Think AI will replace human data creators anytime soon? Think again! The CEO of a $2 billion AI training startup believes humans will be crucial to AI development for decades to come. But here's where it gets controversial: are they right, or are they underestimating the pace of technological advancement?
According to Shubhangi Goel of Business Insider, Matt Fitzpatrick, CEO of Invisible Technologies, a company specializing in AI training data, recently shared his insights on the "20VC" podcast. Fitzpatrick boldly stated that the notion of humans becoming obsolete in AI training within the next few years is a major misconception.
"When I first started this job, the main push back I always got was that synthetic data will take over and you just will not need human feedback two to three years from now," Fitzpatrick explained, having joined Invisible Technologies the previous year. "From first principles, that actually doesn't make very much sense."
Let's break down what he means. Synthetic data is artificially generated data used to train AI models, especially when real-world data is scarce or raises privacy concerns. Imagine training an AI to recognize different types of skin cancer. Acquiring enough real-world images while protecting patient privacy can be incredibly difficult. Synthetic data offers a workaround. Human feedback, on the other hand, involves real people refining, ranking, and improving AI responses. Think of it as a human editor polishing a rough draft written by an AI.
Fitzpatrick argues that the sheer diversity of tasks AI needs to perform, coupled with the nuances of language and cultural context, makes complete reliance on synthetic data impractical in the near future. He pointed to the legal industry as an example, highlighting the vast amounts of non-public information that require careful handling and human understanding. And this is the part most people miss: AI's ability to truly understand context, especially within specialized fields, is still limited.
"On the GenAI side, you are going to need humans in the loop for decades to come," he emphasized. "And I think that is something that most people are starting to realize." This perspective aligns with a growing recognition that while AI is rapidly advancing, human oversight remains essential for ensuring accuracy, fairness, and ethical considerations.
Before joining Invisible Technologies, Fitzpatrick was a senior partner at McKinsey, leading QuantumBlack Labs, their AI research and development division. Invisible Technologies, valued at $2 billion after a $100 million funding round, competes with other data labeling companies like Scale AI and Surge AI. These companies are experiencing massive growth, fueled by tech giants' insatiable demand for high-quality data to train their AI models. They employ millions of human contractors to teach AI everything from basic math and coding to more complex concepts like humor and empathy.
It's worth noting that Fitzpatrick isn't alone in his assessment. Brendan Foody, CEO of Mercor, another data labeling company, stressed the importance of data quality and treating human annotators well. Similarly, Garrett Lord, CEO of Handshake (a job platform that shifted into AI training), believes humans will continue to play a vital role, though the required skill sets are evolving from general knowledge to specialized expertise in fields like math and science. This shift suggests that while the type of human input is changing, the need for human involvement isn't disappearing.
"Now these models have kind of sucked up the entirety of the entire corpus of the internet and every book and video," Lord said on a podcast. "They've gotten good enough where, like, generalists are no longer needed."
So, what does this all mean? The AI landscape is dynamic, and opinions on the future of human involvement vary. But here's a question for you: Do you agree with Fitzpatrick's assessment that humans will be essential for AI training for decades? Or do you believe AI will eventually become capable of training itself, rendering human input obsolete? And if so, what ethical implications might that have? Share your thoughts in the comments below!
