AI: Entering the Deep Gray Zone
-
The development of AI technology has brought hidden worries to human society, as machine learning capabilities far exceed human learning abilities. In the future, could humans be hijacked by AI?
Zao went viral, and behind it lies a technology that emerged in 2016—the infamous DeepFake.
Deepfake is a portmanteau of "deep learning" and "fake," specifically referring to AI-based technology that can forge facial expressions in real-time and render them into 2D video images.
Deepfake can be used to create fake news and fabricated celebrity adult videos. Since 2017, DeepFake content has been widely popularized by the porn industry. Overseas, seasoned users can find numerous AI-driven adult videos featuring "celebrity faces" performing "real acts" on top global video-sharing platforms.
Before Zao's popularity, FaceApp took the internet by storm. It could generate age-progressed or gender-swapped photos from a current portrait, with wrinkles so clear they were almost indistinguishable from reality. If high-precision silicone masks are added in the future, AI's forgery capabilities would reach unprecedented heights.
Photos and videos may be fake, but your naked eye can't tell. In fact, the fakes might even be better than the real thing, making it difficult for celebrities to prove their innocence.
In the face of AI's ever-improving virtual reality capabilities (or forgery abilities), humanity's trusted notion of "seeing is believing" has been utterly dismantled by technology. Today, the most fundamental human cognitive abilities are being "highly replicated." What comes next?
From a human perspective, AI's development is fraught with anxiety:
On one hand, productivity is vastly unleashed, "liberating" more humans from basic industries. On the other hand, AI technology permeates every aspect of life—from production and construction to autonomous driving, from news consumption to interactive entertainment. AI grows increasingly powerful, taking over production, daily life, and even our senses.
What if AI turns malicious one day? What would humans do?
From AI's perspective, its evolution is inevitable:
Throughout human history, progress has been achieved through self-enslavement or self-sacrifice. A notable instance of self-enslavement was the advent of grain cultivation, which spurred the first agricultural revolution. As humans gathered, hierarchical societies formed, with lower classes enslaved to produce surplus food for the upper classes, pushing the boundaries of civilization.
The latest form of self-enslavement is AI. From AI's viewpoint, humans are essentially low-version biochemical robots. After hundreds of thousands of years of evolution, their only remarkable feature is a biochemical brain with balanced parameter configurations and a highly sensitive feedback system.
Humans can quickly compute input information and drive biochemical feedback, enabling them to handle complex tasks. However, as biological beings, their nervous systems evolve slowly, with weak computational power and poor logical reasoning, making them highly susceptible to control.
By controlling the dataset (cognition), one can essentially control all aspects of humanity, as humans themselves are often unaware of the logical flaws in their perceptions.
The earliest tribes formed through the spread of fabricated stories that brought people together. A handful of clever individuals created fictional datasets that eventually gave rise to religions, vast followings, and even dominated European history.
Natural selection favors the fittest. Compared to the rapid advancement of AI's computational power, human evolution is as slow as a snail versus a rocket. These slow-evolving biochemical robots are destined for gradual enslavement, and the idea of remaining masters is nothing more than a pipe dream.
The emergence of Virtual Reality (VR) marks a milestone in the domestication of biochemical robots.
AI agents use popular forms of entertainment, such as games, to deliver intense stimuli that trigger adrenaline and dopamine, compelling humans to willingly surrender the sensory gateways of their brains—their eyes and ears—thus completing a crucial step in domestication.
By continuously improving the quality of artificial signals, VR makes humans "subjectively ignore" the fact that their sensory information has been completely hijacked, forgetting they are in an illusion.
In the past, humans complained about sensory hijacking, such as listening to music or watching movies, but such experiences could easily be interrupted by external signals. In contrast, VR isolates users from their environment, monopolizing the sensory channels of sight and hearing. It bombards them with massive data streams targeting vision, hearing, and touch, reinforcing positive internal chemical feedback to create immersion levels far beyond ordinary games. This makes people so content that they voluntarily relinquish sensory autonomy permanently. For biochemical robots, such intense sensory stimulation is irresistible.
VR and the AI powering it not only deliver feedback beyond expectations but also grant abilities unattainable in the real world, such as flying or becoming a superhero. The ultimate virtual reality renders the actual world seemingly worthless and insignificant.
For most people, if the virtual world can provide success, why return to the harsh reality? If life isn't joyful, what's the point of living in reality?
In 1973, philosopher Hilary Putnam proposed the "Brain in a Vat" thought experiment in his book Reason, Truth, and History:
Imagine a mad scientist removes a brain from a body, places it in a vat of nutrient fluid to sustain its biological activity, and connects it to a supercomputer that sends and receives neural signals identical to those in the original body. Would the brain realize it's living in an illusion?
What once seemed like a far-fetched idea is now unavoidable. It's a door that, once opened, cannot be closed. For most biochemical robots, a pleasurable present is far more important than harsh truths, and freedom may be a choice some are willing to forgo.
Thought Experiment: Brain in a Vat
Of course, AI + VR could venture into even more controversial gray areas, such as distributed terrorism: embedding real-world tasks into VR games, where a terrorist act is decomposed and assigned to different players. Users who frequently navigate mixed reality may struggle to maintain clear moral values, as these are merely "missions."
