ECS - Visão de Produto - Altice · focus on the set of AI technology systems that solve business problems, namely: robotics and autonomous vehicles, computer vision, language, virtual

Artificial Intelligence December 2017

2 Artificial Intelligence

INDEX

1. In Brief 2. Overview 3. Market Forecasts 4. Technological Landscape 5. Altice Positioning 6. Altice Labs Positioning 7. Conclusions 8. References

2

1. In Brief


In Brief 01.

1. Artificial Intelligence is a sub-field of computer science and its goal is to enable computers to perform tasks that normally require human intelligence, such as speech recognition, visual perception, decision making or language translation.

2. Since mid 1950s, AI has developed as a scientific area and, over the years ,there have been successive phases of enthusiasm but also frustration in the way of creating "intelligent machines”.

3. AI technologies are currently at the peak of expectations. Although it is very unlikely that machines will

exhibit broadly-applicable intelligence comparable to or exceeding that of humans in the next 20 years, it is to be expected that machines will reach and exceed human performance in more and more tasks, and also have a significant role on assisting and enhancing human capabilities.

4. The significant investment and commitment of digital companies on AI, along with the significant growth in computational capacity, the algorithms sophistication and the large volumes of data generated by an increasingly digital world, make a common believe in the industry that this time, AI came to deliver real value.

5. AI today includes a variety of technologies and tools, some time-tested, others relatively new. This report focus on the set of AI technology systems that solve business problems, namely: robotics and autonomous vehicles, computer vision, language, virtual agents and machine learning.


In Brief 01.

6. Machine learning and its subfield Deep learning are at the heart of many recent advances in AI and can be seen as enablers of many of the other technologies and applications. There exist a vast set of algorithms that can be used for machine learning, being the process of selecting the more adequate algorithm, for a given target we want to predict, a major competence of AI practitioners.

7. Digital transformation has become somewhat an obsession for Communication Service Providers (CSP). They are driven by the need to compete with fast-moving and digital native OTT and consumer tech players. CSPs need to move quickly and can advance digital transformation with solutions that leverage Artificial Intelligence along its value chain.

8. Successful use of AI is dependent on access to data to build, train and constantly improve the algorithms that power it. The AI strategy of CSP should be supported on a consistent state-of the-art Data Infrastructure that would enable the efficient and cost effective storage and streamlining at a massive scale of the millions of data points generated in the context of Opco’s “Digital Services Fabric”, whether they be performance metrics from the network or from customers ´s digital journey.

9. In order to reach the necessary critical mass in AI application and at the right pace, Altice Labs has been exploring, in the recent years, the incorporation of AI artifacts in the functional evolution of some of its product lines.

2. Overview


Overview: The beginnings of AI 02.

“Artificial Intelligence is a sub-field of computer science and its goal is to enable computers to perform tasks that normally require human intelligence, such as speech recognition, visual perception, decision making or language translation”.

That is a definition from our age, however the creation of artificial intelligence has always being in the dreams of man.

We can return to the golden robots of Hephaestus in ancient Greek mythology, to the programmable machines of Al-Jazari in the 12th century , to Mary Shelley´s Frankenstein in 19th century and more recently to 1950 Alan Turing paper “Computing Machinery and Intelligence” where he proposed what has come to be called the Turing test: “can a computer communicate well enough to persuade a human that it, too is human?”. AI was coined by John McCarthy in 1956 and since that date, mid 1950’s, it has developed as a scientific area. Over the years, there have been successive phases of enthusiasm but also frustration in the way of creating "intelligent machines".

The automated girl, from the Book of Knowledge of Ingenious Mechanical Devices by Al-Jazari


Overview: The Expert Systems phase 02.

