Matt Fitzpatrick, CEO of data labeling startup Invisible Technologies, says artificial or synthetic data will not replace human involvement in training AI systems for decades. Speaking on the "20VC" podcast, Fitzpatrick challenged a widespread industry belief that synthetic data would soon render human feedback obsolete within just a few years. “When I first started this job, the main pushback I always got was that synthetic data would take over and you wouldn’t need human input in two to three years,” Fitzpatrick said. “From first principles, that actually doesn’t make very much sense.” Synthetic data—artificially generated information used to train AI models—is often employed when real-world data is limited or sensitive. However, Fitzpatrick emphasized that the diversity and complexity of real-world tasks, especially those involving nuanced language, cultural context, and domain-specific knowledge, cannot be adequately addressed by AI alone. He pointed to industries like law, where vast amounts of confidential and highly specialized information exist. “On the GenAI side, you’re going to need humans in the loop for decades to come,” he said. “And I think that’s something most people are starting to realize.” Fitzpatrick, a former senior partner at McKinsey and head of QuantumBlack Labs, joined Invisible Technologies last year. The company raised $100 million in September at a $2 billion valuation. It competes with major data labeling firms like Scale AI and Surge AI, which have collectively attracted billions in funding as tech giants race to secure high-quality training data. These startups rely on millions of human contractors to perform tasks ranging from labeling images and text to ranking AI-generated responses based on accuracy, tone, and empathy. The work helps train models in areas like math, science, coding, and even emotional intelligence. Fitzpatrick’s view aligns with other industry leaders. In September, Brendan Foody, CEO of Mercor, stressed that data quality hinges on having “phenomenal people” who are well-treated and highly skilled. Similarly, Garrett Lord, CEO of Handshake—a job platform that shifted into AI training—said the industry is evolving from generalists to specialists. “Now these models have absorbed nearly all the content on the internet, every book, every video,” Lord said on a podcast. “They’ve gotten good enough that generalists are no longer needed. The future is in experts—people with deep knowledge in math, science, and other technical fields.” As AI continues to advance, the consensus among data labeling leaders is clear: human oversight remains essential, not just for now, but for the foreseeable future.