The CEO of $2 billion AI training startup says that humans will stay involved in data creation for decades

The landscape of artificial intelligence is set to keep humans in the loop for the foreseeable future, according to Matt Fitzpatrick, CEO of the data labeling startup Invisible Technologies. In a recent episode of the '20VC' podcast, Fitzpatrick addressed a prevalent myth in the AI training sector: the belief that synthetic data will soon replace human involvement in data creation. Fitzpatrick, who joined Invisible Technologies last year, recalled the skepticism he faced when he began his role. "When I first started this job, many believed that synthetic data would dominate, eliminating the need for human feedback within just a couple of years. However, that perspective lacks a solid foundation," he remarked. Synthetic data is artificially generated to train AI and machine learning models, particularly in areas where real data is either scarce or fraught with privacy issues. In contrast, human feedback involves actual individuals who help filter, rank, and refine AI outputs. Fitzpatrick emphasized the complexity of tasks that AI must handle, particularly regarding language and cultural nuances, suggesting that a significant amount of time will be needed for these systems to perform accurately. He pointed out that certain sectors, like the legal field, contain vast amounts of sensitive information that require human insight. "For generative AI applications, human involvement will be critical for decades," he asserted, a sentiment he believes is gaining wider acceptance. Fitzpatrick's insights come as Invisible Technologies, which recently secured $100 million at a valuation of $2 billion, competes with other data labeling firms like Scale AI and Surge AI. These companies have collectively raised billions recently as major tech players strive to acquire the necessary data for training their AI systems. They employ millions of human contractors to instruct AI models in various subjects, including mathematics, science, coding, and even traits like humor and empathy. His views resonate with other leaders in the data labeling industry. In a recent statement, Brendan Foody, CEO of Mercor, emphasized that top-notch data quality hinges on having exceptional people who are well-treated. Similarly, Garrett Lord, CEO of Handshake, indicated a shift in the industry’s requirements, noting that while human trainers remain essential, the focus is moving from generalists to specialized experts in fields like math and science. As AI continues to evolve, the role of humans in training these advanced systems is likely to remain crucial for the foreseeable future.