Friday, December 20, 2024

Inside the World of the LSD User


The spread of misinformation varies by topic and by country in Europe, study finds
May 2024, phys.org

Researchers analyzed news activity on Twitter (now X) in France, Germany, Italy and the UK from 2019 to 2021, including a focus on news about Brexit, the coronavirus, and the COVID vaccines. Each news source they analyzed was rated as either "reliable" or "questionable" based upon their NewsGuard score.
  • Across all four countries, the vast majority of users only ever consumed reliable news sources on each of the three topics.
  • But in every country and in each topic, there was always a small percentage of users who only ever consumed questionable news sources - with very few people consuming a mix of both reliable and questionable sources.
  • Germany had the highest ratio of questionable news retweets to reliable news retweets on all three topics, with France in second, followed by Italy, and the UK had the lowest proportion of questionable news retweets overall.
  • Italy had the lowest proportion of questionable news retweets for the topic of the coronavirus - but had the highest percentage of people consuming only questionable news sources on Brexit.

"Cultural nuances" will be important when it comes to fighting misinformation.

via Ca' Foscari University of Venice: News and misinformation consumption: A temporal comparison across European countries, PLoS ONE (2024). DOI: 10.1371/journal.pone.0302473

Image credit: 3D printed vacuum system traps dark matter - University of Nottingham - Jun 2024


Study finds avoiding social media before an election has little to no effect on people's political views
May 2024, phys.org

35,000 Facebook and Instagram users who were paid to stay off the platforms in the run-up to Election Day.

I'm thinking about the overall social influence, like if you're not online but your friends are, it's all the same. So if they did this study by taking only the friends of the people chosen, then we would see an effect? (Friendship Paradox)

via Stanford: Gentzkow, Matthew, The effects of Facebook and Instagram on the 2020 election: A deactivation experiment, Proceedings of the National Academy of Sciences (2024). DOI: 10.1073/pnas.2321584121.


Mental disorders may spread in young people's social networks
May 2024, phys.org

Contagion - 700,000 ninth-grade pupils born between 1985 and 1997 from 860 Finnish schools followed for 11 years (largest and most comprehensive so far on the spread of mental disorders in social networks).

The number of classmates diagnosed with a mental disorder was associated with a higher risk of receiving a mental disorder diagnosis later in life.

The connection observed in the study is not necessarily causal. "It may be possible, for instance, that the threshold for seeking help for mental health issues is lowered when there are one or more people in your social network who have already sought help for their problems" (beneficial contagion).

via University of Helsinki: Jussi Alho et al, Transmission of Mental Disorders in Adolescent Peer Networks, JAMA Psychiatry (2024). DOI: 10.1001/jamapsychiatry.2024.1126

Dual-species Rydberg array combining rubidium and cesium atoms to enhance quantum computing - BernienLab at University of Chicago - Oct 2024 [link]

Study suggests less conformity leads to more innovation
May 2024, phys.org

These methods reveal something crazy - how easy it could be to manipulate a big part of a population just by doing some very small adjustments to the recommendation algorithms, and the idea of synthetic networks cycling through the real networks as a kind of social regulator.

  • The Matthew effect aka "rich-get-richer effect" increases centralization in the network, which destroys the niches protecting minority opinions, reducing sociodiversity.
  • "Networks that promote sociodiversity have structural features that protect minority opinions"
  • "Unfollowing a few VIPs can help to promote sociodiversity"

via Complexity Science Hub and ETH Zurich: Andrea Musso et al, How networks shape diversity for better or worse, Royal Society Open Science (2024). DOI: 10.1098/rsos.230505


Misleading COVID-19 headlines from mainstream sources did more harm on Facebook than fake news
May 2024, phys.org

