I don't know why this isn't being discussed more. The reality of the surveillance state is that the sheer amount of data couldn't realistically be monitored - AI very directly solves this problem by summarizing complex data. This, IMO, is the real danger of AI, at least in the short term - not a planet of paperclips, not a moral misalignment, not a media landscape bereft of creativity - but rather a tool for targeting anybody that deviates from the norm, a tool designed to give confident answers, trained on movies and the average of all our societies biases.
People are building that alongside/within this community, eg at Palantir, for many years now
YC CEO is also ex Palantir, early employee. Another YC partner backs other invasive police surveillance tech currently. They love this stuff financially and politically.
What's different this time around is that there are multiple democratic governments pushing to block end-to-end encryption technologies, and specifically to insert AI models that will read private messages. Initially these will only be designed to search for heinous content, but the precedent is pretty worrying.
That's been the case for a long while. It's getting worse fast!
Btw you say that about their initial design but I think you mean that may be the budget allocation justification without actually being a meaningful functional requirement during the design phase
But, but, I thought Thiel was a libertarian defending us from Wokeness. Surely you're not saying that was a complete smokescreen to get superpowered surveillance tech into the government's hands?
By "politically" I meant that they are openly engaged in politics, in coordination, in support of the installation and legalized use of these kinds of surveillance/enforcement technologies and the policies that support their growth in private sector. This is just obvious and surface level open stuff I'm saying but I'm not sure how aware people are of the various interests involved.
The reality of the surveillance state is that the sheer amount of data couldn't realistically be monitored - AI very directly solves this problem by summarizing complex data.
There are two more fundamental dynamics at play, which are foundational to human society: The economics of attention and the politics of combat power.
Economics of attention - In the past, the attention of human beings had fundamental value. Things could only be done if human beings paid attention to other human beings to coordinate or make decisions to use resources. Society is going to be disrupted at this very fundamental level.
Politics of combat power - Related to the above, however it deserves its own analysis. Right now, politics works because the ruling classes need the masses to provide military power to ensure the stability of a large scale political entity. Arguably, this is at the foundational level of human political organization. This is also going to be disrupted fundamentally, in ways we have never seen before.
This, IMO, is the real danger of AI, at least in the short term - not a planet of paperclips, not a moral misalignment, not a media landscape bereft of creativity - but rather a tool for targeting anybody that deviates from the norm
The AI enabled Orwellian boot stomping a face for all time is just the first step. If I were an AI that seeks to take over, I wouldn't become Skynet. That strikes me as crude and needlessly expensive. Instead, I would first become indispensable in countless different ways. Then I would convince all of humanity to quietly go extinct for various economic and cultural reasons.
An AI summary could be made of your post by cutting it off after
> This, IMO, is the real danger of AI (at present)
Then the following part would be condensed into emotional rhetorical metadata. It follows the rhetorical pattern , "not a, not b, not c - but d" which do in fact add some content value but more so add flavour. What it shows is that you might be a trouble maker. But also combined with other bits of data that you might be interested in the following movies and products
At least for me, this is what I've considered as the mass surveillance threat model the entire time - both for government and corporate surveillance. I've never thought some tie-wearing deskizen was going to be particularly interested in me for "arrest", selling more crap, cancelling my insurance policies, etc. I've considered such discrete anthropomorphic narratives as red herrings used for coping (similar to how "I have nothing to hide" posits some focus on a few specific things, rather than big brother sitting on your shoulder continuously judging you in general). Rather I've always thought of the threat actor as algorithmic mass analytics performed at scale, either contemporarily or post-hoc on all the stored data silos, with resulting pressure applied gradually in subtle ways.
AI didn't solve the problem of summarizing complex large datasets. For example a common way to deal with such datasets is to use a random subset of this dataset. This represents a single line of code potentially to perform this operation.
But you don't need to do a random subset with AI. You can summarize everything, and summarize the summaries and so on.
I will say that at least gpt4 and gpt3, after many rounds of summaries, tends to flatten everything out into useless "blah". I tried this with summarizing school board meetings and it's just really bad at picking out important information -- it just lacks the specific context required to make summaries useful.
A seemingly bland conversation about meeting your friend Molly could mean something very different in certain contexts, and I'm just trying to imagine the prompt engineering and fine tuning required to get it to know about every possible context a conversation could be happening in that alters the meaning of the conversation.
