RobotTaxis to Thrive in Vegas

 Where Else?  Good place for live tests, proof of use in complex use, Uber complexity ...  On to LA they say ... 

Uber and Motional's robotaxis arrive in Las Vegas

It marks the first time public riders and access the autonomous vehicles using Uber.

Steve Dent|@stevetdent|December 7, 2022 6:00 AM

Uber has launched public robotaxi rides in Las Vegas using Motional's Hyundai Ioniq 5 autonomous EVs with the aim of offering a full driverless service to the public in 2023. It will eventually expand to Los Angeles, where the two companies have been testing autonomous Uber Eats deliveries since May 2022. It's all part of a 10-year agreement between Uber and Motional to offer autonomous ride-hailing and deliveries. 

The taxi rides will be monitored by safety drivers, with the goal of launching a fully driverless service to the public in 2023. "Today, Motional becomes the first AV company to conduct all-electric autonomous rides on the Uber network for public passengers," said Motional VP Akshay Jaising. (Uber offered "autonomous" taxi rides using its own self-driving tech back in 2016, but riders were accompanied by engineers ready to take the wheel.) 

Lyft also partners with Motional, a joint venture between Hyundai and Aptiv. It beat Uber to the punch by launching Motional rides in Las Vegas in August this year with safety drivers on board. Lyft has also said it will ditch those drivers and offer true autonomous service by next year. 

Don't be surprised if that timeline changes, though. Apart from Motional, only Alphabet division Waymo and GM's Cruise are offering true driverless services at a reasonably large scale. The Waymo One service is operating in Phoenix and San Francisco, while Cruise rides are currently limited to San Francisco. Both operate only in specific areas of cities (which can be mapped out in great detail) and some vehicles still use safety riders.

Uber co-founder Travis Kalanick once said that the his company had to be "tied for first, at least" in the race to offer true driverless rides. Since both Uber and Lyft rely on Motional for self-driving tech, that's exactly what might happen. ... ' 

Rephrasing Communications in Context

 Advances in assistant communications.  In Google Blog,  As always context is very important. 

Contextual Rephrasing in Google Assistant

Tuesday, May 17, 2022

Posted by Aurelien Boffy, Senior Staff Software Engineer, and Roberto Pieraccini, Engineering Director, Google Assistant

When people converse with one another, context and references play a critical role in driving their conversation more efficiently. For instance, if one asks the question “Who wrote Romeo and Juliet?” and, after receiving an answer, asks “Where was he born?”, it is clear that ‘he’ is referring to William Shakespeare without the need to explicitly mention him. Or if someone mentions “python” in a sentence, one can use the context from the conversation to determine whether they are referring to a type of snake or a computer language. If a virtual assistant cannot robustly handle context and references, users would be required to adapt to the limitation of the technology by repeating previously shared contextual information in their follow-up queries to ensure that the assistant understands their requests and can provide relevant answers.

In this post, we present a technology currently deployed on Google Assistant that allows users to speak in a natural manner when referencing context that was defined in previous queries and answers. The technology, based on the latest machine learning (ML) advances, rephrases a user’s follow-up query to explicitly mention the missing contextual information, thus enabling it to be answered as a stand-alone query. While Assistant considers many types of context for interpreting the user input, in this post we are focusing on short-term conversation history.

Context Handling by Rephrasing

One of the approaches taken by Assistant to understand contextual queries is to detect if an input utterance is referring to previous context and then rephrase it internally to explicitly include the missing information. Following on from the previous example in which the user asked who wrote Romeo and Juliet, one may ask follow-up questions like “When?”. Assistant recognizes that this question is referring to both the subject (Romeo and Juliet) and answer from the previous query (William Shakespeare) and can rephrase “When?” to “When did William Shakespeare write Romeo and Juliet?”

While there are other ways to handle context, for instance, by applying rules directly to symbolic representations of the meaning of queries, like intents and arguments, the advantage of the rephrasing approach is that it operates horizontally at the string level across any query answering, parsing, or action fulfillment module.  ... '

Context In Machine Intelligence

I often make this point ... context is everything ... in fact I like to make the expansion of this as well:  'Context is everything', and its always changing"      Its like how we use all 'intelligence' that  is presented in a context, and included in that context are usually a number of uncontrolled time dimensions, like history and customer status and  other domain considerations, random context elements, etc  

Advancing Machine Intelligence: Why Context Is Everything  in TowardsDataScience  By Gadi Singer May 10

Most of us have heard the phrase, “Image is everything.” But when it comes to taking AI to the next level, it’s context that is everything.

