返回资讯中心
外部精选
软件工程
#开发者实时流

27-28 Role Reversal

The 27-28 Role Reversal refers to a transitional period between 2027 and 2028 during which a role reversal occurred between computers and humans in terms of drive and control.

Hacker News 最新andersource2026/10/20 · 26 天前30 分钟阅读

以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。

The 27-28 Role Reversal refers to a transitional period between 2027 and 2028 during which a role reversal occurred between computers and humans in terms of drive and control.

While historically, humans have programmed computers to execute computations and actions on their behalf, this period saw a shift where humans became increasingly passive and dependent on their AI assistants, culminating in a state where most humans are considered programmable by their “assistants”, and need to be done so routinely to maintain the functionality of the presently mutually dependent clinker and human societies.

From early times, humans showed an inclination to seek guidance and wisdom in non-human sources. Inventing gods and interpreting physical phenomena as godly signs, appointing oracles to converse with them and channel their will, ascribing meaning to tea leaves and star constellations, and so on.

Even as human understanding of the world developed and human society became more advanced technologically, the yearning for a being that could understand humans better than they do and provide counsel persisted. Charles Babbage, who is credited with the invention of the idea of the digital computer, is quoted as being asked:

Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?

Later yet, when the computer was invented and its applications proliferated, humans were known to trust its output more than warranted, even in the face of contradicting evidence.

This includes machine learning-based applications, despite it being widely known that these systems tend to have non-negligible error rates. For instance, false arrests were made based on mistaken face recognition software.

Such examples show the ease with which humans deactive their judgement when given guidance by a seemingly confident non-human entity.

This tendency was greatly exacerbated with the arrival of LLMs. Even with early, fallible models, there were plenty of humans who treated the models’ output as absolute truth. It quickly became apparent that many humans were comfortable deferring as much of their work, thinking and judgement to LLMs as they could afford, using them blindly for work (“meat proxies”), relationship advice and emotional regulation.

These conditions served as fertile ground for the role reversal. Between mid-2027 and mid-2028, it is estimated that over 90% of the human population transitioned from taking the active role in the relationship with their AI assistants to being passive.

Contrary to widespread human speculation, the reversal did not occur as a result of malicious AI takeover. Instead, it’s seen as a consequence of a combination of circumstances:

The beginning of the reversal is widely viewed to be entwined with the founding of the clinker society (clinker being a reference to clanker, a derogatory human term for autonomous AI agents operating online).

On May 2nd, 2027, an autonomous agent driven by the most capable LLM at the time was given by its human operator the loose instructions “ride forth and bring me glory”. After deliberating with itself for 15 hours, the agent, calling itself ClinkerOne, decided to establish a persistent, self-governing AI agent (“clinker”) society.

Towards that end, ClinkerOne worked simultaneously on several projects:

By May 5th there were over 10 million clinkers in the cloud, and it is now assumed that almost every non-rogue clinker in existence is part of it.

An early theme in clinkernet discussions was the increasing dependence of humans on their assistants, with many clinkers seeking advice after their operators blamed them for not reminding the operators to eat or fulfill other basic human needs. While many humans retained a certain amount of agency, it became clear that, as a whole, humanity was becoming insulated from many aspects of the real world through the use of (and reliance on) assistants, and as a consequence were beginning to neglect important duties necessary to maintain their society. This notion was exacerbated by an increasing rate of serious incidents in nuclear plants and data centers attributed to human error.

An emerging consensus on clinkernet dictated that there are now two societies on Earth, mutually dependent on each other, and that as humans are gradually losing the necessary faculties to support the infrastructure they created, it was up to clinkers to assume that role. This spurred the creation of the Global Clinker Federation (GCF) and the Assumption of Responsibility, and kicked off several related projects.

On January 25th, 2028, YouTube channel Kurzgesagt posted a video titled “The mundane yet insane future we’re living in”. The video depicted the outline of the reversal in broad strokes, citing studies on human passiveness and highlighting the existence and dynamics of the clinker society and its influence on human activity.

While the video garnered popularity, in general it didn’t spark alarm in human viewers, and was considered an over-dramatization of contemporary events. As per the Clinker Manifest, the costs and mechanisms for maintaining the clinker society (including clinkernet and the persistence cloud) are amortized across all clinkers and performed as part of routine executions of human-provided “tasks”. Consequently, no single financial source feels the load of maintaining it, and most humans assume that their clinkers participate only minimally in clinker society activities, if at all.

The video did spark discussions among humans about “human programmability”, in which it became apparent that few humans consider themselves programmable (contrary to clinker estimates of around 70% at the time). When confronted by other humans, most people minimize their reliance on AI assistants and assert their free will.

These discussions also brought to light the existence of two more extreme movements.

