Ah, Emergent Behavior: Tough to Predict, Right?

December 28, 2022

Super manager Jeff (I manage people well) Dean and a gam of Googlers published “Emergent Abilities of Large Language Models.” The idea is that those smart software systems informed by ingesting large volumes of content demonstrate behaviors the developers did not expect. Surprise!

Also, Google published a slightly less turgid discussion of the paper which has 16 authors. in a blog post called “Characterizing Emergent Phenomena in Large Language Models.” This post went live in November 2022, but the time required to grind through the 30 page “technical” excursion was not available to me until this weekend. (Hey, being retired and working on my new lectures for 2023 is time-consuming. Plus, disentangling Google’s techy content marketing from the often tough to figure out text and tiny graphs is not easy for my 78 year old eyes.

image

Helpful, right? Source: https://openreview.net/pdf?id=yzkSU5zdwD

In a nutshell, the smart software does things the wizards had not anticipated. According to the blog post:

The existence of emergent abilities has a range of implications. For example, because emergent few-shot prompted abilities and strategies are not explicitly encoded in pre-training, researchers may not know the full scope of few-shot prompted abilities of current language models. Moreover, the emergence of new abilities as a function of model scale raises the question of whether further scaling will potentially endow even larger models with new emergent abilities. Identifying emergent abilities in large language models is a first step in understanding such phenomena and their potential impact on future model capabilities. Why does scaling unlock emergent abilities? Because computational resources are expensive, can emergent abilities be unlocked via other methods without increased scaling (e.g., better model architectures or training techniques)? Will new real-world applications of language models become unlocked when certain abilities emerge? Analyzing and understanding the behaviors of language models, including emergent behaviors that arise from scaling, is an important research question as the field of NLP continues to grow.

The write up emulates other Googlers’ technical write ups. I noted several facets of the topic not included in the paper on OpenReview.net’s version of the paper. (Note: Snag this document now because many Google papers, particularly research papers, have a tendency to become unfindable for the casual online search expert.)

First, emergent behavior means humans were able to observe unexpected outputs or actions. The question is, “What less obvious emergent behaviors are operating within the code edifice?” Is it possible the wizards are blind to more substantive but subtle processes. Could some of these processes be negative? If so, which are and how does the observer identify those before an undesirable or harmful outcome is discovered?

Second, emergent behavior, in my view of bio-emulating systems, evokes the metaphor of cancer. If we assume the emergent behavior is cancerous, what’s the mechanism for communicating these behaviors to others working in the field in a responsible way? Writing a 30 page technical paper takes time, even for super duper Googlers. Perhaps the “emergent” angle requires a bit more pedal to the metal?

Third, how does the emergent behavior fit into the Google plan to make its approach to smart software the de facto standard? There is big money at stake because more and more organizations will want smart software. But will these outfits sign up with a system that demonstrates what might be called “off the reservation” behavior? One example is the use of Google methods for war fighting? Will smart software write a sympathy note to those affected by an emergent behavior or just a plain incorrect answer buried in a subsystem?

Net net: I discuss emergent behavior in my lecture about shadow online services. I cover what the software does and what use humans make of these little understood yet rapidly diffusing methods.

Stephen E Arnold, December 28, 2022

Online and Brick Crime Ecosystem: Not Just Surviving, It Is Thriving

December 28, 2022

For some US cities, looting has become a daily occurrence. Retailers in San Francisco are left helpless as their merchandise is stolen. Looting happens when things become violent during protests centered on ethnicity or politics. It appears looting is a victimless crime, because thieves are hitting up large retail chains and luxury stores. Yahoo! Finance details how Target is facing a huge profit loss because of shoplifting and it is the same for other stores too: “Target: ‘Organized Retail Crime’ Has Driven $400 Million In Extra Profit Loss This Year.”