Contextual awareness embodies all the subtle nuances of human learning. It is the ‘who’, ‘why’, ‘when’, and ‘why’ that inform human decisions and behavior. Without context, the current foundation models are destined to spin their wheels and ultimately interrupt the trajectory of expectation for AI to improve our lives.

This blog will discuss the significance of context in ML, and how late binding context could raise the bar on machine enlightenment.

Why Context Matters

Context is so deeply embedded in human learning that it is easy to overlook the critical role it plays in how we respond to a given situation. To illustrate this point, consider a conversation between two people that begins with a simple question: How is Grandma?

In a real-world conversation, this simple query could elicit any number of potential responses depending on contextual factors, including time, circumstance, relationship, etc.:

Fig 1. A proper answer to “How’s Grandma?” is highly context-dependent. 

The question illustrates how the human mind can track and take into account a vast amount of contextual information, even subtle humor, to return a relevant response. This ability to fluidly adapt to a variety of often subtle contexts is well beyond the reach of modern AI systems.

To grasp the significance of this deficit in machine learning, consider the development of reinforcement learning (RL)-based autonomous agents and robots. Despite the hype and success that RL-based architectures have had in simulated game environments like Dota 2 and StarCraft II, even purely gaming environments like NetHack pose a formidable obstacle to current RL systems due to the highly conditional nature and complexity of policies that are required to win the game. Similarly, as noted in many recent works, autonomous robots have miles to go before they can interact with previously unseen physical environments without the need of a serious engineering effort to either simulate the correct type of environment prior to deployment, or to harden the learned policy.

Current ML and Handling of Contextual Queries

With some notable exceptions, most ML models incorporate very limited context of a specific query, relying primarily on the generic context provided by the dataset that the model is trained or fine-tuned on. Such models also raise significant concerns about bias which makes them less suited for use in many business, healthcare, and other critical applications. Even state-of-the-art models like D3ST used in voice assistant AI applications require manually creating descriptions of schemata or ontologies with possible intents and actions that the model needs to identify context. While this involves a relatively minimal level of handcrafting, it nonetheless means that an explicit human input is required every time the context of the task is to be updated.

That’s not to say there haven’t been significant developments in context awareness for machine learning models. GPT-3, a famous large language model from the OpenAI team, has been used to generate full articles that rival human composition — a task that requires keeping track of at least local context. The Pathways Language Model (PaLM), introduced by Google in April 2022, demonstrates even greater capability, including the ability to understand conceptual combinations in appropriate contexts to answer complex queries.  ... 

On Investment in Explainable AI

Needs always to be explained in its usage and risk contexts.

AI Is Explaining Itself to Humans. It's Paying Off.   By Reuters, April 15, 2022, Comments

Startups and major technology companies are investing heavily in explainable artificial intelligence (XAI), as U.S. and EU regulators campaign for fairness and transparency in automated decision-making.  XAI advocates say it has helped make AI more effective in fields such as healthcare and sales.

Microsoft saw its LinkedIn subscription revenue increase 8% after providing its sales team with CrystalCandle software, which identifies clients potentially at risk of cancellation, while explaining its reasoning.

Skeptics say an AI’s explanations of why it made the predictions it did are still too unreliable. LinkedIn says an algorithm's integrity cannot be judged without understanding its reasoning, while tools like CrystalCandle, for example, could help physicians learn why AI predicts someone is at greater risk of disease.

From Reuters   

Can AlphaZero Solving Problems and Rule Variations

 Sharing variations of a problem.   All problems have variations, which specify their context, can this give us a hint for other solutions?

Reimagining Chess with AlphaZero

By Nenad Tomašev, Ulrich Paquet, Demis Hassabis, Vladimir Kramnik

Communications of the ACM, February 2022, Vol. 65 No. 2, Pages 60-66 10.1145/3460349

Modern chess is the culmination of centuries of experience, as well as an evolutionary sequence of rule adjustments from its inception in the 6th century to the modern rules we know today.17 While classical chess still captivates the minds of millions of players worldwide, the game is anything but static. Many variants have been proposed and played over the years by enthusiasts and theorists.8,20 They continue the evolutionary cycle by altering the board, piece placement, or the rules—offering players "something subtle, sparkling, or amusing which cannot be done in ordinary chess."1