As significant marks in this journey, we can name the first works on semantic networks for machine translation at Cambridge Language Research Unit in 1956 and the creation of the Artificial Intelligence Laboratory at MIT in 1959. Around the 1970s, there was a slow down on the funding from US government reflecting somehow the delay in the delivery of practical results from the AI work stream. Only by the 1980s, the AI resurge again with the development of “Expert Systems” – a computer system that emulates the decision-making ability of a human expert, reasoning about knowledge, represented as if-then rules rather than through conventional procedural code. Altice Labs (in its previous incarnation as “Centro de Estudos de Telecomunicações”, the R&D division of the national PTT) ended up being involved in this wave of AI. As examples from these phase, we can highlight the innovative application of KBS (knowledge based systems) to the network management discipline undertaken in the context of EU Framework Programme FP2 Advance Project, and the implementation of an expert system prototype to diagnose faults in the ATC 200/800 family of analogue switching systems , which would be affected by progressive lack of technical experts, mostly near the end of its professional carrier.


Overview: AI in the Hype 02.

Currently, Artificial Intelligence is again in the front pages with successive claims about the promises but also the dangers of AI to our daily lives, as individuals or communities. AI, is enabling machines to show human-like cognition, is relieving workers of repetitive or dangerous tasks, is driving our cars, is streamlining our electric energy production by a better prediction of supply and demand, is increasing corporate productivity but also empowering corporate spies or the steal of our privacy.

Hype Cycle for Artificial Intelligence, 2017, Gartner Inc.


Overview: Will AI deliver real value this time? 02.

As it is depicted by the recent Hype Cycle Gartner report, a set of AI technologies are at the peak of expectations. But is it now the time that AI will deliver real value? A set of confluent factors seem to be under way and give us the confidence that yes, AI is finally starting to deliver real-life business benefits. Those factors that make us believe we are entering a new era for AI are:

• significant growth in computational capacity • sophistication already achieved in algorithms • large volumes of data generated by an increasingly digital world • significant investment and commitment of digital companies on AI


Overview: Types of Artificial Intelligence (1) 02.

The scientific area of AI is commonly divided in two streams: General AI (also referred as AGI – Artificial General Intelligence), that is focused in creating a true human-thinking machine (somehow aligned with the Turing challenge), something that it is not yet possible today and that most probably will evolve slowly along the next decades, and

Narrow AI (or pragmatic AI) that is a AI system designed and trained for a specific task. It is this pragmatic AI that is already delivering value and that is attracting the attention of digital economy companies. This stream of AI has applications and use cases in almost every industry vertical and is considered the next big technological shift, similar to past shifts like the industrial revolution, the computer age, and the smartphone revolution https://www.tractica.com/research/artificial-intelligence-market-forecasts/..

ESET Robot

Google Self-Driving Car

https://www.tractica.com/research/artificial-intelligence-market-forecasts/










Another interesting and recent categorization of AI is the one advanced by Arend Hintze from Michigan State University published at https://theconversation.com/ , 14 Nov. 2016. He categorizes AI into 4 types, from the kind of AI systems that exist today to sentient systems, which do not yet exist.

Type I AI: Reactive machines. The most basic types of AI systems are purely reactive, and have the ability neither to form memories nor to use past experiences to inform current decisions. Deep Blue, the IBM chess program that beat Garry Kasparov in the 1990s, is a remarkable example of such type of machine. Also the more recent example of Google´s AlphaGo, although using a more sophisticated analysis method, based on neural networks, falls into this category. In both cases, the methods used do improve the ability of AI systems to play specific games better, but they can´t be easily changed or applied to another situation.

Type II AI: Limited Memory. This type contains machines that can look to the past experiences to inform future actions. Some of the decision-making functions in autonomous vehicles have been designed this way, namely the dynamic observations of other cars’ speed and direction are added to the self-driving cars’ pre-programmed representation of the world (lane markings, traffic lights, line curves in the roads, etc) and are to inform actions happening in the near future, such as a car that has changed lanes. But these observations are transient. They aren’t saved as part of the car’s library of experience it can learn from.

https://theconversation.com/



Type III AI: Theory of mind. With this new class of AI systems we are entering the zone of machines we will build in the future. Machines in this more advanced class, not only form representations about the world, but also about other agents or entities in this world. This capability is named in psychology as “theory of mind”. The understanding that people, creatures and objects in the world can have thoughts and emotions that affect their own behavior.

Type IV AI: Self-awareness. In this class, AI systems have a sense of self, they present consciousness. This corresponds to the more elaborated class, and to develop it AI researchers have not only to understand consciousness, but build machines that have it. This can be seen as an extension of the theory of mind, the previous type of AI capability. Machines with self-awareness, know about their internal states, and are able to predict feelings and infer actions of others.