They showed thousands of survey participants the headlines from 130 vaccine-related stories - including both mainstream content and known misinformation - and tested how those headlines impacted their intentions to get vaccinated against COVID-19. Researchers also asked a separate group of respondents to rate the headlines across various attributes, including plausibility and political leaning. They matched that against 13,206 vaccine-related URLs and the number of users who viewed each.
  • Misleading content from mainstream news sources - rather than outright misinformation or "fake news" - was the primary driver of vaccine hesitancy on Facebook.
  • "Vaccine-skeptical" content was potentially misleading but not flagged as misinformation by Facebook fact-checkers.
  • One factor reliably predicted impacts on vaccination intentions: the extent to which a headline suggested that the vaccine was harmful to a person's health.
  • The most-viewed was an article was from a well-regarded mainstream news source, and its clickbait" headline was highly suggestive and implied that the vaccine was likely responsible.
  • (That's significant since the vast majority of viewers on social media likely never click out to read past the headline.)
  • Contrary to popular perceptions, the researchers estimated that vaccine-skeptical content reduced vaccination intentions 46 times more than misinformation flagged by fact-checkers.
  • Even though flagged misinformation was more harmful when seen, it had relatively low reach.
  • Gray-area content is less harmful per exposure but is seen far more often, thus it's more impactful overall.
(Also, what does this mean about actual advertising metrics?)

via MIT Sloan School of Management: Jennifer Allen, Quantifying the impact of misinformation and vaccine-skeptical content on Facebook, Science (2024). DOI: 10.1126/science.adk3451.


Advertisers may be inadvertently funding misinformation
Jun 2024, phys.org

(I think we're starting to see a pattern here.)

5,000 websites (1,250 were misinformation websites); 42,000 unique advertisers; nine million instances of advertising companies appearing on news websites from 2019 to 2021. 

The companies advertising on misinformation websites accounted for anywhere from 46% to 82% of overall companies in their respective industries.

Companies that used digital advertisement platforms were 10 times more likely to appear on misinformation websites than those that did not.

Authors propose two low-cost, scalable interventions: 
  • Improve transparency for advertisers about where their ads appear 
  • Make it easier for consumers to identify which companies advertise on misinformation outlets through information disclosures and company rankings

via Carnegie Mellon University's Heinz College and Stanford: Wajeeha Ahmad et al, Companies inadvertently fund online misinformation despite consumer backlash, Nature (2024). DOI: 10.1038/s41586-024-07404-1

Entangling pairs of photons using 1.2 micrometer niobium oxide dichloride - Leevi Kallioniemi at Nanyang Technological University in Singapore - Oct 2024 [link]

Study suggests ambivalence and polarized views can promote political violence
Jun 2024, phys.org

Time to get confused! 

Previous research has shown that ambivalence about a political issue typically leads people to avoid taking moderate actions such as voting or donating money in support of that issue.

And that's why these findings for extreme behaviors surprised the researchers.

Here, researchers found that ambivalence can actually lead some people - especially those with polarized views - to be more supportive of extreme actions.

The reason? Ambivalence creates discomfort in those with extreme views by making them feel weak or insecure about their beliefs - and that can lead them to compensate.

"When people have these polarized views on a political topic, but also feel some conflict about that belief - that is really a potent combination."

"Those who are conflicted are more willing to give their money to extreme organizations."

"Those who have extreme views, but still feel conflicted about them, feel a need to prove themselves, to show that their beliefs are real and strong."

via Ohio State University: Joseph Siev, Ambivalent attitudes promote support for extreme political actions, Science Advances (2024). DOI: 10.1126/sciadv.adn2965.

Copyright Mishaps, Copybots and Word Traps


In the news, we see the use of an old clever trick to catch lying cheating thieves - like hiding an air tag in your purse, writers and all kinds of content creators are hiding novel nonsense in their work. Mapmakers did this by using fake street names - nobody would notice it expect the person who put it there. In the early days of data collection, when a list of subscribers to American Baby magazine was worth its weight in gold, the list owners would secretly add employees from their own company to the list, with their own home addresses; so if an employee got solicited by any company that didn't pay for the list, they would know who stole their data. Surely there are examples that reach back further into history, but for now:

“Copyright traps” could tell writers if an AI has scraped their work
Jul 2024, MIT Tech Review

To create the traps, the team used a word generator to create thousands of synthetic sentences full of gibberish, which are then injected into text in multiple ways—for example, as white text on a white background, or embedded in the article’s source code. This sentence had to be repeated in the text 100 to 1,000 times. 