Key Insights

Technological progress is the new driver of the evolutionary cycle. Chess engines increase in strength, and players have access to millions of computer games and volumes of opening theory. Consequently, the number of decisive games in super-tournaments has declined, and it takes longer for players to move from home preparation to playing original moves on the board.14 While classical chess remains a fascinating game and is unlikely to ever fall out of fashion, alternative variants provide an avenue for more creative play. In Fischer random chess, the brainchild of former world champion Bobby Fischer, the initial position is randomized to counter the dominance of opening preparation in a game.7 One could consider not only entirely new ideas, but also reassess some of the newer additions to the game. For example, the "castling" move was only introduced in its current form in the 17th century. What would chess have been like had castling not been incorporated into the rules? Without recourse to repeating history, we reimagine chess and address such questions in silico with AlphaZero.25

AlphaZero is a system that can learn superhuman chess strategies from scratch without any human supervision.19,22 It represents a milestone in artificial intelligence (AI), a field that has ventured down the corridors of chess more than once in search of challenges and inspiration. Throughout the history of computer chess, the focus was on creating systems that could spar with top human players over the board.3 Computer chess has progressed steadily since the 1950s, with better-tuned evaluation functions and enhanced search algorithms deployed on increasingly more computational resources.2,3,9,13,18,24 Alan Turing already envisioned more in 1953 by asking, "Could one make a machine to play chess, and to improve its play, game by game, profiting from its experience?"27 Unlike its predecessors, AlphaZero learns its policy from scratch from repeated self-play games, answering the second part of Turing's question. The result is a unique approach to playing classical chess22 and a new era in the development of chess engines, as spear-headed by Leela Chess Zero.15

AlphaZero's ability to continually improve its understanding of the game, and reach superhuman playing strength in classical chess and Go,25 lends itself to the question of assessing chess variants and potential variants of other board games in the future. Provided only with the implementation of the rules, it is possible to effectively simulate decades of human experience in a day, opening a window into top-level play of each variant. In doing so, computer chess completes the circle, from the early days of pitting man vs. machine to a collaborative present of man with machine, where AI can empower players to explore what chess is and what it could become.11

Back to Top  ... ' 

Building AIs that Feels Emotion, Empathy

Work by Microsoft's  'Human Understanding and Empathy' group.  New to me but likely useful as a component of other sensory measures.   In context, as always.  

Building an AI That Feels

AI systems with emotional intelligence could learn faster and be more helpful

By Mary Czerwinski, Javier Hernandez and Daniel McDuff

In the past year, have you found yourself under stress? Have you ever wished for help coping? Imagine if, throughout the pandemic, you’d had a virtual therapist powered by an artificial intelligence (AI) system, an entity that empathized with you and gradually got to know your moods and behaviors. Therapy is just one area where we think an AI system that can recognize and interpret emotions could offer great benefits to people.

Our team hails from Microsoft’s Human Understanding and Empathy group, where our mission is to imbue technology with emotional intelligence. Why? With that quality, AI can better understand its users, more effectively communicate with them, and improve their interactions with technology. The effort to produce emotionally intelligent AI builds on work in psychology, neuroscience, human-computer interaction, linguistics, electrical engineering, and machine learning.

Lately, we’ve been considering how we could improve AI voice assistants such as Alexa and Siri, which many people now use as everyday aides. We anticipate that they’ll soon be deployed in cars, hospitals, stores, schools, and more, where they’ll enable more personalized and meaningful interactions with technology. But to achieve their potential, such voice assistants will require a major boost from the field of affective computing. That term, coined by MIT professor Rosalind W. Picard in a 1997 book by the same name, refers to technology that can sense, understand, and even simulate human emotions. Voice assistants that feature emotional intelligence should be more natural and efficient than those that do not.

Consider how such an AI agent could help a person who’s feeling overwhelmed by stress. Currently, the best option might be to see a real human psychologist who, over a series of costly consultations, would discuss the situation and teach relevant stress-management skills. During the sessions, the therapist would continually evaluate the person’s responses and use that information to shape what’s discussed, adapting both content and presentation in an effort to ensure the best outcome.

While this treatment is arguably the best existing therapy, and while technology is still far from being able to replicate that experience, it’s not ideal for some. For example, certain people feel uncomfortable discussing their feelings with therapists, and some find the process stigmatizing or time-consuming. An AI therapist could provide them with an alternative avenue for support, while also conducting more frequent and personalized assessments. One recent review article found that 1 billion people globally are affected by mental and addictive disorders; a scalable solution such as a virtual counselor could be a huge boon. .... " 

Explanations and Contexts

An example of the need for explain-ability. Like in a conversation with a human, we could want the option of getting an explanation of a solution.  But the nature of an explanation does often depend on context. Is is for management or an engineer?  Is it for a current set of data or a generalization?    Does it depend on some regulation or special constraints?  Context is often key.  Often occurred in our work. 

