Algorithm Tuning: Zeros and Ones Plus Human Judgment
October 23, 2020
This is the Korg OT-120 Orchestral Tuner. You can buy it on Amazon for $53. It is a chromatic tuner with an eight octave detection range that supports band and orchestra instruments. Physics tune pianos, organs, and other instruments. Science!
This is the traditional piano tuner’s kit.
You will need ears, judgment, and patience. Richard Feynman wrote a letter to a piano tuner. The interesting point in Dr. Feynman’s note was information about the non-zero stiffness of piano strings affects tuning. The implication? A piano tuner may have to factor in the harmonics of the human ear.
The Korg does hertz; the piano tuner does squishy human, wetware, and subjective things.
I thought about the boundary between algorithms and judgment in terms of piano tuning as I read “Facebook Manipulated the News You See to Appease Republicans, Insiders Say”, published by Mother Jones, an information service not happy with the notes generated by the Facebook really big organ. The main idea is that human judgment adjusted zeros, ones, and numerical recipes to obtain desirable results.
The write up reports:
In late 2017, Zuckerberg told his engineers and data scientists to design algorithmic “ranking changes” that would dial down the temperature.
Piano tuners fool around to deliver the “sound” judged “right” for the venue, the score, and the musician. Facebook seems to be grabbing the old-fashioned tuner’s kit, not the nifty zeros and ones gizmos.
The article adds:
The code was tweaked, and executives were given a new presentation showing less impact on these conservative sites and more harm to progressive-leaning publishers
What happened?
We learn:
for more than two years, the news diets of Facebook audiences have been spiked with hyper conservative content—content that would have reached far fewer people had the company not deliberately tweaked the dials to keep it coming, even as it throttled independent journalism. For the former employee, the episode was emblematic of the false equivalencies and anti-democratic impulses that have characterized Facebook’s actions in the age of Trump, and it became “one of the many reasons I left Facebook.”
The specific impact on Mother Jones was, according to the article:
Average traffic from Facebook to our content decreased 37 percent between the six months prior to the change and the six months after.
Human judgment about tool use reveal that information issues once sorted slowly by numerous gatekeepers can be done more efficiently. The ones and zeros, however, resolve to what a human decides. With a big information lever like Facebook, the effort for change may be slight, but the impact significant. The problem is not ones and zeros; the problem is human judgment, intent, and understanding of context. Get it wrong and people’s teeth are set on edge. Unpleasant. Some maestros throw tantrums and seek another tuner.
Stephen E Arnold, October 23, 2020
Music and Moods: Research Verifies the Obvious
October 21, 2020
It has been proven that music can have positive or negative psychological impacts on people. Following this train of research, Business Line reports that playlists are a better reflection of mood than once thought, “Your Playlist Mirrors Your Mood, Confirms IIIT-Hyderabad Study.”
The newest study on music and its effect on mood titled “Tag2risk: Harnessing Social Music Tags For Characterizing Depression Risk, Cover Over 500 Individuals” comes from the International Institute of Information Technology in Hyderabad (IIIT-H). The study discovered that people who listen to sad music can be thrown into depression. Vinoo Alluri and her students from IIIT-H’s cognitive science department investigated if they could identify music listeners with depressive tendencies from their music listening habits.
Over five hundred people’s music listening histories were studied. The researchers discovered that repeatedly listening to sad music was used as an avoidance tool and a coping mechanism. These practices, however, also kept people in depressive moods. Music listeners in the study were also drawn to music sub genres tagged with “sadness” and tenderness.
We noted:
“ ‘While it can be cathartic sometimes, repeatedly being in such states may be an indicator of potential underlying mental illness and this is reflected in their choice and usage of music,’ Vinoo Alluri points out. She feels that music listening habits can be changed. But, in order to do that, they need to be identified first by uncovering their listening habits. It is possible to break the pattern of “ruminative and repetitive music usage”, which will lead to a more positive outcome.”
Alluri’s study is an amazing investigation into the power and importance of music. Her research, however, only ratifies what music listeners and teenagers have known for decades.
Whitney Grace, October 21, 2020
Infohazards: Another 2020 Requirement
October 20, 2020
New technologies that become society staples have risks and require policies to rein in potential dangers. Artificial intelligence is a developing technology. Governing policies have yet to catch up with the emerging tool. Experts in computer science, government, and other controlling organizations need to discuss how to control AI says Vanessa Kosoy in the Less Wrong blog post: “Needed: AI Infohazard Policy.”
Kosoy approaches her discussion about the need for a controlling AI information policy with the standard science fiction warning argument: “AI risk is that AI is a danger, and therefore research into AI might be dangerous.” It is good to draw caution from science fiction to prevent real world disaster. Experts must develop a governing body of AI guidelines to determine what learned information should be shared and how to handle results that are not published.