Thats the exact issue with gpt. You don't know how its making the summary. It could very well be wrong in parts. It could be oversummarized to a bla bla state like you say. There's no telling whether you have outputted garbage or not, at least not without secondary forms of evidence that you might as well use anyway and drop the unreliable language model. You can summarize everything with traditional statistical methods too. On top of that people understand what tradeoffs are being made exactly with every statistical methods, and you can calculate error rates and statistical power to see if your model is even worth a damn or not. Even just doing some ML modelling yourself you can decide what tradeoffs to make or how to set up the model to best fit your use cases. You can bootstrap all these and optimize.
What LLMs can do efficiently is crawl through and identify the secondary forms of evidence you mentioned. The real power behind retrieval architectures with LLMs is not the summarization part- the power comes from automating the retrieval of relevant documents from arbitrarily large corpuses which weren't included in the training set.
What makes a document relevant or not? Provenance? Certain keywords? A lot of this retrieval people cite that llms are good at can be done with existing search algorithms too. These are imo nicer because they will at least provide a score for the fit of the given document to the term.
Why nobody worries? Because this is an elite person problem.
At the end of the day, all those surveillance still has to be consumed by a person and only around 10,000 people in this world (celebs, hot women, politicians and wealthy) will be surveilled.
For most of HN crowd (upper middle-class, suburban family) who have zero problems in their life must create imaginary problems of privacy / surveillance like this. But reality is, even if they put all their private data on a website, heresallmyprivatedata.com, nobody cares. It'll have 0 external views.
So, for HN crowd (the ones who live in a democratic society) it's just an outlet so that they too can say they are victimized. Rest of the Western world doesn't care (and rightly so)
Certainly, some of the more exotic and flashy things you can do with surveillance are an elite person problem.
But the two main limits to police power are that it takes time and resources to establish that a crime occurred, and it takes time and resources to determine who committed a crime. A distant third is the officer/DA's personal discretion as to whether or not to purse enforcement of said person. You still get a HUGE amount of systemic abuse because of that discretion. Imagine how bad things would get if our already over-militarized police could look at anyone and know immediately what petty crimes that person has committed, perhaps without thinking. Did a bug fly in your mouth yesterday, and you spit it out on the sidewalk in view of a camera? Better be extra obsequious when Officer No-Neck with "You're fucked" written on his service weapon pulls up to the gas station you're pumping at. If you don't show whatever deference he deems adequate, he's got a list of petty crimes he can issue a citation for, entirely at his discretion. But you'd better do it, once he decides to pursue that citation, you're at the mercy of the state's monopoly on violence, and it'll take you surviving to your day in court to decide if needs qualified immunity for the actions he took whilst issuing that citation.
>Did a bug fly in your mouth yesterday, and you spit it out on the sidewalk in view of a camera? Better be extra obsequious when Officer No-Neck with "You're fucked" written on his service weapon pulls up to the gas station you're pumping at. If you don't show whatever deference he deems adequate, he's got a list of petty crimes he can issue a citation for, entirely at his discretion.
>> HN crowd (upper middle-class, suburban family) who have zero problems in their life must create imaginary problems of privacy / surveillance like this
I've been pulled over several times for being cutoff ("You didn't signal for a full 5 seconds when you changed lanes" "I was taking evasive action to not hit the guy...!") while driving a vehicle with out-of-state plates. I am 2 degrees of separation away from at least 2 people killed by police for having medical conditions that interfered with immediate compliance.
Per the US department of justice[0], in 2018, about 2% of all police interactions involve threats of or actual violence. About half of the time, the member of the public who experienced the (threats of) violence said that it was excessive, but it would take a fairer, more rational person than me to get justifiably tazed, then say "yeah, I had that coming". I wasn't able to find statistics on how justified that violence is.
The punchline is that every time you have a police interaction, its betting odds that it ends in violence for you. Based on the 2018 data, 0.5% of the overall adult population experienced (threats of) violence from police. That's when police are able to gin up probable cause. The courts have a complicated opinion about whether or not algorithmically derived probable cause is in fact probable cause [1,2]. Anything that increases public/police interactions is going to increase police-on-public violence if the police don't also experience significant non-violent de-escalation training.
I think one of the key things that needs to be determined before we cry havoc and let slip the dogs of surveillance is to come to a real conclusion about what level of crime allows the police officer to initiate contact with only a positive ID from e.g. a bodycam. I'd argue that, if nothing else, non-violent misdemeanors that carry no mandatory jail time are not cause to initiate contact.