Researchers Develop 'Explainable' Algorithm

University of Toronto (Canada), Matthew Tierney, March 31, 2021

An "explainable" artificial intelligence (XAI) algorithm developed by researchers at Canada's University of Toronto (U of T) and LG AI Research was designed to find and fix defects in display screens. XAI addresses issues with the "black box" approach of machine learning strategies, in which the artificial intelligence makes decisions entirely on its own. With XAI's "glass box" approach, XAI algorithms are run simultaneously with traditional algorithms to audit the validity and level of their learning performance, perform debugging, and identify training efficiencies. U of T's Mahesh Sudhakar said LG "had an existing [machine learning] model that identified defective parts in LG products with displays, and our task was to improve the accuracy of the high-resolution heat maps of possible defects while maintaining an acceptable run time." The new XAI algorithm, Semantic Input Sampling for Explanation (SISE), outperformed comparable approaches on industry benchmarks.... ' 

Building Multilingual, Multipurpose, Multi Context Wikipedias

A long time user, and supporter of Wikipedia.  Have run into the problem that this describes,   Posts in different languages in the WP vary in their knowledge content.  In fact in too many examples,  there may be useful, detailed posts in one language, but the same topic is uncovered in another language. Within 50 million articles in 300 languages.   We ran into some related problems when we scoped out a wikipedia for a large corporation.  So I expanded the title of this piece, to include other aspects. But the article referenced below uses a 'language' knowledge problem example. Sharing knowledge.   Tough problem, let me know if you solve it effectively.

Building a Multilingual Wikipedia,   By Denny Vrandečić  in CACM

Communications of the ACM, April 2021, Vol. 64 No. 4, Pages 38-41  10.1145/3425778

Wikipedia has more than 50 million articles in approximately 300 languages. The content in these languages is independently created and maintained. The knowledge in Wikipedia is very unevenly distributed over the languages: some languages have more than a million articles, but more than 50 languages have only a few hundred articles or less. More importantly, also the number of contributors is very unevenly distributed: English Wikipedia has more than 418,000 contributors, the second-most active one, Spanish, drops down to 90,000. More than half of language editions have fewer than 10 contributors doing more than four edits per month. To assume that fewer than 10 active contributors can write and maintain a comprehensive encyclopedia in their spare time is optimistic at best.

In order to close these knowledge gaps we are building a multilingual Wikipedia where content is created only once but made available in all languages. The multilingual Wikipedia has two main components: Abstract Wikipedia where the content is created and maintained in a language-independent notation, and Wikifunctions, a project to create, catalog, and maintain functions. For the multilingual Wikipedia, the most important function is one that takes content from Abstract Wikipedia and renders it in natural language, which in turn gets integrated into Wikipedia proper.

This will considerably reduce the effort required to create a comprehensive and maintain a current encyclopedia in many languages. It will allow more people to share more knowledge in more languages than ever before. It will be particularly useful for under-served languages, providing an important way to help improve education and ready access to knowledge in many countries. ... "

Search Not Completely Solved in Context

Indeed it works very well.  But we all experience examples where it falls flat.  Often because it does not know the context of its use.   We cheerfully act as the intelligent editor of the process, unless it induces risk or chaos or takes far more time than we expect. 

Think Search Is Solved? Think Again

Alex Woodie in DataNami

Search is one of the oldest technologies around. Ever since the dawn of the World Wide Web, a search engine has been the portal through which we obtain information. The search for a better search engine index kick started the Hadoop craze, and it continues to drive Google to push the limits of technology. But don’t for a second think that search has been solved.

“Who said it’s solved?” barked Coveo’s Director of AI Ciro Greco in a recent interview with Datanami. “Search is far from being solved. It’s the hardest thing we do. It’s the hardest thing everybody does.”

Coveo is one of a handful of companies building the next generation of search engines, although even calling it that seems to be a disservice to someone. Forrester calls the field “Cognitive Search,” thanks to the abundance of machine learning, natural language understanding, and deep learning embedded in these products. Gartner, in turn, calls them “Insight Engines,” because they deliver more context than mere search engines.

“In contrast to search engines that provide links to original source materials such as documents and videos,” Gartner analysts write in a September 2019 Magic Quadrant report, “insight engines can also provide contextual information about the fact or entity in question.”

