Models, Models Everywhere: Not a Doubt in Sight
October 7, 2021
In 2017, computers became better at generating and understanding human language when Google researchers designed the natural language AI, Transformers. Fast Company explains why natural language is important in the article, “Ex-Googlers Raise $40 Million To Democratize Natural-Language AI.”
Three of the Transformers AI researchers, Nick Frosst, Ivan Zhang, and Aidan Gomez, began their own startup, Cohere, and raised $40 million in funding. They started Cohere to commercialize and further develop the natural language processing AI. The Cohere plan to address biases accidentally programmed into AI when they are taught with bad datasets. These biases are unfavorable to ethnic minorities and women, basically anyone who is not a white man.
Transformers AI models need huge amounts of data in order to be programmed, but only organizations with supercomputers have the necessary high quality natural language models. The Cohere team want to democratize NLP models and make them available to organizations that otherwise would not have the funds for the technology. Cohere wants to guarantee its NLP AI will not contain any biases:
“To address the risks, Cohere’s engineers have implemented quality control tests to look for any issues with the model before release, and the company continues to monitor its models after launch as well. In addition, Gomez says Cohere will publish “data statements,” which will including information about training data, its limitations, and any risks—a concept first popularized by Gebru. Cohere has also established an external Responsibility Council that will help oversee the safe application of the company’s AI. The company declined to share who is part of the council.”
Frosst, Zhang, and Gomez embrace the technological biases in AI, but instead of reacting poorly, like Google did with Timnit Gehru, they admit the mistake and are actively creating a solution. They also made their own company, will probably earn handsome salaries, and help shape future AI.
Whitney Grace, October 6, 2021
Key Words: Useful Things
October 7, 2021
In the middle of nowhere in the American southwest, lunch time conversation turned to surveillance. I mentioned a couple of characteristics of modern smartphones, butjec people put down their sandwiches. I changed the subject. Later, when a wispy LTE signal permitted, I read “Google Is Giving Data to Police Based on Search Keywords, Court Docs Show.” This is an example of information which I don’t think should be made public.
The write up states:
Court documents showed that Google provided the IP addresses of people who searched for the arson victim’s address, which investigators tied to a phone number belonging to Williams. Police then used the phone number records to pinpoint the location of Williams’ device near the arson, according to court documents.
I want to point out that any string could contain actionable information; to wit:
- The name or abbreviation of a chemical substance
- An address of an entity
- A slang term for a controlled substance
- A specific geographic area or a latitude and longitude designation on a Google map.
With data federation and cross correlation, some specialized software systems can knit together disparate items of information in a useful manner.
The data and the analytic tools are essential for some government activities. Careless release of such sensitive information has unanticipated downstream consequences. Old fashioned secrecy has some upsides in my opinion.
Stephen E Arnold, October 7, 2021
Facebook: Why Change?
October 6, 2021
I read “Facebook Can’t Be Saved.” The main point struck me as:
Facebook has experienced years of intense scrutiny over the exact issues that are being discussed in the wake of Haugen’s revelations, and has only succeeded in making its inherent problems worse. During the hearing, Haugen compared fixing Facebook’s issues to mandating that cars come with seat belts. But maybe Facebook doesn’t need a seat belt. Maybe it just needs to stop being given more chances.
This is an interesting analogy. I would ask this question, “Why should Facebook change?” The company has loyal users, lobbyists, and friends in high places. The available consequences are fines and enduring hearings and legal proceedings.
After watching the testimony by the whistle blower, my hunch is that Facebook will evolve. But the deep machine is chugging along.
Stephen E Arnold, October 6, 2021
The Darknet: a Dangerous Place
October 6, 2021
Criminal activity on the Darknet is growing and evolving. One person who has taken it on themselves to study the shadow realm shares some of their experiences and observations with reporter Vilius Petkauskas in, “Darknet Researcher: They Said They’ll Come and Kill Me—Interview” at CyberNews. The anonymous interviewee, who works with research firm DarkOwl, describes a threat to their life, one serious enough to prompt them to physically move their family to a new home. They state:
“There was one specific criminal actor I was going after, trying to figure out where they were operating, who they were involved with, what groups they were affiliated with. I became a target. They turned on me and said, we will find whoever wrote this and come kill them. We will destroy them.”
Yes, poking around the Darknet can be dangerous business. What sorts of insights has our brave explorer found? Recently, there has been a substantial uptick in ransomware, and for good reason. The researcher explains:
“Look at ransomware as a service (RaaS). First and second-generation ransomware lockers were developed by incredibly smart malware developers, cryptologists, and encryption specialists. Those who designed and employed such software were some of the most sophisticated malware developers or ‘elite’ hackers around if you want to label them that. But with the RaaS affiliate model, they’re giving others the chance to ‘rent’ ransomware for as little as a few hundred bucks a year, depending on which strain they’re using. Anyone interested in getting into the business of ransomware can enter the market without necessarily having any prior or expert knowledge of how to conduct an enterprise-level attack against a network. Some of the gangs, like Lockbit 2.0 are nearly entirely automated, and their affiliates don’t need to have the slightest clue what they’re doing. You just push, plug, and play. Identify the victim, drop it onto the network, and the rest is taken care of.”
