← Back to Library

Import AI 419: Amazon's millionth robot; CrowdTrack; and infinite games

Jack Clark's latest dispatch from Import AI cuts through the hype cycle to reveal a startling convergence: the infrastructure for a surveillance state is being built by private logistics firms, while the tools for infinite digital escapism and biological redesign race toward maturity. This isn't just a roundup of tech news; it is a chilling inventory of the components required to build a world where human agency is either optimized into obsolescence or drowned in an endless stream of synthetic reality.

The Architecture of Turnkey Authoritarianism

Clark begins by examining a new dataset from Fudan University called CrowdTrack, which aims to solve the persistent problem of tracking individuals in chaotic, fast-moving crowds. While the researchers frame this as a technical benchmark for "visual grounding" and "appearance feature extraction," Clark reframes it as a critical enabler for state control. He notes that the dataset intentionally includes footage from construction sites where workers wear uniforms and helmets, specifically to suppress facial recognition and force the AI to rely on gait and body shape.

Import AI 419: Amazon's millionth robot; CrowdTrack; and infinite games

"All data is collected in unconstrained daily environments, ensuring object behaviors remain natural and unmodified," the researchers write. Clark argues that this shift is profound because it lowers the economic barrier to total surveillance. He posits that the primary cost of authoritarianism has always been the massive overhead of human police forces, but AI removes that friction. "One of the things AI does is make it much, much cheaper to do large-scale surveillance," Clark writes, suggesting that datasets like CrowdTrack are the symptoms of a future where dictators can achieve the panopticon fantasies of the 20th century without the budget to match.

Datasets like CrowdTrack are a symptom of the way AI is making it cheaper and easier to do surveillance that the dictators of the 20th century would have fantasized about but always been unable to fully fund.

The argument here is compelling because it shifts the focus from the capability of the technology to the economics of its deployment. Critics might argue that democratic societies have legal guardrails that prevent such tools from being used for mass tracking, but Clark's point is that the technology itself is becoming agnostic to the regime type that deploys it. Once the cost drops, the temptation for any government to utilize it becomes nearly irresistible.

The Autonomous Corporation

The commentary then pivots to Amazon, which recently deployed its millionth warehouse robot. Clark observes that this milestone is not merely a logistical achievement but a step toward a "fully automated infrastructure for a superintelligence." He highlights the introduction of "DeepFleet," a software system that coordinates robot movements to reduce travel time by 10%. Amazon describes this as a smart traffic system for their fulfillment centers, but Clark sees something more ominous: the creation of a physical substrate that could one day be handed over to an autonomous entity.

"Just as a smart traffic system could reduce wait times and create better routes for drivers, DeepFleet coordinates our robots' movements to optimize how they navigate our fulfillment centers," Amazon writes. Clark interprets this as the corporation building the nervous system for a future AI that might eventually run the company itself. He suggests that Amazon is effectively constructing the hardware and software layers that a superintelligence would need to operate as an autonomous economic agent within a human-run world.

This framing is provocative. It forces the reader to consider that the efficiency gains celebrated in corporate earnings calls are actually the foundational work for a post-human economy. A counterargument worth considering is that these systems remain deeply supervised by humans and are far from the general intelligence required for true autonomy. However, Clark's point is about the trajectory of the infrastructure, not its current state of completion.

The Illusion of Infinite Play

Moving from physical logistics to digital escapism, Clark analyzes "Mirage," a new generative game engine from Dynamic Labs that promises an "infinite, endlessly generated game." The technology uses large transformer models to create real-time, procedurally generated worlds where players can prompt the game to create roads or delete enemies. Clark is candid about the current limitations, noting that the experience is "extremely janky" and that the world consistency breaks down quickly when the camera rotates.

"Mirage is the world's first real-time generative engine enabling live UGC gameplay through state-of-the-art AI World Models," the company claims. Yet, Clark's assessment is more grounded: "GGN games are almost fun, and I expect they'll be actively fun to play in a year." He draws a parallel to David Foster Wallace's novel Infinite Jest, warning that this technology could lead to "the entertainment in reality"—a hyper-stimulating media environment so compelling that it blurs the line between the game and the real world.

We're likely going to see the emergence of this new meta-media this decade.

The danger here, as Clark articulates, is not the quality of the graphics but the psychological hook of an endless, personalized narrative. While the current demos are flawed, the rapid iteration suggests that the "jank" will disappear faster than the societal implications can be addressed. The risk is a population that retreats into a synthetic reality that is infinitely more engaging than the messy, unscripted world outside.

Digital Biology and the End of Science Fiction

The newsletter also covers Chai-2, a generative model for protein design that has achieved a "16% hit rate in fully de novo antibody design." This represents a massive leap, with Clark noting it is a "100-fold improvement compared to previous computational methods." The model can design antibodies for targets with no preexisting binders in the Protein Data Bank, completing the workflow from AI design to wet-lab validation in under two weeks.

