Is AI Headed in the Wrong Direction?
Alihan Beba, BF Youth Program Advisor / Design and Business
History has a habit of warning us before we listen. The automobile was supposed to revolutionize how we move through the world - and it did, just not always for the better. Despite the existence of more efficient alternatives like buses and rail, policy decisions and infrastructure investments locked cities into car-dependent systems, bringing with them congestion, pollution, urban sprawl, and oil dependence. We didn't choose the best transportation technology. We chose the one that private incentives made easiest to adopt.
Artificial intelligence may be following the same road.
AI's potential is real and significant. In education, adaptive learning platforms can deliver personalized tutoring at scale. In manufacturing, AI-powered systems predict equipment failures, streamline supply chains, and improve quality control. Across nearly every sector, the technology promises to make human work faster, smarter, and more productive.
But potential and trajectory are two different things. Recent evidence suggests that firms are not primarily investing in AI that complements human labor - they're investing in AI that replaces it. IBM data shows that among companies using AI to address labor or skills shortages, 55% are deploying it to reduce repetitive tasks and 47% to automate customer service functions. The World Economic Forum reports that 40% of employers expect to shrink their workforces where AI can take over. As economists Daron Acemoglu and Pascual Restrepo have argued, technological change has become heavily biased toward automation, with too little focus on creating new roles where workers can thrive.
The reason for this bias is not hard to find. Automation delivers immediate cost savings. Labor-augmenting AI on the other hand - the kind that makes workers more productive rather than redundant - requires complementary investments in training, organizational redesign, and process change before any gains materialize. For firms optimizing short-term profits, the math is not close. The socially better technology loses to the privately cheaper one.
This is how path dependence takes hold. Early adoption patterns shape infrastructure, regulation, and incentives. Once automation-heavy AI becomes embedded in production systems and corporate structures, transitioning toward more worker-friendly alternatives becomes increasingly costly - much like trying to build a subway system in a city designed entirely around highways.
The good news is that we are not there yet. Policymakers still have meaningful leverage, but the window is certainly narrowing. Three interventions have the potential to be effective: a Pigouvian-style tax on labor-displacing automation that forces firms to internalize the social costs they currently pass onto workers and communities; targeted subsidies and R&D support for AI that demonstrably augments human productivity; and regulatory frameworks designed to prevent the firms most invested in automation from writing the rules that govern it.
The challenge is not to slow AI down. It is to guide it. The automobile succeeded - and its success gave us a century of sprawl and pollution that no city has escaped. We have a rare chance to get this one right before the concrete sets.