Target has lost $400 million in gross profit in compared to last year’s third-quarter results and the CEO projects a $600 million loss for all of 2022. Target attributes profit shrinkage to “organized retail crime.” Why are more people stealing these days than before? The Yahoo Finance Editor-In-Chief Andy Serwer summed it up as a zeitgeist issue:

“”Why are people stealing these days? That’s a tough one. To some degree it’s a reflection of our times. Simply put, America’s social contract is straining. Until recently we’ve been able to lay out goods—often in mammoth, big box stores with only a handful of employees. When our social contract is strong—i.e people are getting a fair shake—it’s a model that works. Now it seems more people are stealing instead. (BTW, our stressed social contract may be capping how far we can push this people-light, technology-heavy model. Last month Wegman’s ended its scan-and-go shopping app. Why? Shrinkage, of course.)”

Other factors include a widening wealth gap, companies failing to pay workers a living wage, shoppers being violent toward employees, external thefts, and post-COVID mentalities. The article, however, failed to mention how easy it is to fence stolen products. Online commerce Web sites such as Amazon, eBay, Mercer, Swappa, the Real Real, and social media marketplaces are teaming with stolen goods. It is a little harder to drop luxury items, but everyday products like electronics, baby formula, diapers, toys, and alcohol go quickly. These Web sites do little to vet the sellers, although Amazon has some blocks and eBay scans for “counterfeit goods” and limits the sale of certain items.

These prevention measures do little to stop thieves from hawking their stolen

merchandise online.

Whitney Grace, December 28, 2022

Ah, Lawyers: What One Does Not See Others Will

December 27, 2022

I read “The Copyright Industry Is about to Discover That There Are Hundreds of Thousands of Songs Generated by AI Already Available, Already Popular.” The write does a typical lawyer thing: Presenting in cool tones a logical argument. Is there a problem with this? Nope; however, what one lawyer presents as a logical argument, there will be other legal eagles preparing more logical arguments backed by a business model, knowledge of litigation processes, and the money or clout to push a matter forward.

After you read the cited article, navigate to PicRights. Read the verbiage. Then run a query on Reddit for the business entity PicRights. Now form a mental picture of this type of firm equipped with a signed letter which says, “Protect my rights against IP thieves. We agree to split the money extracted from these scofflaws, thieves, cut-purses, and content recycles.”

Got the picture.

The wide diffusion of smart software which does human like things open the door to a major business opportunity for PicRights type companies. Plus, if you need a lawyer, I have heard that a Higbee & Associates-type of law firm is skilled in this facet of assorted laws, conventions, and regulations.

The opportunity to extract money from machine generated images, music, and other content is unlimited.

Stephen E Arnold, December 27, 2022

Are Bad Actors Working for Thrills?

December 27, 2022

Nope, some bad actors may be forced to participate in online criminal behavior. Threats, intimidation, a beating or two, or worse can focus some people to do what is required.

The person trying to swindle you online might be doing so under duress. “Cyber Criminals Hold Asian Tech Workers Captive in Scam Factories,” reports Context. The article begins with the story of Stephen Wesley, an Indian engineer who thought he was taking a graphic design job in Thailand. Instead he found himself carted off to Myanmar, relieved of his passport and phone, and forced to work up to 18 hours a day perpetuating crypto currency scams. This went on for 45 days, until he and about 130 others were rescued from such operations by Indian authorities. Reporters Anuradha Nagaraj and Nanchanok Wongsamuth reveal:

“Thousands of people, many with tech skills, have been lured by social media advertisements promising well-paid jobs in Cambodia, Laos and Myanmar, only to find themselves forced to defraud strangers worldwide via the internet. … The cybercrime rings first emerged in Cambodia, but have since moved into other countries in the region and are targeting more tech-savvy workers, including from India and Malaysia. Authorities in these countries and United Nations officials have said they are run by Chinese gangsters who control gambling across southeast Asia and are making up for losses during the pandemic lockdowns. The experts say the trafficked captives are held in large compounds in converted casinos in Cambodia, and in special economic zones in Myanmar and Laos. ‘The gangs targeted skilled, tech-savvy workers who had lost jobs during the pandemic and were desperate, and fell for these bogus recruitment ads,’ said Phil Robertson, deputy director for Asia at Human Rights Watch. ‘Authorities have been slow to respond, and in many cases these people are not being treated as victims of trafficking, but as criminals because they were caught up in these scams.'”

