Humans have long been astonished by computers. When these machines first emerged, they felt almost magical—capable of solving problems at speeds no human could match, executing tasks with perfect precision, and transforming industries one algorithm at a time.
Fast forward a few decades, and the fascination has only deepened—but the object of awe has shifted. It’s no longer just about computers that follow commands, but about machines that can learn, adapt, and evolve. Artificial Intelligence (AI) has moved humanity beyond the boundaries of traditional hard-coded software.
Instead of being told exactly what to do, AI systems can now observe, predict, and even create.
This transformation—from predictable logic to dynamic intelligence—has opened up entirely new possibilities. AI is no longer just a tool; it’s becoming a collaborator, a problem solver, and in some cases, a storyteller. Whether it’s powering recommendations, driving cars, or generating artwork, AI has started to feel less like a program and more like a presence.
Trenton Bricken, researcher at Anthropic, sat down in an interview with AI podcaster Dwarkesh Patel, with another fellow Anthropic researcher Sholto Douglas.
The conversation discuss the potential of combining Reinforcement Learning (RL) and Large Language Models (LLMs) to achieve Artificial General Intelligence (AGI). They explore whether these technologies, when integrated, can lead to machines with human-like cognitive abilities.
It delves into the strengths and limitations of RL and LLMs, examining how each contributes to learning and decision-making processes. They discuss the challenges of creating systems that can generalize knowledge across diverse tasks and adapt to new situations—key characteristics of AGI.
Additionally, the speakers consider the role of embodiment and interaction with the physical world in developing general intelligence. They debate whether purely digital models can achieve AGI or if a form of physical presence is necessary for true understanding and learning.
Throughout the video, Douglas and Bricken provide insights into current research directions, potential future developments, and the philosophical implications of pursuing AGI through RL and LLMs.
And so, the story continues—not of man versus machine, but of humans working alongside something new, something both familiar and unfamiliar.
The question now is no longer "What can computers do?" but rather, "What can we imagine—together?"
Bricken foresee a dystopian future where AI take over, and humans work for AI overlords.
"So you have human meat robots."
Trenton Bricken used the phrase “human meat robots” during a discussion, to discuss a possible future, where AIs can do nearly everything except physical robotic tasks.
In this imagined world, humans become the physical appendages of AI systems—executing instructions, moving objects, and interfacing with the real world, but stripped of autonomy. Intelligence, agency, and decision-making would be offloaded entirely to machines, with people reduced to laboring bodies directed moment-by-moment through wearable devices.
He wasn’t suggesting this will definitely happen, but rather raising it as a plausible economic scenario.
He warns about what could happen if we allow market forces and technological development to proceed without deliberate human values steering the course.
As AIs become more capable—particularly in areas like decision-making tasks, planning, coding, and creative work—the pressure to reduce human involvement to mere physical labor may grow.
The remaining economic value in humans might just be their ability to physically interact with the world—turning them into extensions of AI systems, guided moment-to-moment like literal puppets
.
Bricken clarifies he’s not claiming AIs would want to do this—but if things are left to raw economic efficiency, it’s a future that could emerge.