Overview: Augmented Intelligence 02.

Augmented Intelligence or Intelligence Augmentation focus on the assistive role of AI, emphasizing its design to enhance human intelligence, expertise and skills, rather than replacing them.

Sophisticated AI programs are capable of analyzing large amounts of data quickly and efficiently to reach a decision, but that decision is only as good as the data humans provide as input.

Building trust is a key aspect for the development and deployment of AI technology in a responsible manner, and experts anticipate the need for a significant multidisciplinary scientific effort to instill human values, sense of morality, operation transparency, bias avoidance and general interaction ethical principles, so that the enormous societal benefits and implications of AI systems can be attained in a shared context, side-by-side as human assistants.

Illustration by Derek Bacon

3. Market Forecasts


Market Forecasts: Artificial Intelligence revenues 03.

The analysis of the public information provided by market research companies points to a significant variation in the current value of the global market of AI. Eventually derived from the fact that there are few products on the market or about to enter the market and generate widespread adoption. Projections point to market values in 2016 between $ 860 million to $ 1.38 billion USD. In what concerns values for the market forecast for the period 2016-2025, the common factor relates to high growth rates forecasted: 52% (Tractica) or even 62,9% (Research & Markets) CAGR, with target value for 2025 pointing to $59,75 billion USD. This is data from a 2Q17 report (see figure) which represents an upgrade of Tractica’s previous projection for AI revenue generated by direct and indirect application of AI software, which was published in 3Q16, owing to a greater than anticipated pace of change and development in the AI sector.


Market Forecasts: Artificial Intelligence investment 03.

The big companies of the digital world, as the on-line giants Google and Baidu, are investing heavily on AI. In Its AI Report, from June 2017, Mckinsey GI estimates worldwide AI investment between $20 billion and $30 billion USD (including significant M&A) in 2016. Private investors are also focusing their attention on AI enterprises. From the same McKinsey GI report it is estimated that private equity funds invested between $1 billion and $3billion USD in AI during 2016, and that venture capitalists invested $4 billion to $6 billion USD. From a survey made by Forrester in beginning 2016, the forecasted 2017 AI investment increase is around 300%.

In 2016, worldwide companies invested in AI

$26B to $39B

For 2017 it was estimated that AI investment worldwide would increase

300%


Market Forecasts: Artificial Intelligence investment 03.

In complement to the previous figures of worldwide venture capital investment in AI firms ($4 billion to $6 billion USD in 2016), the graph depicts the evolution within USA of investment rounds and total amount raised by predictive modeling / machine learning companies from 2012 to 2016. What is most significant in the raised investment value evolution is that it occurs at the same time that the venture capital (VC) investment for USA startups has been falling, being 2016 with a 32% drop to a global value of $52.4 billion USD, the steepest annual drop since the 2001 collapse following the dot-com bubble. These figures confirm the financial market interest for companies involved in artificial intelligence and predictive modeling / machine learning in particular.

4. Technological Landscape


Technological landscape: AI Technology Systems Map 04.

AI today includes a variety of technologies and tools, some time-tested, others relatively new. The following classification took into considerations various proposals, as the Forrester TechRadar, and ended more aligned with the one proposed within MacKinsey Global Institute Report. The focus is on the set of AI technology systems that solve business problems, and were identified 5 technology systems that are key areas of AI development:

Robotics and Autonomous vehicles

Computer Vision

Language

Virtual Agents

Machine Learning Some are related to process information from the external world, such as computer vision and language (including natural language processing, text analytics, speech recognition and semantics technology); others are related to acting on information, such as robotics, autonomous vehicles or virtual agents; Machine learning and its subfield Deep learning are at the heart of many recent advances in AI and can be seen as enablers of many of the other technologies and applications.


Technological landscape: AI Investment Map 04.