How convenient. Getting into the target’s network, though, is another matter. For that criminals turn to
initial access brokers (IABs), also located on the Darknet, who help breach networks through vulnerabilities, leaked credentials, and other weaknesses. See the write-up for more of the researchers hard-won observations. They close with this warning—there is more going on here than opportunists looking to make a buck. Espionage and cyber terrorism are also likely involved, they say. We cannot say we are surprised.
Cynthia Murrell, October 6, 2021
Ex-Googlers Work On Biased NLP Solutions
October 6, 2021
Google is on top of the world when it comes to money and technology. Google is the world’s most used search engine, its Chrome Web browser is used by two-thirds of users, and about 29% of 2021 digital advertising were Google ads. Fast Company asks and investigates important questions about Google’s product quality in: “It’s Not Just You. Google Search Really Is Getting Worse.”
Over 80% of Alphabet Inc.’s revenue, Google’s parent company, comes from advertising revenue and about 85% of the world’s search engine traffic feeds through Google. Google controls a lot of users’ screen time. The search engine’s quality results have been studied and researchers have learned that very few users scroll past the “fold” (all of the available content on a screen). Advertising space at the top of search results is incredibly valuable. It also means that users are forced to scroll further and further to reach non-paid results.
Alphabet Inc. has another revenue generating platform, YouTube. A huge portion of videos include multiple ads. Users can avoid ads by paying for a premium subscription, but very few do.
Google does want to improve its search quality. Currently a lot of information from queries are distributed across multiple Web sites. Google wants to condense everything:
“Google is working on bringing this information together. The search engine now uses sophisticated “natural language processing” software called BERT, developed in 2018, that tries to identify the intention behind a search, rather than simply searching strings of text. AskJeeves tried something similar in 1997, but the technology is now more advanced.
BERT will soon be succeeded by MUM (Multitask Unified Model), which tries to go a step further and understand the context of a search and provide more refined answers. Google claims MUM may be 1,000 times more powerful than BERT, and be able to provide the kind of advice a human expert might for questions without a direct answer.”
Google controls a huge portion of the Internet and how users utilize it. Alphabet Inc. is here to stay for a long time, but there are alternatives such as Bing, DuckDuckGo, Ecosia, and Tor browsers. Google, however, will one day fade. Sears Roebuck, Blockbuster, Kmart, cassettes, etc. were al household names, until they became obsolete.
Whitney Grace, October 6, 2021
Internet Defreedoming: An Emerging and Surging Market Sector
October 6, 2021
Check out “The Global Drive to Control big Tech.” The information in the article and the report suggest that Internet freedom is decreasing. The write up makes this point and supports it with data and research:
In the high-stakes battle between states and technology companies, the rights of internet users have become the main casualties.
The data can be viewed from a different perspective; namely, censorship is a growth business. Products and services needed to censor content at scale are available, but these are often clunkers or complex add ins to network components which are under load and often less reliable than a used Lada.
Opportunities include:
- Repurposed software designed for artificial intelligence operations; for example, identifying and flagging problematic content
- Workflow software which can automate the removal, posting of flags, or once in a while notifying a person his or her content is problematic
- Tools for locating objectionable content and triggering removal, logging the issue, and locating other instances of the “problem.”
After a decade of consistent censorship growth, opportunities abound.
Censorship appears to be a hot business segment.
Stephen E Arnold, October 6, 2021
Yarchives: a Multi-Topic Repository of Information
October 5, 2021
Here is a useful resource, a repository of Usenet newsgroup articles collected and maintained by computer scientist Norman Yarvin. The Yarchive houses articles on twenty-two wide-ranging topics, from air conditioning to jokes to space. We note a couple that might be of interest to today’s assorted revolutionaries (or those tasked with countering them): explosives and nuclear technologies. Hmm. Perhaps there is a need to balance unfettered access to information with wisdom. The site’s About page reveals some details about Yarvin’s curation process. He writes:
“Articles are not put up here immediately; only a year or three after first saving them do I look at them again, sort them out, and make index pages for them. (By that time I’ve forgotten enough of them to make them worth rereading — and if I find they are not worth rereading, I discard them.) I’ve largely automated the making of index pages; the programs I’ve written for it (mostly in Perl) are available as a tar file (tools.tar). The making of the links to search for Google’s copy of each article is also automated. If it stops working because Google changed their query syntax, please let me know. Links that are on the Message-ID line of the header should link straight to the article in question; other links (from articles I’ve lost the Message-ID for) should invoke a search. For articles from the linux-kernel mailing list, links that are on the Original-Message-ID line of the header are to kernel.org’s copy of the article. (They used to be to GMANE, but that service went away.) Some changes have been made to these articles, but nothing that would destroy any possible meaning.”
The project seems to be quite the hobby for Yarvin. He goes on to describe the light corrections he makes, articles’ conversion to the UTF-8 character encoding, and his detailed process of checking the worthiness of URLs and making the valuable ones clickable.
