Facebook: Fooled by Ranking?
April 1, 2022
I sure hope the information in “A Facebook Bug Led to Increased Views of Harmful Content Over Six Months.” The subtitle is interesting too. “The social network touts downranking as a way to thwart problematic content, but what happens when that system breaks?”
The write up explains:
Instead of suppressing posts from repeat misinformation offenders that were reviewed by the company’s network of outside fact-checkers, the News Feed was instead giving the posts distribution, spiking views by as much as 30 percent globally.
Now let’s think about time. The article reports:
In 2018, CEO Mark Zuckerberg explained that downranking fights the impulse people have to inherently engage with “more sensationalist and provocative” content. “Our research suggests that no matter where we draw the lines for what is allowed, as a piece of content gets close to that line, people will engage with it more on average — even when they tell us afterwards they don’t like the content,” he wrote in a Facebook post at the time.
Why did this happen?
The answer may be that assumptions about the functionality of online systems must be verified by those who know the mechanisms used. Then the functions must be checked on a periodic business. The practice of slipstreaming changes may introduce malfunctions, which no one catches because no one is rewarded for slowing down the operation.
Based on my work for assorted reports and monographs, there are several other causes of a disconnect between what a high technology outfits and its systems actually do. Let me highlight what I call the Big Three:
- Explaining something that might be is different from delivering the reality of the system. Management wants to believe that code works, and not too many people want to be the person who says, “Yeah, this is what the system is actually doing?” Institutional momentum can crush certain types of behavior.
- The dependencies within complex software systems are not understood, particularly by recently hired outside experts, new hires, or — heaven help us — interns who are told to do X without meaningful checks, reviews, and fixes.
- An organization’s implicit policies keep feedback contained so the revenue continues to flow. Who gets promoted for screwing up ad sales? As a result, news releases, public statements, and sworn testimony operates in an adjacent but separate conceptual space from the mechanisms that generate live systems.
It has been my experience that when major problems are pointed out, reactions range from “What do you mean?” to a chuckled comment, “That’s just the way software works.”
What intrigues me is the larger question, “Is the revelation that Facebook smart software does not work as the company believed it did, the baseline for the company’s systems. On the other hand, the information could be an ill considered April Fool’s joke.
My hunch is that the article is not humor. Much of Facebook’s and Silicon Valley behavior does not tickly my funny bone. My prediction is that some US regulators and possibly Margrethe Vestager will take this information under advisement.
Stephen E Arnold, April 1, 2022
FinFisher Slips Beneath the Cold, Dark Sea
April 1, 2022
I read “Stage Win: FinFisher Is Bankrupt.” The main idea is that the somewhat controversial vendor of intelware is out of business. The article states:
“The end of FinFisher is not the end of the state Trojan market,” says Thorsten Schröder, who conducted the CCC FinSpy analysis together with Linus Neumann. “The employees who are now laid off will look for new jobs – presumably at competitors, who will probably also take over the customer base.” More important than the company going bankrupt, therefore, is a conclusion to the criminal proceedings. “We all hope that the end of FinFisher is just the beginning and that the competitors will also finally face legal and financial consequences,” says Linus Neumann.
What is interesting that as information moved from “secret” to open sources, the reaction to specialized software and services has been interesting. The MBA, cowboy emulating, entrepreneurial approach to generating revenue has had an impact.
Blub, blub. Is this the sound of that will haunt other vendors of specialized software and services? What if companies with fish names may face a similar fate?
Stephen E Arnold, April 1, 2022
The Art and Craft of Sending Document Copies to Legal Eagles: The Googley Method
April 1, 2022
Not joke. I read an allegedly accurate write up. It is called “Justice Department Accuses Google of Hiding Business Communications.” The idea is that in the US communications between a lawyer and his/her/them clients are privileged. I am not attorney, but the idea is to allow the lawyer to discuss sensitive issues with the his/her/them paying the bills.
The write up states:
The DOJ writes in its brief that Google teaches employees to request advice from counsel around sensitive business communications, thereby shielding documents from discovery in legal situations. Once counsel is involved, the company can treat the documents as protected under attorney-client privilege.
My view is that Google is just being “Googley.” When people who perceive themselves as entitled and really smart, those his/her/thems get advice from bright, often lesser individuals. The Googlers process the advice and when a suggestion measures up to Googzilla’s standards, the suggestion just sorta maybe becomes a way to handle certain issues.
Those who are Googley understand. Individuals who are not Googley — presumably like those in the Department of Justice — don’t understand the Googliness of the action.
Laws. Rules of the road. Those are often designed for the non Googley. The Googley must tolerate the others. But having the cash to throw legal cannon fodder in the path of the lesser lights who would do the Google harm is a useful tactic.
Stephen E Arnold, April 1, 2022
Google: The Quantum Supremacy Turtling
April 1, 2022
Okay, Aprils’ Fool Day.
“Google Wants to Win the Quantum Computing Race by Being the Tortoise, Not the Hare” explains that the quantum supremacy “winner” which captured “time crystals” has a new angle:
it’s clear that Google — or, to be more accurate, its parent company Alphabet — has its sights set on being the world’s premiere quantum computing organization.
Machines? Nah, think cloud, gentle reader. Google has it together, but the non Googley may struggle to get the picture. The write up says:
Parent company Alphabet recently starbursted its SandboxAQ division into its own company, now a Google sibling. It’s unclear exactly what SandboxAQ intends to do now that it’s spun out, but it’s positioned as a quantum-and-AI-as-a-service company. We expect it’ll begin servicing business clients in partnership with Google in the very near-term.
But? The write up says:
We can safely assume we haven’t seen the last of Google’s quantum computing research breakthroughs, and that tells us we could very well be living in the moments right before the slow-and-steady tortoise starts to make up ground on the speedy hare.
Maybe turtle? An ectotherm like Googzilla? Eye glass frames with a relevant Google product review? So many questions.
Stephen E Arnold, April 1, 2022
How Robots Learn: Fits and Starts
April 1, 2022
Navigate to “This Video of the Mini Cheetah Running Is Both Inspiring and Hilarious.” The video is on YouTube at this link. The idea is that instead of programming to teach a dog-like robot to run, the machines learns. The approach is called “model free learning.”
Several observations:
- Model free means that a human does not have to figure out each instruction and then write it up for the robot. The implication is that the robot learns like a human.
- Humans learning to walk typically have a humanoid around to make sure the learner does not end up in a pickle. Example: The robot runs into walls in the video. That is, I think, the hilarious part.
- The failures of the robot make it clear that if devices using model free learning methods are released before the robot learns the ins and outs of a process, disaster is probable.
The failures are a-okay for online ad matching and fiddling in a lab. Applied to an autonomous weapon, the hilarity could be short lived. What about a smart device taking a detour through a kindergarten? So maybe a slight error but no big deal. The method is less time consuming and more economical. Get those priorities straight.
Stephen E Arnold, April 1, 2022