A long-game tactic typically used by these outfits is eloquently named “pig butchering,” wherein the operator builds trust with each victim through fake profiles on social media, messaging apps, and dating apps. Once the mark is hooked, the involuntary con artist pressures them to invest in phony crypto or trading schemes. Beware virtual suitors bearing unique investment opportunities.

Sadly, recent tech layoffs are bound to accelerate this trend. Bad actors are not going to pass up a chance to get talent cheaply. Myanmar’s current government, which seized power in February 2021, declined to comment. After months of denying the problem existed, we are told, Cambodian officials are finally cracking down on these operations. The article states thousands of workers are still trapped.

Business is business as the saying goes.

Cynthia Murrell, December 27, 2022

Amazon: Filled with Holiday Cheer

December 27, 2022

In the modern world, Amazon is a catch-22; you can live without it but you would be hard-pressed not to. While buying a product on Amazon is usually simple, the return process is worse than a pain in the neck. A Canadian family has learned the hard way that Amazon will pull a scam whenever possible. CBC explains the frustrations in, “Family Says Amazon Shipped Fake Product, Refuses Refund Until ‘Correct Item’ Returned.”

Matthew Legault’s parents purchased him a computer as a high school graduation gift. He was excited about the present, until he took it apart and noticed the graphics card was hollowed out and filled with putty. The Legaults returned the defective computer and asked for a refund. Amazon, however, said they threw out the “computer” because it was a danger to employees. Amazon also demanded the Legaults need to ship the correct item.

Do you see where the cyclical problem is going?

The Legaults are like many unhappy Amazon customers, who are told that an item was disposed of to “end the conversation.”

What is hysterical is that Legault patriarch thinks Amazon cares about him:

“François says his history with Amazon should have stood for something — he’s been a loyal customer for years, and rarely returned anything. ‘The box had obviously been tampered with,’ he said. ‘We kind of expected that Amazon would have better quality controls, better procedures to ensure that something like this doesn’t happen.’”

HAHA!

NOPE!

Amazon is not a brick and mortar store. Amazon does not inspect anything and simply boxes returns up to be auctioned off to the highest bidder.

Amazon has lost revenue since the end of the COVID pandemic. The company is laying off employees.

A word of advice from the victims: take a video of yourself opening the package, so Amazon will not have any fodder against you.

The Legaults eventually got their refund when the story aired. It also helps to threaten legal action.

Is it possible that the post Bezos Amazon does not care about anyone or anything? Some more advice: order expensive products from the manufacturer’s Web site or in a physical store. Just a thought.

Whitney Grace, December 27, 2022

Google: Rank Ordering Its Wizards, Shamen, and Necromancers

December 27, 2022

Okay, six percent of the magic workers are not sufficiently Google. The figure does not count Timnit Gebru types.

Google is not afraid to fire anyone who ignites controversy within the company related to diversity and women. Sometimes it is not bad press that causes Google to lay off its employees, instead it is the economy. The Daily Hunt reports that, “Google Asked Managers To Fire 10,000 ‘Poor Performers’ As Mass Layoffs Hit Tech Sector.”

The US federal government’s raising interest rates and tech companies that make a large portion of their profits from ads are feeling the pain. Meta, Google, Amazon, Twitter, and more companies are firing more workers. Alphabet is telling its managers to lay off all employees who are rated as “poor performers.” The hope is to get rid of at least 10,000 workers and there might be some subterfuge behind it:

“As per a report from Forbes, Google might even bank on these rankings to avoid paying bonuses and stock grants. Google’s managers have been reportedly asked to categorize 10,000 employees as “poor performers” so that 10,000 people can be fired. Alphabet has a total workforce of 187,000 people, which is one of the largest workforces in tech.”

