Instagram Reins in Trolls

July 21, 2017

Photo-sharing app Instagram has successfully implemented DeeText, a program that can successfully weed out nasty and spammy comments from people’s feeds.

Wired in an article titled Instagram Unleashes an AI System to Blast Away Nasty Comments says:

DeepText is based on recent advances in artificial intelligence, and a concept called word embeddings, which means it is designed to mimic the way language works in our brains.

DeepText initially was built by Facebook, Instagram’s parent company for preventing abusers, trolls, and spammers at bay. Buoyed by the success, it soon implemented on Instagram.

The development process was arduous wherein a large number of employees and contractors for months were teaching the DeepText engine how to identify abusers. This was achieved by telling the algorithm which word can be abusive based on its context.

At the moment, the tools are being tested and rolled out for a limited number of users in the US and are available only in English. It will be subsequently rolled out to other markets and languages.

Vishal Ingole, July 21, 2017

Darktrace Delivers Two Summer Sizzlers

July 17, 2017

Darktrace offers an enterprise immune system called Antigena. Based on the information gathered in the writing of the “Dark Web Notebook,” the system has a number of quite useful functions. The company’s remarkable technology can perform real time, in depth analyses of an insider’s online activities. Despite the summer downturn which sucks in many organizations, Darktrace has been active. First, the company secured an additional round of investment. This one is in the $75 million range. This brings the funding of the company to the neighborhood of $170 million, according to Crunchbase.

Details about the deal appear in this Outlook Series write up. I noted this statement:

The cyber security firm has raised a $75 million Series D financing round led by Insight Venture Partners, with participation from existing investors Summit Partners, KKR and TenEleven Ventures.

On another front, Darktrace has entered into a partnership with CITIC. This outfit plans to bring “next-generation cyber defense to businesses across Asia Pacific.” Not familiar with CITIC? You might want to refresh your memory bank. Beyond Search believes that this tie up may open the China market for Darktrace. If it does, Darktrace is likely to emerge as one of the top two or three cyber security firms in the world before the autumn leaves begin to fall.

Here in Harrod’s Creek we think about the promise of Darktrace against a background of erratic financial performance from Hewlett Packard. As you may recall, one of the spark plugs for Darktrace is Dr. Michael Lynch, the founder of Autonomy. HP bought Autonomy and found that its management culture was an antigen to its $11 billion investment. It is possible to search far and wide for an HP initiative which has delivered the type of financial lift that Darktrace has experienced.

Information about Darktrace is at A profile about this company appears in the Dark Web Notebook in the company of IBM Analyst’s Notebook, Google/In-Q-Tel Recorded Future, and Palantir Technologies Gotham. You can get these profile at this link:

Stephen E Arnold, July 17, 2107

Can an Algorithm Tame Misinformation Online?

June 23, 2017

UCLA researchers are working on an algorithmic solution to the “fake news” problem, we learn from the article, “Algorithm Reads Millions of Posts on Parenting Sites in Bid to Understand Online Misinformation” at TechRadar. Okay, it’s actually indexing and text analysis, not “reading,” but we get the idea. Reporter Duncan Geere tells us:

There’s a special logic to the flow of posts on a forum or message board, one that’s easy to parse by someone who’s spent a lot of time on them but kinda hard to understand for those who haven’t. Researchers at UCLA are working on teaching computers to understand these structured narratives within chronological posts on the web, in an attempt to get a better grasp of how humans think and communicate online.

Researchers used the hot topic of vaccinations, as discussed on two parenting forums, as their test case. Through an examination of nearly 2 million posts, the algorithm was able to come to accurate conclusions, or “narrative framework.” Geere writes:

While this study was targeted at conversations around vaccination, the researchers say the same principles could be applied to any topic. Down the line, they hope it could allow for false narratives to be identified as they develop and countered by targeted messaging.

The phrase “down the line” is incredibly vague, but the sooner the better, we say (though we wonder exactly what form this “targeted messaging” will take). The original study can be found here at eHealth publisher JMIR Publications.

Cynthia Murrell, June 23, 2017


Algorithms Are Getting Smarter at Identifying Human Behavior

June 19, 2017

Algorithm deployed by large tech firms are better at understanding human behaviors, reveals former Google data scientist.

In an article published by Business Insider titled A Former Google Data Scientist Explains Why Netflix Knows You Better Than You Know Yourself, Seth Stephens-Davidowitz says:

Many gyms have learned to harness the power of people’s over-optimism. Specifically, he said, “they’ve figured out you can get people to buy monthly passes or annual passes, even though they’re not going to use the gym nearly enough to warrant this purchase.