Today’s enterprise search engines use the index as the starting point for surfacing insights. But beyond just returning information based on the degree to which an entered term matches a predefined keyword stored in an index, modern search engines bring other data and technology to the party, including text analytics and machine learning technology that try to predict what the user is trying to find.

For Greco, the nature of search itself opens up such a rich field for exploration that it could never possibly be perfectly solved. For example, if people could always be relied upon to enter the perfect search term, there wouldn’t be much of a need for more elaborate search technology. But, of course, we’re all human, and so we can’t be relied upon to do that.

“If you go in a website because you have a problem with a product and you want to use a search engine to find the right information in your knowledge base, what are you going to do?” he asks. “Do you know exactly what you’re looking for? No, because you have a problem right. Do you know exactly where that thing is stored and how it is expressed in the documentation that this company has? Obviously not.

“So what you’re trying to do when you do search is actually you’re trying to the best of your knowledge to guess the intent of the human being through different layers that separate you from this person,” he continues. “It is as hard as trying to understand the map of New York City from the people that go in and out from JFK, if that is the little information that you have.”  ... '

Machine Learning Underspecification

 Yes, often a problem, often  mainly for lack of sufficient data in context.   Seems Google know this too.  Key issue.  Credibility/value is the key.  Note COVID example.  Technical. 

Googlers Speak Out on the Scourge of ML Underspecification

Oliver Peckham    in Datanami

A few days ago, 40 authors (all but a handful hailing from Google) published a 59-page paper. https://arxiv.org/pdf/2011.03395.pdf   The topic at hand: why so many machine learning models, borne out by internal testing, proceed to then fail spectacularly in real-world applications. The answer, the Googlers say, is underspecification – a blight on machine learning that, they stress, requires substantive solutions.

“An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain,” they write. In plain language: an underspecified model can think of a bunch of reasonably accurate explanations for why a dataset looks the way it does. The problem comes in when researchers assume that all of those explanations are equivalently valid based solely on the model’s training results, without accounting for real-world factors that may have escaped the model’s training process. In those situations, the authors say, “ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains.”

By way of illustration, the Googlers highlight examples spanning “computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics.” In epidemiology, for instance, they discuss how early data from an epidemic (such as the COVID-19 pandemic) is easily explained by a variety of models that do not substantively account for major factors – such as the gradually diminishing number of susceptible people in an area as an epidemic infects (and then renders immune) larger and larger portions of the populace.

“Importantly, during the early stages of an epidemic … the parameters of the model are underspecified by this training task,” they write. “This is because, at this stage, the number of susceptible is approximately constant at the total population size (N), and the number of infections grows approximately exponentially.”

As a result, they say, “arbitrary choices in the learning process” determine which parameters are deemed most predictive by the model, despite different models predicting “peak infection numbers, for example, that are orders of magnitude apart.” .... '

The paper, titled “Underspecification Presents Challenges for Credibility in Modern Machine Learning,”

Voice Interoperability Initiative: Combining Assistant Contexts?

Had seen this mentioned before, but had not completely understood its implications.  Considerable piece in Venturebeat.    Overall implications for ultimate intelligence of an assistant are unclear to me,but this takes an interesting step. Still think that intelligent interaction based on context is most important.   But if we can make multiple assistants collaborate, say based on particular focused context?     Looking further.    Have some other docs you can point me to?  See intro and key links below:

Facebook and Dolby join Amazon’s Voice Interoperability Initiative   By Kyle Wiggers / VentureBeat:

Amazon's Voice Interoperability Initiative, which lets multiple assistants work on a single device, adds 38 new members including Facebook, Dolby, and Xiaomi  —  Last September, Amazon unveiled the Voice Interoperability Initiative, a program aimed at ensuring voice-enabled products …
Last September, Amazon unveiled the Voice Interoperability Initiative, a program aimed at ensuring voice-enabled products like smart speakers and displays allow users to choose among multiple voice assistants. Today, the company announced the addition of 38 new members including Dolby, Facebook, Garmin, and Xiaomi to the initiative, bringing the total number of member companies to 77. (Google remains conspicuously absent from the list.) To mark the milestone, Amazon published what it’s calling the Multi-Agent design guide, a whitepaper outlining design recommendations Voice Interoperability Initiative members should use in building multi-assistant products.