Individuals and single organizations cannot make these decisions alone, even if they do have their own governing policies. Governing organizations and people must coordinate their knowledge regarding AI and develop a consensual policies to control AI information. Kozoy determines that any AI policy shoulder consider the following:
• “Some results might have implications that shorten the AI timelines, but are still good to publish since the distribution of outcomes is improved.
• Usually we shouldn’t even start working on something which is in the should-not-be-published category, but sometimes the implications only become clear later, and sometimes dangerous knowledge might still be net positive as long as it’s contained.
• In the midgame, it is unlikely for any given group to make it all the way to safe AGI by itself. Therefore, safe AGI is a broad collective effort and we should expect most results to be published. In the endgame, it might become likely for a given group to make it all the way to safe AGI. In this case, incentives for secrecy become stronger.
• The policy should not fail to address extreme situations that we only expect to arise rarely, because those situations might have especially major consequences.”
She continues that any AI information policy should determine the criteria for what information is published, what channels should be consulted to determine publication, and how to handle potentially dangerous information.
These questions are universal for any type of technology and information that has potential hazards. However, specificity of technological policies weeds out any pedantic bickering and sets standards for everyone, individuals and organizations. The problem is getting everyone to agree on the policies.
Whitney Grace, October 20, 2020
Tickeron: The Commercial System Which Reveals What Some Intel Professionals Have Relied on for Years
October 16, 2020
Are you curious about the capabilities of intelware systems developed by specialized services firms? You can gat a good idea about the type of information available to an authorized user:
- Without doing much more than plugging in an entity with a name
- Without running ad hoc queries like one does on free Web search systems unless there is a specific reason to move beyond the provided output
- Without reading a bunch of stuff and trying to figure out what’s reliable and what’s made up by a human or a text robot
- Without having to spend time decoding a table of numbers, a crazy looking chart, or figuring out weird colored blobs which represent significant correlations.
Sound like magic?
Nope, it is the application of pattern matching and established statistical methods to streams of data.
The company delivering this system, tailored to Robinhood-types and small brokerages, has been assembled by Tickeron. There’s original software, some middleware, and some acquired technology. Data are ingested and outputs indicate what to buy or sell or to know, as a country western star crooned, “know when to hold ‘em.”
A rah rah review appeared in The Stock Dork. “Tickeron Review: An AI-Powered Trading Platform That’s Worth the Hype” provides a reasonably good overview of the system. If you want to check out the system, navigate to Tickeron’s Web site.
Here’s an example of a “card,” the basic unit of information output from the system:
The key elements are:
- Icon to signal “think about buying” the stock
- A chart with red and green cues
- A hot link to text
- A game angle with the “odds” link
- A “more” link
- Hashtags (just like Twitter).
Now imaging this type of data presented to an intel officer monitoring a person of interest. Sound useful? The capability has been available for more than a decade. It’s interesting to see this type of intelware finds its way to those who want to invest like the wizards at the former Bear Stearns (remember that company, the bridge players, the implosion?).
DarkCyber thinks that the high-priced solutions available from Wall Street information providers may wonder about the $15 a month fee for the Tickeron service.
Keep in mind that predictions, if right, can allow you to buy an exotic car, an island, and a nice house in a Covid-free location. If incorrect, there’s van life.
The good news is that the functionality of intelware is finally becoming more widely available.
Stephen E Arnold, October 16, 2020
A New Role for Facial Recognition
October 6, 2020
The travel industry is finding its way around COVID-provoked limitations. Where once travelers were promised a “seamless” experience, they are now promised a “touchless” one, we learn from PhocusWire’s piece, “Touchless Tech: The Simple—and Advanced—Ways Ground Transport Providers Are Encouraging Travel.” Some measures are low-tech, like pledges to clean thoroughly, glove and mask use, and single-passenger rides instead of traditional shuttles. However, others are more technically advanced. The role of facial recognition in “touchless tickets” caught our eye. Writer Jill Menze reports:
“On the rail front, Eurostar has tapped facial-verification technology provider iProov to enable contactless travel from United Kingdom to France. With the solution, passengers can be identified without a ticket or passport when boarding the train, as well as complete border exit processes, at St. Pancras International station without encountering people or hardware. ‘What we’re trying to facilitate for the first time ever is a seamless process of going through ticket and border exit checks contactlessly and more fluidly than it’s ever been possible before using face verification,’ iProov founder and CEO Andrew Bud says. ‘That means, instead of checking people’s ID when they arrive, you check their ID long before. The idea is that you move the process of checking IDs away from the boarding point to the booking point.’ During booking, Eurostar will offer travelers an accelerated pre-boarding option, which allows passengers to scan their identity documentation using Eurostar’s app before using iProov’s facial biometric check, which uses patented controlled illumination to authenticate the user’s identity against the ID document. After that, travelers would not have to show a ticket or passport until they reach their destination.”