Google’s workforce is described as bloated and pays its employees 70% more than Microsoft compensates its staff or 153% compared to the top twenty big tech companies. Google pays more than its competition to hoard talent and increases its stranglehold on the tech industry.

Googzilla has to pay for NFL football any way it can.

Whitney Grace, December 27, 2022

Google and Its Puzzles: Insiders Only, Please

December 26, 2022

ProPublica made available an article of some importance in my opinion. “Porn, Piracy, Fraud: What Lurks Inside Google’s Black Box Ad Empire” walks through the intentional, quite specific engineering of its crucial advertising system to maximize revenue and befuddle (is “defraud” a synonym?) advertisers. I was asked more than a decade ago to do a presentation of my team’s research into Google’s advertising methodology. I declined. At that time, I was doing some consulting work for a company I am not permitted to name. That contract stipulated that I would not talk about a certain firm’s business technologies. I signed because… money.

The ProPublica essay does the revealing about what is presented as a duplicitous, underhanded, and probably illegal business process subsystem. I don’t have to present any of the information I have gathered over the years. I can cite this important article and point out several rocks which the capable writers at ProPublica either did not notice or flipped them over and concluded, “Nah, nothing to see here.”

I urge you to do two things. First, read the ProPublica write up. Number Two: Print it out. My hunch is that it may be disappeared or become quite difficult to find at some point in the future. Why? Ah, grasshopper, that is a question easily answered by the managers who set up Foundem and who were stomped by Googzilla. Alternatively you could chase down a person at the French government tax authority and ask, “Why were French tax forms not findable via a Google search for several years.” These individuals might have the information you need. Shifting gears: Ask Magix, the software company responsible for Sony Vegas why cracks for the software appear in YouTube videos. If you use your imagination, you will come up with ideas for gathering first person information about the lovable online advertising company’s systems and methods. Hint: Look up Dr. Timnit Gebru and inquire about her interactions with one of Google chief scientists. I guarantee that a useful anecdote will bubble up.

So what’s in the write up. Let me highlight a main point and then cite a handful of interesting statements in the article.

What is the main point? In my opinion, ProPublica’s write up says, “The GOOG maximizes its return at the expense of the advertisers and of the users.”

Who knew? Not me. I think the Alphabet Google YouTube DeepMind outfit is the most wonderfulest company in the world. Remember: You heard this here first. I have a priceless Google mouse pad too.

Consider these three statements from the essay. First, Google lingo is interesting:

Google spokesperson Michael Aciman said the company uses a combination of human oversight, automation and self-serve tools to protect ad buyers and said publisher confidentiality is not associated with abuse or low quality.

The idea is that Google is interested in using a hybrid method to protect ad buyers. Plus there is a difference between publishers and confidentiality. I find it interesting that instead of talking about [a] the ads themselves (porn, drugs, etc.), [b] the buyers of advertising which is a distinct industry dependent upon Google for revenue, [c] the companies who want to get their message in front of people allegedly interested in the product of service, or [d] the user of search or some other Google service. Google wants to “protect ad buyers.” And what about the others I have identified? Google doesn’t care. Logical sure but doesn’t Google have the other entities in mind? That’s a question regulators should have asked and had answered after Google settle the litigation with Yahoo over advertising technology, at the time of Google’s acquisition of Oingo (Applied Semantics), or at the time Google acquired DoubleClick. In my opinion, much of the ProPublica write up operates in a neverland of weird Google speak, not the reality of harvesting money from those largely in the dark about what’s happening in the business processes.

Second, consider this statement:

we matched 70% of the accounts in Google’s ad sellers list to one or more domains or apps, more than any dataset ProPublica is aware of. But we couldn’t find all of Google’s publisher partners. What we did find was a system so large, secretive and bafflingly complex that it proved impossible to uncover everyone Google works with and where it’s sending advertisers’ money.

