“The field of AI is moving so fast, putting up guardrails is extremely difficult,” says Greg Dobler, an astrophysicist turned urban data scientist at UD.

The Wild West of AI

Assessing AI regulation

Diane Stopyra November 26, 2024

If regulating artificial intelligence feels about as likely as flying to Jupiter for a space odyssey, there’s good reason.

“The field of AI is moving so fast, putting up guardrails is extremely difficult,” says Greg Dobler, an astrophysicist turned urban data scientist at UD. “It’s nearly impossible to contain some of these developments.”

Dobler is not an AI doomsdayer. He uses the technology to improve lives in various ways; during the height of the COVID-19 pandemic, for example, algorithms he created helped hospitals predict—and plan for—bed demand. But as someone whose work often intersects with policymaking, he sees regulation as a formidable challenge.

UD Professor Greg Dobler

For starters, we have a literacy problem. The lawmakers tasked with containing AI often do not understand the science. And given the technology’s rapid evolution, any crash-course-for-Congress is nearly obsolete by the time it’s developed.

Additionally, regulation requires establishing an AI code of ethics, a thorny proposition globally—rival nations are unlikely to agree on the use of AI for, say, creation of new biological weapons—and domestically. Who gets to decide whether Americans should be comfortable with an artificially intelligent president? On the one hand, a robot leader may have greater capacity to process every budgetary and geopolitical footnote. On the other, we’re talking about a mechanical commander-in-chief.

We have a literacy problem. Lawmakers tasked with containing AI often do not understand the science.
-

Then there’s the issue of execution. Even if humanity can agree on ethical standards, enforcing them presents a whole new barrel of robots. Policymakers can attempt to regulate the building of AI models—No recurrent neural networks! Only eight layers of encoders!—but engineers will likely find workarounds. Policymakers can also attempt to regulate the output of machines, but that requires an almost unimaginable foresight. Few would have predicted the phenomenon of “revenge porn” (when a jilted lover uses AI to generate fake-but-realistic nude photos in order to shame an ex on the Internet), or ChaosGPT (a rogue autonomous chatbot working as you read this to annihilate humanity.)

For these reasons, attempts at regulation at the state, federal and international levels have thus far been vague, Dobler says: “We’re reacting to what’s happening now. Instead, we need to think about what will be possible in the future and the implications of that.”

If there’s any hope for movement on the regulatory front, decision makers cannot operate in a vacuum. Politicians will need to be in regular conversation with people building the next generation of AI models (tricky due to intellectual property protections), as well as serious AI ethicists. In other words, regulation will require a fundamental change in our legislative process.

Questions coming from all sides of the political spectrum are nearly identical, and it is unique to see such a common purpose.
-

It’s a tall order, but Dobler believes there is reason to feel hopeful. For starters, everyone—from every corner of the globe—is invested and engaged. In April 2024, when the United Nations passed a landmark resolution urging world leaders to recognize the power and pitfalls of AI, all 193 member nations signed on—no vote necessary. And because AI hasn’t yet been politicized, American legislators in particular find themselves at a rare and critical moment in the political arena.

“There’s bipartisan agreement that something big is coming,” Dobler says. “Watching Congressional testimony on AI, questions coming from all sides of the political spectrum are nearly identical, and it is unique to see such a common purpose.”

If nothing else, take the unlikely (and probably temporary) alignment of America’s political forces as irrefutable proof: When it comes to AI, all things—space odysseys, effective regulatory strategies and sci-fi realities yet to be revealed—are possible.