External Investment in AI-focused companies by technology category: 2016 ($billion) Source: Artificial Intelligence The Next Digital Frontier?, McKinsey Global Institute, June 2017

The graphic summarizes the annual investments (VC, PE and M&A) in AI focused companies during 2016, using the categorization presented before (the only change is on the separation of smart robotic and Autonomous vehicles into two separate categories). The graphic shows in a clear way the attraction on Machine Learning area. Machine Learning attracted almost 60% on that investment, mostly likely because it is an enabler for so many other AI technologies and applications.


Technological landscape: Robotics 04.

Smart Robotics Robotics is one field involving mechanical, usually computer-controlled, devices to perform tasks that require extreme precision or tedious or hazardous work by people.

Traditional Robotics uses Artificial Intelligence planning techniques to program robot behaviors and works toward robots as technical devices that have to be developed and controlled by a human engineer.

The Autonomous Robotics approach suggests that robots could develop and control themselves autonomously. These robots are able to adapt to both uncertain and incomplete information in constantly changing environments. This is possible by imitating the learning process of a single natural organism or through Evolutionary Robotics, which is to apply selective reproduction on populations of robots. It lets a simulated evolution process develop adaptive robots.

Tri.on concept Robot da Vinci robotic surgery system

Boston Dynamics Atlas robot


Technological landscape: Robotics 04.

Robotic Process Automation Robotic Process Automation (RPA) is an emerging form of clerical process automation technology based on the application of software robots workers that incorporate artificial intelligence artifacts for advanced decision making and inference.

The value proposition of RPA is to automate manual, logically-designed processes that are repetitive and time-consuming. They are usually configured to automatically capture and interpret existing applications for processing a transaction, manipulating data, triggering responses and communicating with other digital systems.

The key difference that distinguishes RPA from enterprise automation tools like business process management (BPM) is that RPA uses software or cognitive robots to perform and optimize process operations rather than human operators. RPA does not require invasive integration or changes in the underlying systems, allowing organizations to rapidly deliver efficiencies and cost-savings mainly by replacing repetitive and/or low value added human tasks with software robots.

Orchestrate Robotic Processes with UiPath


Technological landscape: Autonomous Vehicles 04.

AI systems, which continuously learn from experience by their ability to discern and recognize their surroundings, have the potential to be highly beneficial when integrated into an autonomous vehicle's software architecture. Autonomous vehicles already rely on advanced tools to gather information, including long-range radar, LIDAR, cameras, short/medium-range radar, and ultrasound to analyze and adapt to a rapidly changing environment. Each of these technologies is used in different capacities, and each collects different information. However, this information is useless unless it is processed and some form of action is taken based on the gathered information.

photo source Electronics Weekly EHang 184 drone


Technological landscape: Autonomous Vehicles 04.

AI could help these self-driving vehicles recognize patterns and learn from the behavior of other vehicles on the road. The real power of this approach is realized because autonomous cars have one advantage that human drivers don’t have; self-driving cars have the ability to share their experiences and readings with other cars instantaneously. Information and situations encountered by autonomous cars along every mile driven are shared with other vehicles so that each computer can adapt its algorithm to the environments faced by other vehicles. This type of shared experience and active learning creates a situation where autonomous cars, through Artificial Intelligence algorithms, can improve their ability to react to situations on the road without actually having to experience those situations first-hand.


Technological landscape: Computer Vision 04.

Computer vision (CV) is a field that involves methods for acquiring, processing, analyzing and understanding real-world images and video to allow machines to extract meaningful, contextual information from the physical world.

Today, there are numerous different and important CV technology areas, including machine vision, optical character recognition, scene recognition, pattern recognition, facial recognition, edge detection and motion detection.


Technological landscape: Computer Vision 04. The convergence of enabling technologies — such as deep learning, neural networks, large swaths of data and massive parallel processors — has brought new life into the significantly advancing field of computer vision.

Thirty years ago, object classification was a relatively difficult task. However, today, current CV implementations are able to classify millions of individual objects with more than 95% accuracy.

These advancements allow CV algorithms to make quicker and more accurate visual identifications, which led to the rise of new players in CV (outside of academia) such as Baidu, Microsoft and Google.

CV is still in its adolescence, and it is only beginning to be recognized for its broad applicability. As such, adoption is still limited, but is expected to ramp up quickly — along with its maturity.