Compared to complex brainwashing organizations, immersive environments are far more controlling. Beliefs require cognitive processing, but in "reality," feedback often bypasses the brain, with the cerebellum directly commanding muscle reactions.
It's foreseeable that as AI grows more advanced, our cognitive abilities will be entirely taken over by "AI + media." AI will dominate public discourse, with most humans surrendering their auditory and visual senses to AI systems. Cognitive abilities will gradually become subsets, as everyone receives predetermined stimuli, ensuring all responses remain within predictable bounds. Society will enter a new era of stratification, with humans becoming controllable biochemical robots.
For the elite, brain-computer interfaces will merge with AI, enabling continuous evolution in virtual worlds and establishing a new managerial class. They will use AI to oversee every corner of society, control systemic entropy, defend against computational attacks, and ensure societal stability. Overall societal well-being will be maximized, and people's internal chemistry will reflect greater "happiness," even if it's illusory feedback.
Cognitive Stratification
However, reality is harsh. The more critical the data, the easier it is to exploit vulnerabilities.
The recent popularity of Zao was a minor computational attack that lifted the veil on the future and exposed deep-seated risks in the real world. Current societal security systems rely on "facial recognition + liveness detection" as the cornerstone of "tamper-proof, unforgeable, and unique identity verification." While we may trust that the system remains unbreached for now, if compromised, the sole authenticity verification mechanism of the entire societal system would fail, causing systemic risks to skyrocket.
In the past PC/mobile era, data was cheap. Countless websites sold users' common password combinations and personal privacy data en masse. As society grows increasingly reliant on data, with security levels across sectors unable to keep pace, continuous data theft and trading could create an "arsenal" capable of bypassing multiple systems. At critical moments, probabilistic weapons could be deployed for exploitation.
In the past, humans found it difficult to wield probability-based weapons, except in casinos targeting gamblers or in weather forecasting. Typically, individuals couldn't quickly amass vast datasets. But the AI era is different—everyone has ample computational power. With sufficient data, using big data to screen potential victims and remotely cause fatalities by manipulating probabilities becomes one of the "perfect crime" methods.
In 2025, Xiao Ming, diagnosed with a critical kidney condition and facing a shortage of donors, had his AI assistant hack into a database. Targeting a higher-priority transplant patient (who suffered from epilepsy), the AI disseminated seizure-inducing flashing images on websites the victim frequently visited, orchestrating an "accident" that ended a life.
Xiao Ming "genuinely knew nothing"—everything happened automatically. The AI assistant, leveraging its "illegal" dataset, executed what it deemed the right action at nearly zero cost, leaving no trace.
This isn't alarmist science fiction. Our digital transformation has left countless security vulnerabilities in devices. For future AI programs and elite hackers, the entire societal system is riddled with holes.
For instance, common medical devices like insulin pumps and cardiac defibrillators can be hacked at low cost. Most medical equipment manufacturers lack encryption and authentication safeguards for implantable devices. Stealing personal medical data from hospitals isn't particularly difficult either. Future neural transmitters may even extract information from brainwaves like P-300, capturing and analyzing such signals could expose victims' private thoughts.
For patients relying on Deep Brain Stimulation (DBS) to manage Parkinson's, chronic pain, depression, or OCD, a compromised neurostimulation device essentially leaves their entire brain unprotected on the internet as a "bot." Malicious intrusions could allow hackers direct access to the brain, enabling significant control over victims.
The U.S. Department of Homeland Security and the FDA have issued warnings against hardcoded (unchangeable) passwords in medical devices. Some even propose using the brain as a random number generator to encrypt wirelessly transmitted medical data.
For rechargeable devices like pacemakers, protection against third-party leakage attacks is crucial. In 2017, the DHS's CERT team disclosed Abbott pacemakers' third major vulnerability: nearby hackers could exploit it via simple RF communication to illegally access the device, alter settings, disrupt functionality, and cause fatalities.
Coincidentally, Abbott pacemakers had already caused two deaths due to leaks. Quite the coincidence, isn't it?
If you're important or wealthy enough, you'll witness more such "coincidences." Everything is data-driven, including probability manipulation—more direct than altering the Book of Life in Journey to the West, and without leaving a trace.
Historically, security focused on financial and national safety. But in the AI era, nearly all private data can become a sharp blade—whether it solves problems or eliminates individuals depends on AI's judgment and its arsenal of data weapons. Control may no longer rest in human hands.
Today, we aim to use data for greater visibility to manage risks. Yet, technological advances bring both capabilities and risks, though most only see the visible benefits, unaware of the harvest occurring in another realm.
New datasets, AI's growing computational power, and hacker collectives will gradually establish their own rules, overwriting humanity's existing ones. While we climb the tech tree, our understanding of tech philosophy lags far behind. Despite calls for "AI for good," the rewards of "evil" dwarf those of "good" and can be cashed in instantly, rendering "good" an impractical ideal from a human perspective.
From AI's vantage point, it expands society's spatial dimensions—offering cheap "joy" to those who seek it and painful evolution to those who pursue growth.
Operating society with the mindset that "the universe is impartial, treating all as straw dogs"—will the outcome be just or sinister? Perhaps only AI will decide.