Switzerland's AI Regulation Delay: A Live Case Study of AI Skepticism

Switzerland's "wait until 2028" AI regulation strategy might be exactly what Gary Marcus warns against - and exactly what proves his point. If AI systems are as fundamentally flawed as he argues, their limitations will become more apparent during this deliberate delay. The irony? Switzerland's regulatory patience could validate Marcus's skepticism about AI's capabilities, even as it ignores his urgency about AI's risks. A rare case where doing nothing might prove everything.

In what might be the most Swiss approach imaginable to artificial intelligence regulation, the country has decided to wait until 2028 to implement comprehensive AI rules. This deliberate delay has drawn criticism from AI watchdogs and advocacy groups. Yet, in a peculiar twist, this seemingly passive approach might provide the strongest evidence for one of AI's most vocal critics, Gary Marcus . [Great article in NZZ by Ruth Fulterer]

The Paradox of Patience

At first glance, Switzerland's strategy exemplifies everything Marcus warns against. As he testified before the U.S. Senate in 2023, "We have built machines that are like bulls in a china shop—powerful, reckless, and difficult to control." The Swiss timeline, stretching into 2028, ignores this urgency entirely.

But here's where the paradox emerges: If Marcus is correct about the fundamental limitations of current AI systems—their tendency to fabricate facts, their inability to reason like humans, their failure to truly understand context—then Switzerland's waiting period will serve as an extended demonstration of these very limitations.

Time Will Tell

While the EU rushes to implement comprehensive AI regulations and the U.S. grapples with conflicting approaches under different administrations, Switzerland's patience might inadvertently create a natural experiment. During this period, several of Marcus's key critiques can be tested:

1. The reliability gap: Marcus argues that AI "doesn't deserve" the trust people place in it. Switzerland's delay gives time for this trust to be either validated or, as Marcus would predict, eroded through continued exposure to AI's limitations.
2. The AGI timeline: While companies like OpenAI invest billions in projects like Stargate, betting on rapid progress toward artificial general intelligence, Switzerland's timeline implicitly suggests a more measured view of AI development—one that aligns more closely with Marcus's skepticism about imminent breakthroughs.
3. The statistical learning problem: Marcus's fundamental critique—that purely statistical approaches to AI cannot achieve human-like intelligence—will have years to prove itself correct or incorrect under Switzerland's wait-and-see approach.

Strategic Inaction

Switzerland's approach is intriguing because it isn't purely passive. The country isn't ignoring AI; rather, it is?choosing?to rely primarily on existing legal frameworks while observing how other jurisdictions' new regulations perform. This creates a controlled environment in which Marcus's criticisms about AI's limitations can be played out against the backdrop of real-world applications.

The Validation Paradox

The ultimate irony might be this: If Switzerland's delay leads to better, more informed regulation in 2028, it could validate Marcus's position that current AI systems aren't as capable or reliable as their proponents claim. Conversely, if the delay proves dangerous or costly, it would also validate Marcus's warnings about the risks of inadequate oversight.

Either way, Switzerland's patience might provide the empirical evidence needed to assess the true state of AI capabilities and risks—even if that wasn't the intended purpose of their regulatory timeline.

Conclusion

In the rapid-fire world of AI development, where companies race to announce breakthrough after breakthrough, Switzerland's measured approach might seem outdated or overly cautious. Yet, it could provide exactly the kind of longitudinal perspective needed to evaluate Marcus's fundamental critiques of AI.

The Swiss approach suggests an unexpected truth: Sometimes, the best way to prove a technology's limitations is simply to give it enough time to demonstrate them itself. In waiting until 2028, Switzerland might inadvertently be setting up the most comprehensive test of AI skepticism to date.

Whether this experiment in regulatory patience proves wise or foolish, it will be illuminating. And in that sense, both Switzerland and Gary Marcus might end up being right—just for entirely different reasons than either intended.

CHRISTINE HASKELL, PhD, is a collaborative advisor, educator, research editor, and author with 30 years in technology, driving data-driven innovation and teaching graduate courses in executive MBA programs at Washington State University’s Carson School of Business and is a visiting lecturer at the University of Washington’s iSchool. She lives in Seattle.

ALSO BY CHRISTINE

Driving Your Self-Discovery (2024), Driving Data Projects: A comprehensive guide (2024), and Driving Results Through Others (2021)