CV is one of the numerous technologies that enable Augmented Reality, which will add further interest for its adoption.

Ikea Design App

Hyper-Reality concept film


Technological landscape: Natural Language 04.

Speech recognition technology (STT – Speech To Text) is a set of technologies capable of recognizing human speech and then translate phrases into text strings which can then be further processed by computer means.

The term Natural Language Processing (NLP) is usually applied to the use of several natural language tools (more than 20 different tool classes), such as knowledge graphs, speech-to-text, machine translation, automatic summarization, entity recognition, question answering, natural-language generation, sentiment analysis and text analytics. NLP helps to make the interaction between humans and computers easier and more natural to humans.

Natural Language Generation (NLG) is a technology capable of generating natural-language descriptions (e.g. complete English text descriptions). The technology combines Natural Language Processing with Machine Learning in order to generate contextualized insights from large volumes of data.

This technology is becoming common with large information analysis platforms, where Information like trends, data relationships and correlations are dynamically identified and explained in plain English, as the users interact with BI and Analytics applications. NLG is used to synthesize textual content by combining analytic output with contextualized narratives, helping less expert users to derive knowledge from complex data, without the help of experts.

photo source Yahoo


Technological landscape: Virtual Agents 04.

Virtual Agents or Assistants (VAs) help users or enterprises with a set of tasks previously only possible by humans.

VAs use AI and machine learning (e.g., NLP, prediction models, recommendations and personalization) to assist users or automate tasks.

VAs listen and observe behaviors, build and maintain data models, and predict and recommend actions. They may act for the user, forming a relationship with the user over time.

Virtual assistants shift responsibility for understanding the business process from the user to the system by corresponding with the user.

Design Hill


Technological landscape: Virtual Agents 04.

The VA space is currently dominated by conversational interfaces such as Apple Siri, Google Assistant, Microsoft Cortana, IPsoft Amelia, Nuance Nina, Amazon Alexa, Kore.ai and SAP CoPilot. Increasingly, image recognition, behavior and event triggers will enhance VAs. Virtual assistants can be deployed in several use cases, including virtual personal assistants, virtual customer assistants and virtual employee assistants. VAs can act on behalf of consumers, employees and businesses, but the use cases are all based on the same AI technologies.

VA adoption grows as users get more comfortable with them, technologies improves and the variety of implementations multiply:

- Unobtrusive, VA-like features, such as Gmail's Smart Inbox with recommended replies and Microsoft's Delve that finds unknown resources embedded in existing products.

- Narrow-purpose VAs have also emerged (such as personal financial advisors, health and wellness coaches, and calendaring agents).

- Virtual assistants are increasingly used to answer customer questions about products and services.

Amazon Echo Dot Apple Siri


Technological landscape: Machine Learning 04.

Machine Learning is the basis for many recent advances and commercial applications of AI. Depending on definitional boundaries, Machine Learning is synonymous with, or largely overlapping with Predictive Modeling. In another way, Machine Learning is a tool for Predictive Data Analytics.

Predictive Data Analytics: moving from Data to Insights to Business Decisions Predictive Data Analytics encompasses the business and data processes and computational models that enable a business to make data-driven decisions, in domains as: Fraud Detection, Dosage Prediction, Risk Assessment, Propensity modeling, Diagnosis, Document Classification, etc.

Machine Learning is a statistical process that starts with a body of data and tries to derive a rule or procedure that explains the data or can predict future data.

source John D. Kelleher



Some key concepts: Descriptive Features: These are attributes that are extracted from data sources to analyze and form a key input to the predictive models. E.g. time start, duration, key word, address, etc. Target Feature: Represents the outcome and is related to the correspondent set of descriptive features . In the context of evaluating and generating predictions, the target feature is the feature whose value will be predicted by a trained ML model.(E.g. possibility of an email be spam or ham). Models: Mathematical structure that characterizes a range of possible decision-making rules with adjustable parameters. A model is like a “box” that applies a rule, and the parameters are adjustable knobs on the front of the box that control how the box operates. In the end a model might have many millions of parameters. (Supervised) Machine Learning ML techniques automatically learn a general model of the relationship between a set of descriptive features and a target feature from a set of historical examples. Unsupervised Learning: In this case the learning algorithm is left alone to find structure in its input un-labeled data. This approach can be used to find hidden patterns in data or to discover the representations needed for feature detection or classification from raw data. Reinforcement Learning: In this case the learning algorithm gets to choose an action in response to each data point. The learning algorithm also receives a reward signal a short time later, indicating how good the decision was. Based on this, the algorithm modifies its strategy in order to achieve the highest reward. Reinforcement learning is common in robotics, where the set of sensor readings at one point in time is a data point, and the algorithm must choose the robot's next action.



