AI Experts See Westworld as a Blueprint for Future Machine Intelligence

Compiled by the editorial desk with reference to interviews and statements from Hanson Robotics and the Allen Institute for Artificial Intelligence.

In the HBO series Westworld, robots with artificial intelligence run amok, but experts say the show's portrayal of AI is not as far-fetched as it seems. While the androids in the series—like Dolores Abernathy—are capable of self-awareness and emotional depth, real-world AI, such as Hanson Robotics' Sophia, is still in its infancy. However, researchers at Hanson Robotics and the Allen Institute for Artificial Intelligence see the show as a plausible roadmap for the future of intelligent machines.

Stephan Bugaj, vice president of creative at Hanson Robotics, who leads personality design for the company's robots, explains that two key software systems separate the fictional AI of Westworld from today's reality. The first is generative code—a system that can write its own code to adapt to new situations. Unlike current neural networks, which learn from data, generative code would allow an AI to reason about itself and create new programs as needed. "An A.I. could reason about itself and decide that it needs some new code, and write it itself," Bugaj says.

The second system is memory, but not just any memory. For AI to truly learn, it needs memory that mimics human cognition—short-term, long-term, and episodic. This type of memory allows for associative thinking, which humans use to draw analogies and make intuitive leaps. Oren Etzioni, CEO of the Allen Institute, notes that no current system comes close to this capability. "We use analogies, intuition, this kind of reminding thing, to do a lot of great work," he says. "And that capability—people call it associative, people call it content-addressable, any of a number of things—we ain't got any systems that come even close."

What Today's AI Can Actually Do

Current AI systems can see, speak, and understand enough to make decisions, such as suggesting which friend to tag in a photo or even driving a car. Neural networks, which are state-evolving over time, can learn from new data, but they lack the ability to generate new code or possess spontaneous, creative memory. Bugaj explains that while neural networks evolve, they are not yet capable of the generative code that would mark the next evolutionary step in machine intelligence.

The gap between Westworld and reality is significant, but Bugaj believes the show serves as a valuable thought experiment. "I think they're doing what a good futurist should do, which is making conjectures about the future, and then exploring how those might play out," he says. The show's depiction of AI, with its semantic networks and self-modifying code, is not entirely out of reach, but it remains firmly in the realm of science fiction for now.

As research continues, the lessons from Westworld may inspire new approaches to AI development. But as Etzioni and Bugaj both emphasize, the path to truly intelligent machines requires solving the fundamental challenges of memory and generative code—a task that will take time and extensive research.