In June, data labeling startup micro1 approached Robbie Hogle, CEO of Vitality Mechanical in Troy, Michigan, with a proposal: integrate artificial intelligence into his HVAC business while earning compensation. Hogle, a former investment banker, had previously provided feedback to micro1 on how AI models performed financial modeling tasks. Now, micro1 wanted him to test various AI models on his daily work, including processing supplier invoices, filling out credit applications, and calculating payroll for the company's 14 employees. Hogle agreed, saying he would identify tedious tasks for automation and rate the AI models on speed, efficiency, and accuracy, delivering evaluations to micro1 without knowing which company's model he was assessing, as all model information is anonymized. Like other data labeling startups, micro1 likely collaborates with leading AI labs.

Micro1 is pursuing similar partnerships with small business owners across industries like marketing and professional services, aiming to address a persistent challenge: despite rapid AI progress in fields like math and physics, many enterprise clients complain that AI struggles with simple office tasks such as inventory ordering or new employee onboarding. To solve this, AI developers are collecting new training data, such as creating reinforcement learning environments that simulate common office software like Salesforce and Excel, to improve AI's ability to operate these tools. Data labeling companies like micro1 now gather real business data from small merchants, like Hogle's firm, to further expand AI's capabilities. Micro1 pays Hogle $1,000 per labeling task, and he expects to complete 150 to 200 tasks this quarter, potentially generating over $150,000 in revenue from data labeling alone.

For Hogle, the partnership is risk-free. He notes that the work of automating backend processes with AI is something he needs to learn for his own business; he already spends time weekly on payroll and invoice processing, and this side gig only adds 3-4 hours of extra work. His HVAC company, established less than a year ago, hasn't hired dedicated backend staff, and these roles are expected to be automated through AI via data labeling. Hogle mentions that other local small businesses face greater challenges, as backend tasks consume time even if core operations aren't automated. He acknowledges this extra income may not last long, stating, "This revenue window is likely only 3 months, 6 months, or 9 months... our window for earning through this side gig is very brief."

Other industry news: Call to slow down frontier AI development

Over 1,100 senior executives and employees from leading AI companies have signed an open letter to the U.S. government, urging "international coordination to build mechanisms for technology and governance, actively managing the pace of development of advanced autonomous AI." This is not the first such appeal; in 2023, tech figures like Elon Musk and Steve Wozniak called for a six-month halt to training AI models more powerful than OpenAI's GPT-4. AI companies ignored that request, with some joking they used the letter as training data. However, industry conditions have changed significantly in three years, and this letter may better advance discussions on global governance policies for high-performance AI.

In recent months, calls for coordinated international controls and slowing AI iteration have intensified. Last month, AI company Anthropic proposed governance concepts addressing the risk of "recursive self-improvement" (AI accelerating its own capabilities). This letter echoes that concern, stating: "Leading global AI companies believe humanity is close to achieving AI that can independently develop its own technology. If realized, AI capabilities will surge rapidly, making these systems incomprehensible and uncontrollable, posing real risks." At the World Economic Forum in Davos earlier this year, Google DeepMind CEO Demis Hassabis expressed willingness to support a temporary pause in the AI race if industry consensus is reached, but his signature does not appear on this letter. Over the past three years, AI practitioners have issued numerous petitions, warning of existential risks, demanding whistleblower protections, supporting AI safety legislation, backing ousted CEOs, opposing AI-military cooperation, and defending Anthropic against Pentagon-related allegations. Interestingly, if AI were to achieve full autonomous development, it could independently handle even organizing a public petition.