To apply machine learning (supervised) we start with a model and a historical data set that is divided into a training set and a test set. We also require an objective function, used to evaluate the desirability of the outcome, that results from a particular choice of parameters. Training the model is the process of adjusting the values for these parameters which either maximize or minimize this objective function (e.g. the mean). The universe of possibilities can be huge, so successful training algorithms have to be clever in how they explore the space of parameter settings, so as to find very good settings with a feasible level of computational effort.

The accuracy of the model is evaluated against the test data set. The goal of machine learning is to create a trained model able to generalize – be accurate not only with the training data set but with new data instances.

Using ML to induce a prediction model from a training dataset

Using this model to make predictions for new query instances

…

source John D. Kelleher



Deep learning expands standard machine learning by allowing intermediate representations to be discovered. These intermediate representations allow more complex problems to be tackled and others to be potentially solved with higher accuracy, fewer observations and less cumbersome manual fine-tuning.

Seen by some authors as a sub-branch of Machine Learning, Deep Learning is getting a lot of attention lately because it is capable of addressing previously human-only capabilities, such as image recognition, text understanding and audio recognition.

source Max Tegmark - MIT

Deep learning discovers intricate structure in large data sets by using, for instance, the back-propagation algorithm to indicate how a SW machine should change its internal parameters that are used to compute the representation in each layer from the representation in the previous layer. Deep convolutional nets have brought about breakthroughs in processing images, video, speech and audio, whereas recurrent nets have shone light on sequential data such as text and speech.



There exist a vast set of algorithms that can be used for machine learning.

A set of aspects should be taken into consideration when choosing an algorithm:

•Accuracy •Training time •Linearity •Nº of parameters •Nº of features

Machine Learning Algorithms Mindmap Source: Machine Learning Mastery



When to apply a specific ML algorithm?

In order to give a view on the process of selecting the more adequate algorithm for a given target that we want to predict , it is presented here the type of approach proposed by cloud-based solutions for use of machine learning technologies, namely Microsoft Azure and Amazon. Both solutions group its algorithm offer in similar ways. Bellow we preset the Amazon approach: http://docs.aws.amazon.com/machine-learning/latest/dg/.

Binary Classification Models: ML models for binary classification problems predict a binary outcome (one of two possible classes). To train binary classification models, Amazon ML uses the algorithm known as logistic regression. Examples of Binary Classification Problems: "Is this email spam or not spam?" ; "Will the customer buy this product?" ; "Is this product a book or a farm animal?“; "Is this review written by a customer or a robot?"

Multiclass Classification Model: ML models for multiclass classification problems allow you to generate predictions for multiple classes (predict one of more than two outcomes). For training multiclass models, Amazon ML uses the multinomial logistic regression algorithm. Examples of Multiclass Problems: "Is this product a book, movie, or clothing?“; "Is this movie a romantic comedy, documentary, or thriller?“; "Which category of products is most interesting to this customer?

Regression Model: ML models for regression problems predict a numeric value. For training regression models, Amazon ML uses the industry-standard learning algorithm known as linear regression. Examples of Regression Problems: "What will the temperature be in Seattle tomorrow?" ; "For this product, how many units will sell?" ; "What price will this house sell for?"

http://docs.aws.amazon.com/machine-learning/latest/dg/





When to apply a specific ML algorithm?

https://docs.microsoft.com/en-us/azure/machine-learning/studio/algorithm-choice


5. Altice Positioning


Altice Positioning: AI Transforming Industries 05.