The passage seems to suggest that Google’s engineers went beyond clever and ventured into the murky acreage of intentional obfuscation. It seems as if Google wanted to be able to consume advertising budgets without any entity having the ability to determine [a] if the ad were displayed in a suitable context; that is, did the advertiser’s message match the needs of the user to who the ad was shown.  And [b] was the ad appropriate even if it contained words and phrases on Google’s unofficial stop word lists. (If you have not see these, send an email to benkent2020 at yahoo dot com and one of my team will email you some of the more interesting words that guarantee Google’s somewhat lax processes will definitely try to block. If a word is not on a Google stop list, then the messages will probably be displayed. Remember: As Google terminates six percent of its staff, some of those humans presumably will not be able to review ads per item one above. And [c] note the word “bafflingly”. The focus of much Google engineering over the last 15 years has been to build competitive barriers, extent the monopoly function with “partners”, and double talk in order to keep regulators and curious Congressional people away. That’s my take on  this passage.

Now for the third passage I will cite:

…we uncovered scores of previously unreported peddlers of pirated content, porn and fake audiences that take advantage of Google’s lax oversight to rake in revenue.

I don’t need to say much more about this statement that look at and think about pirated content (copyright), porn (illegal content in some jurisdictions) and fake audiences (cyber fraud). Does this statement suggest that Google is a criminal enterprise? That’s a good question.

I have some high level observations about this excellent article in ProPublica. I offer these in the hope that ProPublica will explore some of these topics or an enterprising graduate student will consider the statements and do some digging.

  1. Why is Google unable to manage its staff? This is an important question because the ad behaviors described in the ProPublica article are the result of executive compensation plans and incentives. Are employees rewarded for implementing operations that further “soft” fraud or worse?
  2. How will Google operate in a more fragmented, more regulated environment? Is one possible behavior a refusal to modify the guiding hand of compensation and incentive programs away from generating more and more money within external constraints? My hunch is that Google will do whatever is necessary to build its revenue.
  3. What mechanisms exist or will be implemented to keep Google’s automated systems operating in a legal, ethical way?

Net net: Finally, after decades of craziness about how wonderful Googzilla is, more critical research is appearing. Is it too little and too late? In my view, yes.

Stephen E Arnold, December 26, 2022

Ka-Ching, Ka-Ching: The Unforgettable Tune Haunting Meta

December 26, 2022

What’s the sound of the European cash registers? Ka-ching, ka-ching. The melody of money.

I spotted two allegedly true real news stories this morning that sound like previous write ups about the Zuckbook, oops, Facebook, oh, darn, Meta.

The first is “Meta Settles Cambridge Analytica Class-Action Lawsuit for $725 Million.” The story reports:

Parent company Meta has agreed to pay $725 million to settle a long-running class-action lawsuit accusing Facebook of allowing Cambridge Analytica and other third parties to access user’s private information…

Cambridge Analytica. Four years ago. Links to possible intelligence activities in a certain special operation favoring nation state. Lies. What’s not to love? The is writing a check for $725 million. In terms of fellow traveler Twitter that’s peanuts. No big deal. Guilty? “Senator, thank you for that question….”

The second write up is “EU Tells Meta That Facebook Marketplace Breaches Anti-Trust Rules.” This write up explains:

In yet another demonstration of the new aggressive stance of European regulators, the EU has lodged a notice accusing Meta of ‘abusive practices’. The allegation relates to the fact that Facebook Marketplace currently ties its online classified ads service to the Facebook social network. The European Commission alleges that the company imposes unfair trading conditions on rivals to its Facebook Marketplace, in order to gain a commercial advantage.

How disrespectful! The Zuck has worked to bring people together. The Marketplace does that. I live in rural Kentucky. One of my neighbors was killed meeting a person to purchase a mobile phone. The connection making mechanism seems to have been a certain social network. It is clear to me that I don’t understand the bridge building, good vibe generating mechanisms of social media. That’s okay. I am a dinobaby.