Readers may want to peruse the Yarchive and/or bookmark it for future use. Information relevant to many of our readers can be found here, like files on computers, electronics, and security. More generally useful topics are also represented; cars, food, and houses, for example. Then there are the more specialized topics, like bicycles, chemistry, and metalworking. There is something here for everyone, it seems.
Cynthia Murrell, October 5, 2021
Zoom Blunders Can Be Tricky for Employees
October 5, 2021
Zoom meetings seemed like the logical answer to collaborating from home during the pandemic, and its popularity is likely to last. However, asks Hacker Noon, “Has Zoom Made Us ‘Embrace the Dark Side’ of Humanity?” Writer Michael Brooks, a remote-worker since long before COVID, came across some startling information. He tells us:
“I stumbled upon a Bloomberg article with an axing title: ‘Zoom-Call Gaffes Led to Someone Getting Axed, 1 in 4 Bosses Say.’ According to the results of a survey conducted by Vyopta Inc., which included ‘200 executives at the vice president level or higher at companies with at least 500 employees,’ nearly 25% of employees got fired. Why?! Wait! What? What in the world do you need to do during a Zoom call or any other virtual meeting or conference to get fired? It turns out that ‘mortal-virtual-sins’ include ‘joining a call late, having a bad Internet connection, accidentally sharing sensitive information, and of course, not knowing when to mute yourself.’”
A severe penalty indeed for folks working with an unfamiliar platform amidst the distractions of home, all while coping with the stresses of a global plague. Brooks describes how one might handle similar situations with more compassion:
“There was a baby crying loud in the middle of a meeting with my staff. I asked a proud dad, a member of our team, to introduce an adorable noisemaker. The baby joined and stayed throughout a meeting in her father’s arms. There was another team member who kept forgetting to hit the mute button when she wasn’t talking. The background noise was deafening as if she was calling from the busiest construction site in the world. … For the next meeting, the whole team pretended that there was something wrong with her mic. It lasted for a couple of hilarious minutes. Since then, we’ve never had to remind someone to mute themselves.”
Brooks wonders whether some gaffs represent a sort of rebellion against too much Zooming. If so, one’s job is a high price to pay. He suggests frustrated workers discuss the matter with bosses and coworkers instead of passive-aggressively sabotaging meetings. As for employers, they might want to consider lightening up a bit instead of axing a quarter of their talent for very human errors.
Cynthia Murrell, October 5, 2021
Why Good Enough Is a Winner
October 5, 2021
Low fidelity is a thing. “Why Lo-Fi Music Draws Listeners In” explains:
“Lo-fi” means “low-fidelity,” a term for music where you can hear imperfections that would typically be considered errors in the recording process. On YouTube channels like ChillHop music or DreamyCow, however, those “mistakes” become an intentional part of the listening experience.
With fancy technology and bandwidth, what’s with lousy audio data? Pulling in is garden variety magnetism or attraction.
Here’s the answer for music:
“It’s because people respond to the beat.”
American Bandstand made this truism a standard. Mr. Clark would ask a teen, “Why do you like the song?”
The teen would say, “I like the beat.”
So triggering a response based on a pattern is a potent magnetic force. The force operates when a “smart” online service provides content which attracts attention. The familiar generates a desire for more like this. The flaws are irrelevant.
My hunch is that this magnetic force for the “beat” — responding to something familiar, patterned, and emotional — operates within effective social media.
Addictive? Yes. Controllable? Not easily.
Stephen E Arnold, October xxl, 2021
AI: A Digital Mercury?
October 5, 2021
A half century ago, one of my high school science club friends and I gathered several broken mercury thermometers and pooled the mercury. We had liberated a porcelain dish from the chain smoking Mr. Shepherd’s drawer and marvelled at the properties of mercury. Although incredibly stupid, my friend Phil and I had read about mercury and knew that we should not eat it, smear it in our eyes, or get cute and heat it up with potassium ferricyanide.
We concluded that mercury from the thermometers was tough to corral, and it could be a quite problematic substance. My friend and I survived our at home lab work, but the image of the shiny globes and elusive behavior occurred to me when I read “Chinese AI Gets Ethical Guidelines for the First Time, Aligning with Beijing’s Goal of Reining in Big Tech.” The write up states:
Humans should have full decision-making power, the guidelines state, and have the right to choose whether to accept AI services, exit an interaction with an AI system or discontinue its operation at any time….The goal is to “make sure that artificial intelligence is always under the control of humans,” the guidelines state.
The approach is interesting for two reasons. First, some in China are concerned that smart software might be difficult to pin down. Like mercury perhaps? Second, deus ex machina forces are needed to deal with AI. Human “good judgment” may not do the trick.
Which path will work?
Does one trust a bureaucrat or a SAIL master in Palo Alto? This week’s DarkCyber video touches upon this black box filled with mercury.
Stephen E Arnold, October 5, 2021
-
- Subscribe to Beyond Search
Feature archive
News archive
-