The Voice Interoperability Initiative is organized around four core principles, the first of which is developing voice services that work “seamlessly” with others while ostensibly preserving privacy. (Amazon in particular has a spotty track record when it comes to voice privacy, but the company claims to have made strides in recent months.) Members seek to build devices that ship with multiple assistants as they work to accelerate conversational AI research, with the goal of enabling users to leverage the capabilities afforded by Alexa, Cortana, and other services on a single platform.  .... " 

Lowe's Goes Virtual for Pro Home Improvement

Most intrigued about how the knowledge is being stored, delivered,utilized.      There are different levels of expertise embedded in 'Pro', so will this context be included?   Ultimately essentially.

Lowe’s ‘virtually’ goes on the job for home improvement pros   by George Anderson in Retailwire

Lowe’s is introducing a new tool that will enable carpenters, electricians, plumbers and other construction professionals to meet with customers to discuss projects without having to go to their homes.

Lowe’s for Pros JobSIGHT makes use of video, computer vision and augmented reality tech to help pros evaluate projects so they can provide quotes to consumers on a wide variety of repair and home improvement projects. Pros using the tool chat directly with homeowners and are able to conduct tasks, such as determining product serial numbers and product details. They can use an on-screen laser pointer and augmented reality quick-draw tools to work through the consultation with homeowners. When the virtual meeting is complete, pros receive a one-page summary including video and audio, hi-res photos and notes for follow-up.

Lowe’s is making Pros JobSIGHT free to trade professionals through Oct. 31. Those who sign up for the program also save five percent on the chain’s everyday prices and are eligible for zero-interest purchases using their business accounts with Lowe’s. Extended payment terms are also available.  ... '

Seeking Better Chatbots with Contextual Memory

Another example of looking at memory to establish context understanding for a chatbot

Deploying more conversational chatbots
Startup Posh has created chatbots that use “conversational memory” to have more natural exchanges.

Zach Winn | MIT News Office

The comedian Bill Burr has said he refuses to call into automated customer service lines for fear that, years later on his death bed, all he’ll be able to think about are the moments he wasted dealing with chatbots.

Indeed, the frustrating experience of trying to complete even the most straightforward task through an automated customer service line is enough to make anyone question the purpose of life.

Now the startup Posh is trying to make conversations with chatbots more natural and less maddening. It’s accomplishing this with an artificial intelligence-powered system that uses “conversational memory” to help users complete tasks.

“We noticed bots in general would take what the user said at face value, without connecting the dots of what was said before in the conversation,” says Posh co-founder and CEO Karan Kashyap ’17, SM ’17. “If you think about your conversations with humans, especially in places like banks with tellers or in customer service, what you said in the past is very important, so we focused on making bots more humanlike by giving them the ability to remember historical information in a conversation.”

Posh’s chatbots are currently used by over a dozen credit unions across voice- and text-based channels. The well-defined customer base has allowed the company to train its system on only the most relevant data, improving performance.

The founders plan to gradually partner with companies in other sectors to gather industry-specific data and expand the use of their system without compromising performance. Down the line, Kashyap and Posh co-founder and CTO Matt McEachern ’17, SM ’18 plan to provide their chatbots as a platform for developers to build on.   .... "

Tesla Autopilot Detecting Traffic Lights

An example of systems including more environmental context for making decisions, ultimately essential.

Tesla's Autopilot could soon detect traffic lights

A video shows a Tesla stopping autonomously at a red light.   ....

By Christine Fisher, @cfisherwrites in Engadget

Google's Summarization Performance: Pegasus

A kind of AI.   Summarization is useful and powerful concept.  But consider that summarization also exists in a context.   Its output is only useful in a particular context, and that exists based  also on the requester.  And that can also change over time, location, requesters current goals, etc.    And influenced by the metadata involved with its construction.   Still a very useful step forward.

Google Brain’s AI achieves state-of-the-art text summarization performance
Kyle Wiggers in Venturebeat

Summarizing text is a task at which machine learning algorithms are improving, as evidenced by a recent paper published by Microsoft. That’s that’s good news — automatic summarization systems promise to cut down on the amount message-reading done by enterprise workers, which one survey estimates amounts to 2.6 hours each day.

Not to be outdone, a Google Brain and Imperial College London team built a system — Pre-retraining with Extracted Gap-sentences for Abstractive SUmmarization Sequence-to-sequence, or PEGASUS — that leverages Google’s Transformers architecture combined with pre-training objectives tailored for abstractive text generation. They say it achieves state-of-the-art results in 12 summarization tasks spanning news, science, stories, instructions, emails, patents, and legislative bills, and that it shows “surprising” performance on low-resource summarization, surpassing previous top results on six data sets with only 1,000 examples.  .... " 

Counting Crowds

Interesting application, note the technique of training in various contexts of measurement.   We don't do that enough with other problems.