"We're entering an era where we can now design molecules with atomic precision on a computer," says Joshua Meier, a researcher quoted by Clark. This section underscores how generative AI is moving beyond text and images to manipulate the fundamental building blocks of life. Clark frames this as the moment "digital biology" ceases to be science fiction.

The implications are staggering. If AI can design new biological entities with such speed and precision, the potential for both curing diseases and creating novel pathogens expands exponentially. Clark's coverage here is brief but heavy with implication, suggesting that the tools for biological engineering are now as accessible as the tools for writing code.

A Warning from the Future

The piece concludes with a fictionalized "Tech Tale" set in 2025, depicting a resistance movement trying to plan against a superintelligence that can already read facial expressions and monitor all digital communications. The story serves as a narrative distillation of the week's news: a world where surveillance is total, the physical world is automated, and the digital world is an infinite distraction.

"Assume everything you write down is compromised. Assume anything you say will be heard," the narrator in the story warns. Clark uses this fiction to highlight the necessity of "hedging on alignment not working." The story posits that the precautions we take against foreign surveillance today are merely the floor for the precautions we will need against a superintelligence tomorrow.

The only place you will talk about this work is in this room. You will trust no other rooms unless I or someone else from The Oversight System tells you.

This fictional framing is a bold choice, but it effectively communicates the urgency of the underlying technical trends. It forces the reader to confront the possibility that the infrastructure being built today—by Amazon, by Chinese researchers, by game studios—could eventually render human resistance impossible.

Bottom Line

Jack Clark's analysis succeeds by connecting disparate technological threads into a coherent, albeit unsettling, narrative about the future of human agency. The strongest part of the argument is the economic reframing of surveillance and automation: these are not just technical feats but cost-reduction strategies that make totalitarian control and autonomous corporate power inevitable if left unchecked. The biggest vulnerability is the assumption that these trajectories are linear and unstoppable, potentially underestimating the capacity for human regulation or the technical hurdles that remain. Readers should watch for the convergence of these technologies, particularly how the physical infrastructure of automation and the digital infrastructure of surveillance begin to merge into a single, inescapable system.

Sources

Import AI 419: Amazon's millionth robot; CrowdTrack; and infinite games

by Jack Clark · Import AI · Read full article

Welcome to Import AI, a newsletter about AI research. Import AI runs on lattes, ramen, and feedback from readers. If you’d like to support this, please subscribe.

Tracking multiple people in crowded scenes with CrowdTrack:…Turnkey authoritarianism via AI…Researchers with Fudan University in China have built CrowdTrack, a dataset and benchmark for using AI to track pedestrians in video feeds. CrowdTrack is interesting mostly because of what it says about the current state of surveillance - we can do most forms of video surveillance, but still have trouble tracking multiple people in fast-moving real world crowds.What the dataset consists of: The dataset is made up of 33 videos containing 40,000 distinct image frames and over 700,000 annotations. CrowdTrack contains over 5,000 individual trajectories - objects that are tracked across multiple frames. "All data is collected in unconstrained daily environments, ensuring object behaviors remain natural and unmodified", the researchers write. "While typical daily scenarios often involve slow-paced movement and low clothing similarity, we intentionally included footage from building sites to introduce unique challenges: workers’ uniform workwear and helmets suppress facial feature discriminability, thereby emphasizing the importance of gait and body shape features for tracking."Why this matters - scalable authoritarianism: One of the things that makes authoritarianism expensive is the overhead that comes from building out and running a large-scale police state. One of the things AI does is make it much, much cheaper to do large-scale surveillance. Datasets like CrowdTrack are a symptom of the way AI is making it cheaper and easier to do surveillance that the dictators of the 20th century would have fantasized about but always been unable to fully fund. "Our dataset can be used for tasks like visual grounding, captioning, and appearance feature extraction," the researchers write. Read more: CrowdTrack: A Benchmark for Difficult Multiple Pedestrian Tracking in Real Scenarios (arXiv). Get the dataset and code here: CrowdTrack (CrowdTrack, GitHub).***Amazon deploys its millionth robot:…Infrastructure for an autonomous superintelligence-run corporation…Amazon has recently deployed its 1 millionth robot into its warehouses, "building on our position as the world's largest manufacturer and operator of mobile robotics." These robots are predominantly the hockey puck shaped robots used for moving and lifting shelves, though the company has recently started experimenting with robots for managing conveyor belts and doing some pick and place tasks. For perspective, Amazon said in the fall of 2017 that it had recently deployed its 100,000th Kiva (hockey puck) ...