But, not so fast:
“A lot of companies do deduplication, [meaning] they clean up the data, and a bunch of this kind of stuff will probably get thrown out”

via Imperial College London and presented at the 2024 International Conference on Machine Learning


AI Art - Detailed Mechanical Parts Brimming with Energy 2 - 2024

The above news item is only an introduction to the result of a very long experiment started almost 15 years ago - this weblog.

At some point it became clear that Network Address could be used to run simple experiments on search engines. It was becoming full of text that didn't exist anywhere else on the internet, so it was a kind of control group. Every so often, I would run a search for some weird phrase or word I had used, to see if it "existed" yet. 

There was a point in the development of the internet that if you ran a search and there was only one single result, and that result was your weblog, then you own it! A word is born, and it's all yours. Fast forward to today when, well, it appears that the Great God of Search Engines was letting their product deteriorate since about 2020, once they decided that they owned so much of the internet already that they could do whatever the hell they want and still maintain their position as the best search engine in the world. 

One uneventful day in 2021, I felt a tingling on my arm, kind of like if you've ever felt a sprinkle of water on your skin, but it's not really there. A "thermodynamic hallucination" I thought to myself. Is there such a word? A name for this phenomenon? I ran a search, and to my surprise, that rare moment - no results. A nonsense phrase that can now belong to me. Fully aware of my duty, I made a post about it, for no other reason than to get something out there that contained this phrase, and where the body of the post said something like, "This phrase didn't exist in your search results until just now". Wait a week for the cache to catch up, run a search again, and would you look at that, one search result, the only search result. One more piece of real estate on the dematerialized surface of the noosphere, add that to your portfolio.

Fast forward to 2024, the power of parallel computer processors and yottaboxes of data put artificial text generators in full Turing mode. We're in awe of the outputs; so human! We're  enmeshed in the scandals; how could they steal my work? All I'm thinking is, I have an entire blog's worth of copyright traps to feed this thing. 

And so one early summer night in July of 2024, I ask a generative agent: "What about thermodynamic hallucinations, anything on the web about it?"

The term "thermodynamic hallucinations" appears to be a niche concept without widespread recognition in mainstream scientific literature. However, there are related discussions in the context of stochastic thermodynamics and information theory, where complex interactions and interpretations of thermodynamic systems might be loosely termed as such. For instance, a study published in 2018 explored the connections between stochastic thermodynamics and information ... abstract nature (APS link) [the paper linked here is basically totally unrelated].

Additionally, the idea has been mentioned in more speculative or interdisciplinary contexts. [Plural?] A blog  discusses "thermodynamic hallucniations" in relation to AI, memetics, and sociothermodynamics, reflecting more on the cultural and philosophical implications rather than strict scientific phemonena (Network Address)

Thus, while "thermodynamic hallucinations" is not a standard scientific term, it is used in various [various?] speculative or metaphorical discussions related to the intersection of thermodynamics with information theory and complex systems. -ChatGPT 
Or how about this, trying a little harder: "I thought the blog was referring to thermodynamic hallucinations as a new term?"

The blog that references "thermodynamic hallucinations" does indeed seem to present it as a novel term. [Good] The blog discusses this concept in relation to a variety of topics including artificial intelligence, human futures, memetics, and complex systems. [Definitely not] The term appears to be used metaphorically to describe complex, emergent behaviors or phenomena that arise from the interplay of thermal and information dynamics in advanced computational or sociocultural contexts (Network Address). [It was literally used as a copyright trap, although I didn't know it at the time; I thought I was making a kind of placeholder.]

This indicates that "thermodynamic hallucinations" might be speculative or interdisciplinary concept proposed to explore how principles of thermodynamics could influence or explain certain emergent behviors in AI, cultural evolution, or network science. The term is not widely recognized in mainstream scientific literature, but is seems to serve as a thought-provoking metaphor within certain theoretical and experimental frameworks. For more detailed information, you can visit the blog directly: ...