Eurostar plans to enact the technology next March, and Bud says other railway entities have expressed enthusiasm. This is an interesting use of facial recognition tech. It seems getting back to business is powerful motivation to innovate.
Cynthia Murrell, October 6, 2020
Twitter Photo Preview AI Suspected of Racial Bias
October 1, 2020
Is this yet another case of a misguided algorithm? BreakingNews.ie reports, “Twitter Investigating Photo Preview System After Racial Bias Claims.” Several Twitter users recently posted examples of the platform’s photo-preview function seeming to consider white people more important that black ones. Well that is not good. We’re told:
“The tech giant uses a system called neural network to automatically crop photo previews before you can click on them to view the full image. This focuses on the area identified as the ‘salient’ image region, where it is likely a person would look when freely viewing an entire photo. But tests by a number of people on the platform suggest that the technology may treat white faces as the focal point more frequently than black faces. One example posted online shows American politician Mitch McConnell and Barack Obama, with the system favoring Mr. McConnell in its preview over the former US president. Meanwhile, another person tried with Simpson cartoon characters Lenny and Carl – the latter who is black – with Lenny appearing to take preference. A third user even tried with dogs, resulting in a white dog in the prime preview position over a black dog.”
That last example suggests this may be an issue of highlight and shadow rather than biased training data, but either way is problematic. The company’s chief design officer posted one test he performed that seemed to counter the accusations, but acknowledges his experiment is far from conclusive. Twitter continues to investigate.
Cynthia Murrell, October 1, 2020
Predictive Analytics: Follow These Puffy Thought Bubbles
September 21, 2020
Predictive analytics is about mathematics; for instance, Bayesian confections and Markov doodling. The write up “Predictive Analytics: 4 Primary Aspects of Predictive Analytics” uses the bound phrase “predictive analytics” twice in one headline and cheerfully ignores the mathy reality of the approach.
Does this marshmallow approach make a difference? Yes, I believe it does. Consider this statement from the write up:
These predictive models can be used by enterprise marketers to more effectively develop predictions of future user behaviors based on the sourced historical data. These statistical models are growing as a result of the wide swaths of available current data as well as the advent of capable artificial intelligence and machine learning.
Okay, marketers. Predictive analytics are right in your wheelhouse. The assumption that “statistical models are growing” is interesting. The statistical models with which I am familiar require work to create, test, refine, and implement. Yep, work, mathy work.
The source of data is important. However, data have to be accurate or verifiable or have some attribute that tries to ensure that garbage in does not become the mode of operation. Unfortunately data remain a bit of a challenge. Do marketers know how to identify squishy data? Do marketers care? Yeah, sure they do in a meeting during which smartphone fiddling is taking place.
The idea of data utility is interesting. If one is analyzing nuclear fuel pool rod placement, it does help to have data relevant to that operation. But are marketers concerned about “data utility”? Once again, thumbtypers say, “Yes.” Then what? Acquire data from a third party and move on with life? It happens.
The thrill of “deep learning” is like the promise of spring. Everyone likes spring? Who remembers the problems? Progress is evident in the application of different smart software methods. However, there is a difference between saying “deep learning” or “machine learning” and making a particular application benefit from available tools, libraries, and methods. The whiz kids who used smart software to beat a human fighter pilot got the job done. The work required to achieve the digital victory was significant, took time, and was difficult. Very difficult. Marketers, were you on the team?
Finally, what’s the point of predictive analytics? Good question. For the article, the purpose of predictive analytics is to refine a guess-timate. And the math? Just use a smart solution, click and icon, and see the future.
Yikes, puffy thought bubbles.
Stephen E Arnold, September 21, 2020
Count Bayesie Speaks Truth
September 10, 2020
Navigate to “Why Bayesian Stats Needs More Monte Carlo Methods.” Each time I read an informed write up about the 18th century Presbyterian minister who could do some math, I think about a fellow who once aspired to be the Robert Maxwell of content management. Noble objective is it not?
That person grew apoplectic when I explained how Autonomy in the early 1990s was making use of mathematical procedures crafted in the 18th century. I wish I have made a TikTok video of his comical attempt to explain that a human or software system should not under any circumstances inject a data point that was speculative.
Well, my little innumeric content management person, get used to Bayes. Plus there’s another method at which you can rage and bay. Yep, Monte Carlo. If you were horrified by the good Reverend’s idea, wait until you did into Monte Carlo. Strapping these two stastical stallions to the buggy called predictive analytics is commonplace.