A bird’s-eye view of a huge crowd of people. AI Could Help Count How Many People are in Large Crowds

New Scientist  by  Chris Stokel-Walker

German Aerospace Center researchers have developed an artificial intelligence (AI) system that can accurately count the number of people in large crowds. The researchers hand-counted nearly 250,000 people in 33 images of a large crowd taken from planes, drones, and helicopters, then used this data to train an algorithm called MRCNet. The algorithm divides each image into small squares and analyzes how many people are in each. The algorithm’s results were at least 15% more accurate than those of other AI-powered crowd estimation systems. The system is much faster than hand counting, taking 0.03 milliseconds to compute the number of people in each square. .... "

Contextual Intelligence: A Next Big Thing

Good overview piece in DSC.   I add some of our own experience using AI techniques.

We struggled with this a number of times.   All assistance exists in some context.  The context itself can be complex or not.   So for example we built a pump replacement system for choosing industrial pumps based on a dozen criteria.   Though this was technical, even including some predictive analytics,  it was straightforward.    Also easy to explain its details to the decision makers involved.   It also produced some additional meta information:  " how much did we, and would we spend on XYZ pumps next year, and what should their maintenance cost be?

But once we looked at complex systems.  Ones that depended on human behavior, required common sense decisions,  competitive reaction, this needed many sensory inputs, depended on the decisions of many other people and systems, the whole thing became difficult.  All this was asked in questions like "What profit is the company likely to make on new product X in the next year?"   At one point we estimated that such a question required several hundred external inputs.  Assistance in complex context is hard.  And you always end up with some risk/uncertainty attached to the assistance.     Even how the decision is implemented can add considerable risk and uncertainty.   Much work to be done.

Contextually Intelligent NLP Assistants – AI’s Next Big Technical Challenge   Posted by William Vorhies 

Summary:  Contextually intelligent, NLP-based interactive assistants are one of the next big things for AI/ML.  The tech is already here from recommendation engines.  The need to be more efficient and to become AI-augmented in our decision making is now.  Getting the contextual awareness is the hard part.

Last week we took the position that from a technical standpoint, ‘deeply inclusive and contextually sensitive’ AI is one of the two ‘next big things’ in AI.

In retrospect I wish there were a more concise agreed naming convention for this bit of technical legerdemain.  “Inclusive” and “contextually sensitive” are in the category of those ‘suitcase words’ Marvin Minsky called out as being so dependent on the user’s experience that agreement on meaning is difficult.

What we’re not talking about is the ability of NLP to hold a contextually appropriate conversation, such as making a reasonable response or request for clarification based on the topic at hand.  For the most part, short of performing psychoanalysis, chatbots can do pretty well with human ad hoc conversation.

Also, we’re not talking about being culturally inclusive as in detecting and eliminating bias.  Important, but not what we’re getting at.

What we’re describing is the next big step in NLP utility in which the NLP puts together facts it knows about us and proactively takes action or makes suggestions that make our life easier.

The example we gave in our previous article is about having the NLP assistant remind me of my mother’s upcoming birthday in a week or so without my having explicitly created a reminder.  More importantly my NLP assistant could make a recommendation for a present.  Presumably my past communications with her both in fact and tone contain some strong signals about my mom’s demographics and perhaps even her interests so why not predict a short list of appropriate gifts.  Now that would be valuable.

So perhaps a better description of this behavior then would be ‘contextually intelligent’.  We’ll stick with that.

On Scoop and Conversation Analytics

Remembering having this idea for interacting with people conversationally on a chatbot, but the technology was not even close then.   And then also using the conversation to route to the write human or machine resource.  It was Suggested that a 'context' map could be set up for a particular customer.   The technology at the time could not support the possibility.

On Scoop.ai: Conversation Intelligence for Maximum Sales Conversion

Close more deals with AI-powered Conversation Analytics.
Because bottom of the Sales funnel is the moment of truth
Scoop mines insights from conversations to prioritize opportunities and coach reps
CRMs don’t fully capture the reasons behind a deal’s win or loss. The details are hidden in conversations a Sales rep has with customer. Scoop uses ML to uncover the critical elements of winning sales calls.