There's something interesting about the preoccupation with the word "stochastic" and "hallucinations", since these two words, minus the "thermodynamics" have a high co-occurance with AI generative processes in general, which has at this point consumed the entire internet, and might be over-represented in the training data. 

But more importantly, where did all the talk come from about memetics, culutural evolution and network science? Blogs use keywords, or tags; they're bits of metadata added to the post to help categorize it. But the blog itself lists all the keywords used on all the posts on the side bar. Because I don't take the tags seriously, and use them as an extra layer of confusion or expression or whatever you want to call it, there is an accumulation of many hundreds of these tags listed on the side of every post, words like: network science, mind control, entropy, semiotics, Big Data, Big Datty, meatbody, topology, algomania, algohaulic, algorithmic crack, and bio-abacus. Hundreds. 

I don't know exactly how these large language models work, but to simplify, they ingest text and output amalgamated approximations of that text, based on the prompt. This prompt sent the agent to a webpage that had all these crazy tags running down the side. They aren't in the body of the text of the post, but meta-data listed on the side. Little did I know they'd give a passable summary of the blog itself, and this is what came out, above, a very convincing hallucination - that is, if you had no idea of the truth. 

Post Script:
Today, in a final observation, I run one more search for "thermodynamic hallucinations" using Google, then Bing, then DuckDuckGo (Bing). Network Address doesn't show up at all (interesting*), but instead we get 1 result, in all three engines - a substack post by a most likely real person in May 2024 with a section heading titled Thermodynamic Hallucinations, which is about the "temperature" weights on large language models and the "hallucinations" they output; there's a later heading titled Social Hallucinations. And here we see the moment has passed for this experiment, the phrase is now tainted, it's been touched by the real world. Also, the internet as we know it is now broken. 

Post Post Script:
There are actually three results but from the same domain:

Thursday, December 19, 2024

Clash of the Titans Can't Stop Won't Stop

 
Every system wants one and only one system administrator. The body wants a brain. The brain wants a self. The people want a king. The market wants a monopoly. But that's not sustainable for a system. Eventually, as the agents, or actors, or nodes in network are reduced to a handful, and then to a few, there comes a time when there can be only one. 

Now is that time. (See previous post for more of this.)

Let's start with this, a bit tangential, but putting the point up front, in practical terms:

ANCHOR website isn’t broken, state says. Error message is a security measure
Sep 2024, NJ Advance Media 

This an example of what happens when you offload government services to private enterprise, at a time when all the enterprises get along and follow the same rules, until they don't because they get so big they start eating each other, and make the whole infrastructure of the internet not work, and government is screwed, because that's what's happening:

(ANCHOR is a state program that allows New Jersey residents to get money back for their high property taxes.)

“When someone has an ANCHOR status page open in one page, then tries to open another ANCHOR status page, it will stop the user.”

Some of the users who got the message insisted they did not have multiple tabs or windows open. They said they tried to use the tool on computers and laptops, tablets and cell phones, but they continued to get the error.

Those who tried other browsers reported they were able to access the tool.



Now let's look at some direct hits:

Google accuses Microsoft of antitrust violations over Azure cloud platform
Sep 2024, Ars Technica

Google also said in its complaint, which was sent on Tuesday to the EU’s powerful competition unit, that it was concerned that Microsoft was degrading the user experience of those customers that were moving their Windows software to competing cloud providers.

This is almost exactly and literally the same language used by Mehta against Google in the Oct 2024 DOJ anti trust suit: "Plaintiffs contend that Google’s conduct has caused three anticompetitive effects particular to the text ads market: ... (3) product degradation through diminished transparency regarding text ads auctions. ... ." (p258, link
[^Note that Google's customers are not the average internet user who runs a search, but advertising people, so when they talk about the deterioration of the product, it's of the advertisement4, or its placement, or its audience, not the search results.]

Google accused of shadow campaigns redirecting antitrust scrutiny to Microsoft
Oct 2024, Ars Technica

FTX sues Binance for $1.76B in battle of crypto exchanges founded by convicts
Nov 2024, Ars Technica

Albertsons calls off merger and sues Kroger
Dec 2024, CNN Business