Being given a 100% AI generated score for your own mental work hurts. It also hurts because the result is presented as absolute and undeniable, with no path for appeal or inspection. A decision presented as a black-box score with no insights revealed how the score has been computed or the specific reasons/patterns which triggered "detection" to begin with[1]... A score produced by a third party which inserted itself as middleman in the communication flow between writers and readers and claims authority and arbiter status over deciding what is slop and what is not. Another layer of statistical machine learning to detect patterns/anomalies of statistical means of language generation...

In this specific case, I don't really care too much what these tools say. I know what I wrote and rewrote, over hours, and to the best of my ability (as non-native English speaker).

The myth of computational objectivity prevails. For decades it was about blind belief in algorithms, now shifted to AI models. The wider public and most decision makers neither seem to understand the difference, nor do they care about the inherent dangers. Pragmatism at all costs. The genie is out and can't be put back. The problems (if acknowledged at all) are always treated as temporary minor hurdles rather than fundamental systemic issues.

It's upsetting how Little Britain's "Computer Says No"[2] attitude and blind outsourcing of black-box classification and decision-making to machines has already been so extremely normalized and is impacting more and more of our lives, for many people/minorities in far more serious and outright deadly ways... 😭

[1] Having worked with many forms of machine learning myself, I know first hand that lack of detailed observability has been plaguing ML disciplines for decades. I think these are also some of the most worrisome aspects of more widely deploying LLM affine tech. Whilst some minor progress has been made, these efforts are severely challenged and thwarted by the exponentially growing model sizes, architectural complexity and distributed nature of agents.

[2] https://en.wikipedia.org/wiki/Computer_says_no

#AI #AIDetection #DecisionMaking

Am I an LLM? Just out of interest, I checked different variations of a piece of my very own writing I've done this weekend with GPTZero, and each time it says it's 100% AI generated, literally for every single sentence...

Such great motivation on a Monday AM! 😭

#Writing #LLM #AIDetection

This will be me tonight... :) #Darkroom

Reminder: Just as with Age ID and Chat Control legislation justifications, the AI build out, the price gauging of memory, storage (and CPU/GPU/SOCs too), the emerging hardware leasing & subscription models — all this is not just driven by generative AI in the form of chatbots, media generators, automation, job replacements, FOMO stories and other "obvious" reasons repeated ad infinitum by CEOs and in the media...

The only way these industry shifts and largest tech investments ever done make any sense (even in the face of the projected severe negative financial returns) is to consider them part of a larger, class-political project of comprehensively centralizing, monitoring/surveilling, renting out, drip-feeding and otherwise controlling future compute resources and access to them. This includes not only pure computation resources, but equally all communication, data & media produced, processed, stored or consumed via these infrastructures. Personal computing is being pushed out and with it privacy and autonomy over your own compute & media resources are going too...

Governments which have been vapidly talking for decades about "Digitization" (and utterly failing) are jumping on rolling out low quality, "good enough" AI services mostly aimed at the working class, allowing govts to maintain existing austerity programs, whilst claiming to be innovative, cutting costs (by pumping billions into overpriced consultancy contracts) and reducing "red tape" (i.e. mostly regulation which was actually beneficial or protecting the population & environment).

The return on these investments is not financial, but mostly political!

As an aside: I see the biggest threads to the fediverse, as self-owned, self-hosted & self-moderated communication space, are not AI bots, nazis or bigtech-owned platforms, but price explosions of the required (and mostly rented) compute infrastructure. In the near future, these ongoing price increases will force more communities to look for alternatives, which will slowly erode this space[1]... Solidarity and crowdfunding are some ways to combat that, but I fear they will only go that far (i.e. not far enough)...

So even when the current AI bubble bursts, this built-up centralized digital architecture will be a key pillar for any authoritarian government (the general trajectory every country is seemingly on, at current...), run in collaboration with oligarch operators whose purposes are served even more so by said infrastructure. At the very least these data centers will be used for vastly increased state surveillance (machine learning hardware/software is uniquely suited for these use cases), at worst to disenfranchise, prosecute & deny entire demographics...

I'd very much like to stand corrected with this, but there are a no publicly available hints to suggest anything drastically different to happen, not with the current breed of politicians and money in play... At least in the EU, all the talk about Digital Sovereignty already has made it clear that it's mostly about building parallel infrastructure to US big tech, but with the exact same anti-democratic bias, justifications, funding structures and neglect of any climate/social considerations. 30 billion Euro investments to create "AI giga factories" and an "AI continent"[2]. Data centers operated by goody, goody European billionaires rather than evil US based ones... How novel and how sovereign!

[1] Related to that I also see the resource-hungry and extremely chatty nature of the ActivityPub protocol and Mastodon itself as primary implementation as large weak points... In order for the fediverse to survive for longer (for the above stated reasons), I think some drastic reconsiderations of existing design decisions are required to achieve a much lower computational footprint & running costs (e.g. storage requirements, consolidation of messages, reducing traffic & worker effort, esp. important for larger accounts or communities etc.)

[2] https://mastodon.thi.ng/@toxi/117010071572799616

#AI #Politics #PersonalComputing #DigitalSovereignty #LongRead