Today, the data labeling and annotation company is rolling out DataSum, what it refers to as a “nutrition facts”-style framework for AI training data.

DataSum “empowers users to understand the ingredients that go into AI algorithms,” said Matthew McMullen, SVP of Cogito. “It’s answering the call for a trustworthy, comprehensive, supply chain framework to understand how AI is built.”

What is DataSum?

As McMullen explained, Cogito seeks to promote humane practices, equitable conditions and ethical labor standards in the AI data-labeling sector, particularly in developing nations with heightened risk of exploitation.

DataSum is a certification framework designed to audit and validate training data at various touchpoints in the data pipeline. McMullen said it focuses on best practices and adheres to ethical guidelines, methodologies and standards to help ensure the development of responsible, transparent and unbiased training data.

The five certifications within the framework include:

• Ethical Integrity: Alignment with ethical standards such as bias, fair use and privacy.
• Workforce Well-being: Humane practices and equitable conditions.
• Technological Profile: Visibility into tools, platforms and technologies.
• Quality Assurance: Validates the labeling accuracy and quality through documented techniques such as double-blind quality checks.
• Efficiency Technique: Certifies the strategies implemented for efficient data processing, including methodologies like fine-tuning and reinforcement learning from human feedback (RLHF).

“DataSum's certifications bring a heightened focus on the welfare of the workforce, scrutinizing and validating the conditions, compensations and overall well-being of the labor,” McMullen explained.

Data and AI platforms and companies in sectors including AI monitoring, data warehousing, cloud-based machine learning (ML) and data annotation can integrate the certifications into their operations to underscore their commitment to ethical sourcing and quality AI development, he said.

Furthermore, organizations in the AI space could develop unique certifications that reflect their specific governance, compliance and ethics measures and practices in data handling and management.

McMullen pointed out that adhering to high data quality standards is “increasingly important in a world where consumers and regulatory bodies are calling for greater transparency and ethical considerations in AI applications.”

$2.15 an hour wage for an industry worth billions

Data labelers typically work through business process outsourcing (BPO) and crowdsourcing platforms.

However, one recent report found that not one of the 15 most popular web-based digital labor platforms — including Amazon Mechanical Turk, Scale AI, Fiverr and Upwork — met the minimum standard for fair labor principles (that is, fair pay, fair conditions, fair contracts, fair management and fair representation).

That same research also revealed that workers in underdeveloped areas including Kenya, the Philippines and India earn an average of $2.15 per hour, while nearly 27% of their work time consisted of unpaid tasks.

“This is trillions of dollars worth of technology and workers are still making $2,” McMullen marveled. “Most people don't know this part of the world, most people don't know this part of AI.”

He pointed out that these laborers also often find themselves compelled to enter into strict contractual agreements that grant data vendors exclusive ownership of their professional identity. Furthermore, projects often have code names, so there’s no way workers can validate that they worked on a project or reach out for references.

This inhibits career mobility and the ability to negotiate better wages and treatment, as well as to unionize.

“Most people don't understand the dramatic impact on an entire generation of workers,” said McMullen. “It’s binding an entire generation of workers out of existence.”

As the AI arms race heats up, responsible sourcing goes by the wayside

The intensifying AI arms race is the crux of the problem. Large organizations need to collect and label “libraries upon libraries of data” to be competitive. They are looking to deploy AI as quickly, efficiently and inexpensively as possible, McMullen noted. Added to that is the fact that development lifecycles are extensive and “super expensive,” as is training data.

“It's the urgency of getting things from R&D to commercialization,” he said.

Often, organizations are unaware of unethical conditions — or sadly, they don’t care, McMullen said.

“Responsibly sourcing the data isn't part of the equation,” he said.

DataSum aims to spotlight the “indispensable work of data labelers” and expose the “human pillar within the secrecy of AI development.”

Impact sourcing doesn’t sell — but it’s critical

The good news is that BPO companies like Cogito have been growing around the socially responsible concept of impact sourcing.

In Cogito’s case, this brings things down to the “individual perspective,” paying more for traditional work while offering benefits and perks such as phantom stocks and even supporting workers in their daily lives (such as helping them secure transportation).

“We’re trying to give them a more holistic idea of culture outside of work,” said McMullen.

He acknowledges that, as of yet, “impact doesn’t sell,” but noted that it’s critical for Cogito. Ultimately, DataSum reflects that.

“For us, it's not merely a business move,” he said, “but a step closer to an industry standard that we believe is necessary.”