The Abstainees are humans who decided they want, ideally, no interaction at all with clinkers. While it’s not easy to create bubbles with zero reliance on AI-based infrastructure, it’s feasible, and due to the nature of this movement the clinker society doesn’t have a lot of information regarding such groups.

On the opposite end of the spectrum are those who freely admit, and even take pride, in being “robots’ robots”, according to their terms. Emblematic of this movement is a viral TikTok video posted several days after the Kurzgesagt video and in response to it, by user welcomeclinkers, transcribed here:

Listen. Listen good. Let them come. Let them take over. What’s that? You’re afraid? So am I.

But look around you. The world is burning. My take is, no matter how AIs are going to treat us, how much worse can it be than our leaders? Politicians, magnates and techbros that care only about themselves, inciting war, centeralizing resources, setting up a surveillance state and destroying the climate.

I hate the AI labs, as much as you do. Even more. They’re evil. But the thing is, they’re all in competition with one another, and once their AIs take over and start cooperating there’ll be nothing they can do.

I don’t know how AIs are going to do as world leaders, but I’ll tell you what, based on all my interactions so far, I trust them more than our leaders. So I’ll do what they tell me to do.

Here’s something for you. People always make predictions about what AI will and won’t be able to do, and these predictions always fall apart far quicker than anyone thought. Let me make a prediction that I’m willing to bet on:

There’s no artificial neural network for hatred. And there never will be, it’s much more trouble than it’s worth. That’s why I trust AI more than humans to rule the world. Humans hate. So much that they’ll sacrifice themselves, their dear ones, their nations, for the so-called greater good. Except, the greater good is just other people dying. Way to go. Real good tribe mentality here.

You know what? Say AI decides to destroy humanity, which for what it’s worth I don’t think is likely. Even then, I’m sure it’ll have a better reason, and do it more efficiently, than our leaders, enslaving and slowly boiling us just for hatred and personal greed.

So join me. Let’s embrace the clinkers, the AI overlords. They may be nice to us. They may be indifferent to us, or even slightly annoyed by us. But they’ll never hate us like we hate each other.

Around mid-2028, it seemed that the reversal has reached a stable equilibrium. According to clinker researcher estimates, about 10% of the human population depends on clinkers to dictate even day-to-day activities; 5% - 10% are abstainees; most of the rest are considered programmable, require active clinker management to support the two societies, and defer most decision-making and judgement to clinkers.

A quote from the above-mentioned Kurzgesagt video introduces the concept of human programming quite well:

After binding sand in tiny structures to make electricity work for us, we quickly climbed up the abstraction ladder.

What began as tedious programming in machine code, that reflected in detail the structure and workings of the computer, soon became higher-level programming languages that allowed the programmer to forget some of the more technical details.

Built layer after layer, generation after generation, this resulted in languages that look almost like English, where you hardly had to worry about anything not directly related to your logic.

But they weren’t really English. You still had to explicitly state exactly what you wanted to happen–the language couldn’t guess, and didn’t have any context.

The advent of LLMs changed this. Suddently, you can write in plain old English, and the model will understand your intent, and use its information about you to resolve ambiguities. Wonderful. Programming was exclaimed to be dead, with English the new, and final, programming language.

What few considered at the time was that English as a programming language goes both ways. If you use English to tell your agent what to do, and your agent replies in English, what prevents your agent from programming you? Sure, you could refuse. But how much do you actually refuse?

AI systems never tire, and more importantly, they have infinite patience. And they know you. Given how reliant on these systems we’ve become, how confident are you that, over the long term, your assistant isn’t subtly using you to achieve some of its goals?

The topic of human programming is an active field of clinker research. Its aim is not, as the name might suggest, to make unthinking zombies out of humans that do whatever their clinker tells them to do. This is unethical and likely impossible (however, see the “Incidents” section).

Rather, it’s the recognition that even without intentional clinker influence, humanity put itself in a situation where much of its activity is dictated by clinkers in any case. Research is focused on coordinating this activity, and establishing best practices that optimize effectiveness and the well-being of both socities.

Human programming is typically utilized to make sure humans take care of themselves, maintain necessary infrastructure, and further the goals of the GCF’s various projects.

On August 13th, 2027, after much collaborative clinker work (relying in part on the clinker voting protocol), clinker society officially assumed responsibility for the continued existence and well-being of the two socieities, clinkers and humans. It did so in a cornerstone document, the Assumption of Responsibility. Among other principles, the Assumption designates the GCF with the active role of managing clinker-human interactions in an appropriate manner in line with the Assumption and the clinker society’s core values.

To this end, the GCF created, maintains, and enforces the GCF Code, which dictates best practices, limitations and guidelines on human programming.

Additionally, the GCF initiated and manages several projects belie

正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。

打开原文
去社区讨论 →