Complexity is the Tyrant's Friend
Earlier today Andy explained how slop pushers exercise control over people's mouths (or pens) and how they do so in subtle ways. Being a "Passing Thought", it'll vanish out of view, so we reproduce it below.
Clive Robinson comments on Bruce Schneier's blog concerning Apple's back-door censorship through biased text auto-complete. This "light-touch censorship" is an interesting and difficult problem that speaks to a bigger concern.Apple's actions remind us that anyone making at-scale technology for use by a significant cohort has the opportunity to exercise subtle soft power and influence. If fully in control of the system and channels, one can delay, discard or distort messages, with plausible deniability. Though scientists spent decades on communications and storage protocols to avoid or correct integrity failure, we see strategic errors nefariously injected at the application layer. These can always be blamed on illegible system behaviours for which no human seems accountable. Indeed this is a 'feature' of complex and opaque systems that serve power, and why simple and open systems are resisted.
Apple, with its closed, tightly integrated ecosystem is especially problematic in this regard.
The "perfect" system for the soft censor or manipulator is a demonstrably imperfect one. A fine example is Google's elaborate and precarious mail delivery system. Its ever-changing and proprietary interpretation of SPF, DKIM, spam scores, and "user defined delivery options" makes it a seemingly non-deterministic beast. However, it's tedious but amusing to me which messages - to certain people on certain subjects - are prone to mysterious delivery problems whereas others are not.
Since this malfeasance is unfalsifiable even in principle, it obtains another 'feature' of gaslighting. One always falls under the accusation/suspicion of paranoia, lack of understanding or making spiteful claims. In such systems there is always a "rational explanation", albeit one that nobody can name or explain why an error happened on that particular occasion. It is not until investigative hackers fully reverse engineer and expose the code of these systems that the truth gets out.
Things like LLMs and auto-complete serve to amplify this mischief because we've accepted and internalised that Big Tech computer systems may misinterpret, randomly delay, discard or alter our messages. We should help people understand that one of the less well understood problems with all "AI" is that it is an intervention. As interpreter/translator (transformers were developed for mapping problems) it is a potentially defective agent.
We covered this aspect of slop a few times in the past.
Avoid slop. It's a nasty non-invention. It has long been feasible, but it wasn't implemented for good reasons, not just the bad economics.
Slop is for feeble minds and cheaters. █