The CEO of the $2 billion AI training startup, Invisible Technologies, asserts that human involvement in data creation will remain crucial for the foreseeable future. During a recent episode of the “20VC” podcast, Matt Fitzpatrick emphasized that the widespread belief that synthetic data will eliminate the need for human input within a few years is fundamentally flawed.
Fitzpatrick, who joined Invisible Technologies in 2022 and previously served as a senior partner at McKinsey, expressed concerns over this prevailing misconception. He stated, “When I first started this job, the main pushback I always got was that synthetic data will take over and you just will not need human feedback two to three years from now. From first principles, that actually doesn’t make very much sense.”
The dialogue surrounding synthetic data highlights its role in training AI models, especially where real data is limited or restricted due to privacy issues. However, Fitzpatrick argues that human feedback, which involves real individuals filtering and ranking AI responses, is irreplaceable. He pointed out the diverse range of tasks AI must tackle, underscoring that it would take a significant amount of time to execute these tasks accurately while considering language and cultural nuances.
Continued Demand for Human Expertise
Fitzpatrick’s comments resonate within the broader AI training landscape, where demand for human labor remains robust. He noted that in sectors like law, a wealth of nonpublic information requires human insight to navigate effectively. He stated, “On the GenAI side, you are going to need humans in the loop for decades to come, and I think that is something that most people are starting to realize.”
Invisible Technologies recently raised $100 million in funding, achieving a valuation of $2 billion. This positions the company alongside other notable data labeling firms such as Scale AI and Surge AI, both of which have attracted substantial investments as technology companies compete for quality data to train their AI systems. These firms employ millions of human contractors who assist in teaching AI models various subjects, from mathematics and science to more nuanced characteristics like humor and empathy.
Fitzpatrick is not alone in his perspective. Other leaders in the data labeling sector echo similar sentiments regarding the necessity of human involvement. Brendan Foody, CEO of Mercor, highlighted in September 2023 that the cornerstone of their business lies in “data quality and having phenomenal people that you treat incredibly well.”
In July 2023, Garrett Lord, CEO of Handshake, which pivoted into AI training, noted that while humans remain essential in AI training, the qualifications for these roles are evolving. He remarked that the industry is moving away from generalist roles towards a demand for specialists with expertise in areas such as math and science. “Now these models have kind of sucked up the entirety of the entire corpus of the internet and every book and video,” Lord explained, indicating that AI systems have reached a level of proficiency that diminishes the need for generalist input.
As AI continues to evolve, the discourse surrounding the intersection of human capability and synthetic data will likely persist. With industry leaders like Fitzpatrick advocating for sustained human involvement, it is clear that the future of AI training will remain a collaborative effort between technology and human expertise.