From DSSResources.com:
AI-powered conversational analytics platform, Scoop to help brands glean unique insights from all customer conversations

Scoop has been incubated by Netcore Solutions, a global Marketing Automation provider

NEW YORK, Aug. 28, 2019 /PRNewswire/ -- In a recent development that highlights how rep-customer interactions can be leveraged to gain market intelligence, Scoop, a Netcore incubated startup, has developed an AI-powered conversational analytics platform designed to help Inside Sales teams to maximize sales conversion optimization.

Scoop leverages machine learning to mine insights from customer conversations across voice, video, and text sources to discover sales opportunities. The insights gleaned by the AI can be utilized to coach sales representatives, helping them win more deals with more efficient conversations over time.

Conversations between a sales representative and a customer are a storehouse of valuable information that usually lies untapped. These conversations usually happen across channels and mediums. Scoop ensures that all the conversations are transcribed and searchable, which are then mined by ML algorithms to uncover unique market insights.

The platform provides a clear and timely view of customer intent, while revealing the gaps and strengths of the representatives. This information can thus be analyzed to prepare a personalized coaching plan, shared with each representative, to unravel the mystery of why deals are won or lost. Scoop aims to enable enterprises to leverage conversational analytics, understand the key problems and eventually skyrocket their conversions.  .... "

The Value of Lineage Metadata

Did lots of work with metadata, especially in the corporate laboratory space.   Had never heard the term 'Lineage Metadata', though again it was often considered in our work.   Now would think this is more important than ever, if we are to create useful predictions, and also add some accuracy to future maintenance of any automated systems that emerge.  Lineage predicts changing context.   At very least the lineage can determine what errors might exist in the data, but in reality should provide much more in AI.  Should always be considered.     Lineage can also be considered as part of data asset value, predicting stability of value in context.    Also be included in a semantic representation of data involved.
I like that this is presented here.  Below an excerpt, much more at the link.- FAD

Lineage Metadata: The Fuel for Data Governance in Informationweek
Moshe Kranc is the chief technology officer at Ness Digital Engineering

The best way to achieve data quality is by combining or blending these three techniques: decoded lineage, data similarity lineage and manual lineage mapping.

Enterprises aspire to derive insights from data that can provide a competitive advantage. The most common impediment to achieving this goal is poor data quality. If the data that is being input to a predictive algorithm is “dirty” (with missing or invalid values), then any insights produced by that algorithm cannot be trusted.

To achieve data quality, it’s not enough to clean up the existing historical data. You also need to ensure that all newly generated data is clean by instituting a set of capabilities and processes known collectively as data governance. In a governed data environment, each type of data has a data steward who is responsible for defining and enforcing criteria for data cleanliness. And, each data value has a clearly defined lineage: We know where it came from, what transformations it underwent along the way, and what other data items are derived from this data value.

Data lineage provides an enterprise with many benefits:

The ability to perform impact analysis and root-cause analysis, by tracing lineage backwards (to find all data that influenced the current data) or forwards (to identify all other data that is impacted by the current data) from a given data item;
Standardization of the business vocabulary and terminology, which facilitates clear communication across business units;

Ownership, responsibility and traceability for any changes made to data, thanks to the lineage’s comprehensive record of who made what changes and when.

It sounds great, but where does data lineage information come from? Looking at a specific data value in the database tells us its current value, but it will not provide information about how the data evolved into its current value. What is missing is data about the data (lineage metadata) that automatically remembers the time and source of every change made to every data item, whether the change was made by software or by a human database administrator.

There are three competing techniques for collecting lineage metadata, each of which has its strengths and weaknesses: .... " 

Simulating the Ledger in Context

Simulation has always been a favorite tool.   Create a space that includes the context and metadata of an environment for some need.  Then build a means to explore it.  The exploration can be exploration combinatorially.    Or Monte Carlo Randomly.   Or Human logically.   Or via some need scripted exploration direction (like a flight simulator) .   Or even as a game.   Can get you lots of learning about the possibilities to meet some goal.   Here in another space we are exploring:

SimBlock: A Simulator for Testing Improvements to Blockchain Networks 
Tokyo Tech News via the ACM

Researchers at the Tokyo Tech Cybersecurity Research Center in Japan have developed a public blockchain simulator for use in testing blockchain performance and security. SimBlock allows users to simulate a blockchain network of about 10,000 nodes using a single conventional PC, and to modify behavior of those nodes to test the effects on performance and security. The software helps users test improvements on an existing or original blockchain, and can simulate malicious nodes and measure success rates of attack, before applying countermeasures to test their effects. .... "