Companies like Netflix use this to their benefit. For instance, during initial years, Netflix used to encourage users to create playlists. However, most users ended up watching the same run of the mill content. Netflix thus made changes and started recommending content that was similar to their content watching habits. It only proves one thing, algorithms are getting smarter at understanding and predicting human behaviors, and that is both good and bad.

Vishal Ingole,  June 19, 2017

The Power of Context in Advertising

June 9, 2017

How’s it going with those ad-and-query matching algorithms? The Washington Post reports, “For Advertisers, Algorithms Can Lead to Unexpected Exposure on Sites Spewing Hate.” Readers may recall that earlier this year, several prominent companies pulled their advertisements from Google’s AdSense after they found them sharing space with objectionable content. Writers Elizabeth Dwoskin and Craig Timberg cite an investigation by their paper, which found the problem is widespread. (See the article for specifics.) How did we get here? The article explains:

The problem has emerged as Web advertising strategies have evolved. Advertisers sometimes choose to place their ads on particular sites — or avoid sites they dislike — but a growing share of advertising budgets go to what the industry calls ‘programmatic’ buys. These ads are aimed at people whose demographic or consumer profile is receptive to a marketing message, no matter where they browse on the Internet. Algorithms decide where to place ads, based on people’s prior Web usage, across vastly different types of sites.

The technology companies behind ad networks have slowly begun to address the issue, but warn it won’t be easy to solve. They say their algorithms struggle to distinguish between content that is truly offensive and language that is not offensive in context. For example, it can be hard for computers to determine the difference between the use of a racial slur on a white-supremacy site and a website about history.

To further complicate the issue, companies employing these algorithms want nothing to do with becoming “arbiters of speech.” After all, not every case is so simple as a post sporting a blatant slur in the headline; the space between hate speech and thoughtful criticism is more of a gradient than a line. Google. Facebook, et al may not have signed up for this role, but the problem is the direct consequence of the algorithmic ad-placing model. Whether on this issue, the scourge of fake news, or the unwitting promotion of counterfeit goods, tech companies must find ways to correct the wide-spread consequences of their revenue strategies.

Cynthia Murrell, June 9, 2017

Helping Machines Decode the World of Online Content

June 5, 2017

With voice search poised to overtake conventional search, startups like WordLift are creating an AI-based algorithm that can help machines understand content created by humans in a better way.

The Next Web in an article titled Wordlift Is Helping Robots Understand What Online Articles Are Really About says:

The evolution of today’s search engines and the rapid adoption of personal assistants (PAs) – capable of understanding user intent and behaviors through available data – require an upgrade of the existing editorial workflow for bloggers, independent news providers, and content marketers.

Voice activated search assistants rely on Metadata for understanding what the content is about. Moreover, metadata alone is unable to tell the AI what is the user intent. WordLift intends to solve this problem by applying advanced AI for understanding the content and make it voice search engine friendly. Structured data, understanding of textual content are some of the strategies WordLift will use to make the content voice search engine friendly.

Vishal Ingole, June 5, 2017

Linguistic Analytics Translate Doctor Scribbles

May 31, 2017

Healthcare is one of the industries that people imagine can be revolutionized by new technology.  Digital electronic medical records, faster, more accurate diagnostic tools, and doctors having the ability to digest piles of data in minutes are some of the newest and best advances in medicine.  Despite all of these wonderful improvements, healthcare still lags behind other fields transforming their big data into actionable, usable data.  Inside Big Data shares the article, “How NLP Can Help Healthcare ‘Catchup’” discusses how natural language processing can help the healthcare industry make more effective use of their resources.

The reason healthcare lags behind other fields is that most of their data is unstructured:

This large realm of unstructured data includes qualitative information that contributes indispensable context in many different reports in the EHR, such as outside lab results, radiology images, pathology reports, patient feedback and other clinical reports. When combined with claims data this mix of data provides the raw material for healthcare payers and health systems to perform analytics. Outside the clinical setting, patient-reported outcomes can be hugely valuable, especially for life science companies seeking to understand the long-term efficacy and safety of therapeutic products across a wide population.

Natural language processing relies on linguistic algorithms to identify key meanings in unstructured data.  When meaning is given to unstructured data, then it can be inserted into machine learning algorithms.  Bitext’s computational linguistics platform does the same with its sentimental analysis algorithm. Healthcare information is never black and white like data in other industries.  While the unstructured data is different from patient to patient, there are similarities and NLP helps the machine learning tools learn how to quantify what was once-unquantifiable.