The write up closes poetically, which may be more in line with the fuzzy wuzzy discipline of content management:
It may be tempting to blame the complexity of the details of Bayesian methods, but it’s important to realize that when we are taught the beauty of calculus and analytical methods we are often limited to a relatively small set of problems that map well to the solutions of calc 101. When trying to solve real world problems mathematically complex problems pop up everywhere and analytical solutions either escape or fail us.
Net net: Use what matches the problem. Also, understand the methods. Key word: Understand.
Stephen E Arnold, September 10, 2020
Machine Learning Like A Psychic: Sounds Scientific for 2020
September 8, 2020
DarkCyber thinks most psychics are frauds. They are observers and manipulators of human behavior. They take people’s weaknesses and turn it into profit for themselves. In other words, they do not know the winning lottery numbers, they cannot predict stock options, and they cannot find missing pets.
Machine learning algorithms built on artificial intelligence, however, might have the “powers” psychics claim to have. EurekaAlert! Has a brand new: “Study: Machine Learning Can Predict Market Behavior.” Machine learning algorithms are smart, because they were programmed to find and interpret patterns. They can also assess how effective mathematical tools are predicting financial markets.
Cornell University researchers used a large dataset to determine if a machine learning algorithm could predict future financial events. It is a large task to undertake, because financial markets have tons of information and high volatility. Maureen O’Hara, the Robert W. Purcell Professor of Management at the SC Johnson College of Business said:
“ ‘Trying to estimate these sorts of things using standard techniques gets very tricky, because the databases are so big. The beauty of machine learning is that it’s a different way to analyze the data,’ O’Hara said. ‘The key thing we show in this paper is that in some cases, these microstructure features that attach to one contract are so powerful, they can predict the movements of other contracts. So we can pick up the patterns of how markets affect other markets, which is very difficult to do using standard tools.’”
Companies exist solely on the basis of understanding how financial markets work and they have developed their own machine learning algorithms for that very purpose. Cornell’s study used a random forest machine learning algorithm to examine these models using a dataset with 87 future contracts. The study used every single trade, tens of millions, for their analysis. They discovered that some of the variables worked, while others did not.
There are millions of datasets available since every trade has been recorded since 1980. Machine learning interprets this data and makes predictions, but it acts more like a black box. In other words, the algorithms predict patterns but it does not reveal the determinations.
Psychics have tried to predict the future for centuries and have failed. Machine learning algorithms are better at it, but they still are not 100% accurate. Predicting the future still remains consigned to fantasy and science fiction.
Whitney Grace, September 8, 2020
Predictive Policing: A Work in Progress or a Problem in Action?
September 2, 2020
Amid this year’s protests of police brutality, makers of crime-predicting software took the occasion to promote their products as a solution to racial bias in law enforcement. The Markup ponders, “Data-Informed Predictive Policing Was Heralded as Less Biased. Is It?” Writer Annie Gilbertson observes, as we did, that more than 1,400 mathematicians signed on to boycott predictive policing systems. She also describes problems discovered by researchers at New York University’s AI Now Institute:
“‘Police data is open to error by omission,’ [AI Now Director Rashida Richardson] said. Witnesses who distrust the police may be reluctant to report shots fired, and rape or domestic violence victims may never report their abusers. Because it is based on crime reports, the data fed into the software may be less an objective picture of crime than it is a mirror reflecting a given police department’s priorities. Law enforcement may crack down on minor property crime while hardly scratching the surface of white-collar criminal enterprises, for instance. Officers may intensify drug arrests around public housing while ignoring drug use on college campuses. Recently, Richardson and her colleagues Jason Schultz and Kate Crawford examined law enforcement agencies that use a variety of predictive programs. They looked at police departments, including in Chicago, New Orleans, and Maricopa County, Ariz., that have had problems with controversial policing practices, such as stop and frisk, or evidence of civil rights violations, including allegations of racial profiling. They found that since ‘these systems are built on data produced during documented periods of flawed, racially biased, and sometimes unlawful practices and policies,’ it raised ‘the risk of creating inaccurate, skewed, or systemically biased data.’”
The article also looks at a study from 2016 by the Royal Statistical Society. Researchers supplied PredPol’s algorithm with arrest data from Oakland California, a city where estimated drug use is spread fairly evenly throughout the city’s diverse areas. The software’s results would have had officers target Black neighborhoods at about twice the rate of white ones. The team emphasized the documented harm over-policing can cause. The write-up goes on to cover a few more studies on the subject, so navigate there for those details. Gilberston notes that concerns about these systems are so strong that police departments in at least two major cities, Chicago and Los Angeles, have decided against them. Will others follow suit?
Cynthia Murrell, September 2, 2020
-
- Subscribe to Beyond Search
Feature archive
News archive
-