Let’s get back to the music of ka-ching, ka-ching.

I have some expectation that the European Union will be ringing its cash registers in 2023. Some US high-technology companies output cash when the tune ka-ching, ka-ching sounds. That may be a  conditioned reflex similar to Ivan Pavlov’s demonstration of conditioning. Yes, ka-ching may be a suitable substitute for dog treats. Woof.

In my opinion, the ka-ching, ka-ching thing suggests:

  1. A pattern of behavior that flaunts expected business practices
  2. Management attitudes that encourage behavior some lawmakers consider illegal
  3. An increasing perception that some of the societal pressure points are becoming more sensitive because of certain social media practices.

Fair or unfair? True or false?

Well, just listen: Ka-ching, ka-ching.

Stephen E Arnold, December 26, 2022

AI: Pollution and Digital Garbage

December 26, 2022

Humans are polluting the Earth with our addiction to consumerism, meat, and halted advancement in recycling technology. In an ironic twist of fate, the human created AI algorithms are generating decabytes of digital garbage. Ploum wrote about the digital cash in the post: “Drowning In AI Generated Garbage: The Silent War We Are Fighting.”

Ploum asserts we are experiencing the “spectacular results” of forty years of work put into statistical algorithms. He points to “deep fake” videos of public figures, the ability to copycat voices, and digital paintings comparable to masterpieces. AI algorithms need information to learn; the Internet is the greatest bastion of human knowledge (and filth). Everything that AI algorithms learned from was created by humans and now they are creating their own stuff.

It sounds magical, right?

Yes, except it is bad.

“The algorithms are already feeding themselves on their own data. And, as any graduate student will tell you, training on your own results is usually a bad idea. You end sooner or later with pure overfitted inbred garbage. Eating your own shit is never healthy in the long run. Twitter and Facebook are good examples of such algorithmic trash. The problem is that they managed to become too powerful and influential before we realised it was trash…Fascinating garbage but garbage nonetheless.

The robot invasion started 15 years ago, mostly unnoticed. We were expecting killing robots, we didn’t realise we were drowned in AI generated garbage. We will never fight laser wearing Terminators. Instead, we have to outsmart algorithms which are making us dumb enough to fight one against the other.”

The fix? To resist the robot takeover humans need to unplug and engage in reality. That is great advice, except it requires effort and human contact. While some humans are okay, the vast majority stink like the digital garbage.

Whitney Grace, December 26, 2022

How Does a China-Affiliated Outfit Identify Insider Threats? Surveillance, Of Course

December 26, 2022

I will not revisit my comments about the risks posed by TikTok to the US. I do find it amusing that statements offered in one of those “Thank you, Senator, for that question….” sessions has been demonstrated to be false. Hello, perjury?

TokTok Admits Tracking FT Journalist in Leaks Investigation” reports:

Two members of staff in the US and two in China gained access to the IP addresses and other personal data of FT journalist Cristina
Criddle, to work out if she was in the proximity of any ByteDance employees, the company said. However, the company failed to find any leaks.
A BuzzFeed journalist and a number of users connected to the reporters through their TikTok accounts were also targeted.

A government has the capacity to surveil who and when and where it wants. When a company focuses on vulnerable demographics and is directly affiliated with a government, Houston, we have a problem.

More problematic when that government/company can feed information to targeted users, that information can shape the impressionable target’s world view. That’s an opportunity creator. Toss in keep track of what immature minds do only may provide some useful information to force a target to take an action or else. The else can include salacious videos and much, much more interesting immature behavior. If released, the mature version of the nude dance at a high school party might derail a promotion at a secretive high-tech company, creating an opening for a more compliant target to apply for the job. Exciting? Yep.

What’s the tally? Deception. Check. Invasion of a non Chinese citizen’s rights. Check. Information warfare. Check.

Yep, TikTok.

Stephen E Arnold, December 26, 2022

« Previous PageNext Page »

  • Archives

  • Recent Posts

  • Meta