Whitney Grace, May 31, 2017

SEO Adapts to Rapidly Changing Algorithms

May 30, 2017

When we ponder the future of search, we consider factors like the rise of “smart” searching—systems that deliver what they know the user wants, instead of what the user wants—and how facial recognition search is progressing. Others look from different angles, though, like the business-oriented Inc., which shares the post, “What is the Future of Search?” Citing SEO expert Baruch Labunski, writer Drew Hendricks looks at how rapid changes to search engines’ ranking algorithms affect search-engine-optimization marketing efforts.

First, companies must realize that it is now essential that their sites play well with mobile devices; Google is making mobile indexing a priority. We learn that the rise of virtual assistants raises the stakes—voice-controlled searches only return the very first search result. (A reason, in my opinion, to use them sparingly for online searches.) The article pays the most attention, though, to addressing local search. Hendricks advises:

By combining the highly specific locational data that’s available from consumers searching on mobile, alongside Google’s already in-progress goal of customizing results by location for all users, positioning your brand to those who are physically near you will become crucial in 2017. …


Our jobs as brand managers and promoters will continue to become more complicated as time passes. The days of search engine algorithms filtering by obvious data points, or being easily manipulated, are over. The new fact of search engine optimization is appealing to your immediate markets – those around you and those who are searching directly for your product.

Listing one’s location(s) on myriad review sites and Google Places and placing the address on the company website are advised. The piece concludes by reassuring marketers that, as long as they make careful choices, they can successfully navigate the rapid changes to Google and other online search engines.

Cynthia Murrell, May 30, 2017

The World of Artificial Intelligence: Solving the Color Name Problem

May 24, 2017

I eagerly wait the accuracy and precision of artificial intelligence in my everyday life. My wife has presented me with three color chips. Each chip has a name; for example, almond, parchment, and old ivory. She asks me, “Which do you prefer?”

I reply, “Which one do you like?”

The reason is that the names and the colors make zero sense to me. The color is white and the differences among them are not discernable to me. White is pretty much white to me.

I read “Turdly? Stoner Blue? Stanky Bean? Never Let an AI Name Colors.” The main idea is that a research scientist “taught” smart software to name colors. The results were encouraging. Almond? Parchment? Old ivory. Dull. I simply do not relate to odd ball, metaphorical names.

However, the smart software is on my wave length. Why fool around with poetry when there are AI identified names to make colors come alive; for example:

  1. Tired of the weird names for a mixture of black and white? Go with “Horble Gray.” (Does an AI program know the difference between “horrid” and “horble”?)
  2. Want something to go with that snappy new sofa? Why not select both carpet and trim in “Golder cream”? Sounds good enough to eat, right?
  3. Looking for the perfect highlight for one’s non binary child? I would select without hesitation “Burf Pink”. Descriptive and only one vowel away from my favorite word used to describe AI software, “barf.”

Believe me. I wanted to highlight “Bank Butt” and “Dorkwood.” But I was not sure if these colors were the work of a person with an MFA, a product of the Onion’s editorial team, or just another one of these “real news” items which inform and delight.

I think I am slipping into an AI “Clardic Fug.” Others may embrace “Stoner Blue.” (See I did not reference “Turdly Brown,” you “Stanky Bean.”

Stephen E Arnold, May 24, 2017

New AI on Personal Digital Assistant Horizon

May 22, 2017

Computer scientists at Princeton University have developed a technology that allows the user to fully edit voice recordings using an intelligent algorithm.

Science Daily in a report titled Technology Edits Voices Like Text says that:

The software, named VoCo, provides an easy means to add or replace a word in an audio recording of a human voice by editing a transcript of the recording. New words are automatically synthesized in the speaker’s voice even if they don’t appear anywhere else in the recording.

The system is capable of recreating voice of the user using an intelligent algorithm. This makes adding words to pre-recorded audio recordings easier. The same technology can also be used to create a custom robotic voice for digital personal assistants.

Currently available audio editing software are capable of snipping and patching small segments of a recording and cannot add non-existent words. Algorithm of VoCo after analyzing the entire recording is able to synthesize any word without difficulty. At this speed, do we see the current breed of rock and pop artists disappearing?

Vishol Ingole, May 22, 2017

Next Page »

  • Archives

  • Recent Posts

  • Meta