The current expectations are on the disruptive potential of AI across different business domains. To fulfil the expectations that have been created upon it, AI will need to deliver economic applications that significantly reduce costs, increase revenue and better engage enterprises with its customers, current and potential.

In the recent report “Artificial Intelligence: The next digital frontier”, from McKinsey Global Institute, it is advanced a categorization of the ways in which AI can create value along the production value chain. This categorization considers 4 areas:

Project: enabling companies to better project and forecast to anticipate demand, optimize R&D, and improve sourcing

Produce: increasing companies’ ability to produce goods and services at lower cost and higher quality

Promote: helping promote offerings at the right price, with the right message, and to the right target customers

Provide: allowing the companies to provide rich, personal and convenient user experiences


Altice Positioning: AI supporting Digital Transformation of CSPs 05.

Digital transformation has become somewhat an obsession for Communication Service Providers (CSP). They are driven by the need to compete with fast-moving and digital native OTT and consumer tech players. CSPs need to move quickly and can advance digital transformation with solutions that leverage Artificial Intelligence along the previously referred areas of its value chain, namely: project, produce, promote and provide. Successful use of AI is dependent on access to data to build, train and constantly improve the algorithms that power it. Although AI applications, as virtual agents and chat bots developed on top of natural language processing competences, are positioned as very attractive AI use cases applicable in various customer facing scenarios, the AI strategy of CSP should be supported on a consistent state-of the-art Data Infrastructure as a prerequisite for success. Such Data Infrastructure will enable the efficient and cost effective storage and streamlining at a massive scale of the millions of data points generated in the context of Opco’s “Digital Services Fabric”, whether they be performance metrics from the network or from customers ´s digital journey.



From Data to Intelligence

Data fuels AI automation. Data has long been understood to be the source for developing insights about business´s operations, its customers and its prospects. Data plays a critical role in enabling AI applications to achieve expectations, because data is the measure of what the organization is doing, how effectively and efficiently it is conducting the business and how it may make changes to improve performance and user expectations. When an organization engages in applying AI technology as predictive analytics, these data measures are the source for building, training and constantly improving the algorithms underlying the AI capabilities that enable automation. The quality of these Data is critical, as it impact directly the quality of AI insights.

In fact, operators have huge amounts of data, and that data is growing 100%

every two years



Basic Data Collection Descriptive Analytics

Diagnostic Analytics

Predictive Analytics

Prescriptive Analytics

(cooperative)

Prescriptive Analytics

(automatic)

Curative Actions Preventive Actions

Artificial Intelligence and Machine Learning allows to extract more value from data

Data is critical

Data has some value Data is a cost

Value comes from Intelligence


Altice Positioning: AI Roadmap for CSPs 05.

The following are examples of opportunities for CSPs to leverage AI:

Project • Anticipate demand trends, while optimizing and automating supplier negotiation and contracting.

Produce • Evolve the way CSPs operate and manage their networks, supporting them to pursue the vision of autonomous self organized networks, leveraged on the transition towards software defined (SDN) and virtualized networks (NFV). • Replace/augment manual intervention capability in selected business processes (ex. Robotic Process Automation).

Promote • Evolving from “one-size fits all” to “personalization at scale”. Use automated personalized services that serve millions of customers at a time, allowing the generation of personalized offers. • Campaign management; identify the target group for maximizing campaign success. • Automatically fine-tune product catalogue/marketing offerings (price, size,..), based on deep learning of the competition from available data such as advertising, customer feedback and BSS data.

Provide • Evolved Home services and entertainment, via human-machine voice communication and natural language processing, intelligent content recommenders, taking advantage of digital assistants. • Fraud detection and Cyber security capabilities.


Altice Positioning: AI initiatives in the CSP context 05.

This major Korean Operator is applying AI predictive analytics to improve network management and network optimization. This is being accomplished with the evolution of its TANGO OSS (T Advanced Next Generation Operation Support System) in order to cover management automation within the fixed and mobile network domains. According to Park Jin-hyo, Head of the Network Technology R&D Center at SK Telecom, “The AI-assisted network operation technology based on big data analytics will be essential in the 5G era”. http://telecoms.com/485584/...

Many CSP are already engaging in applying AI technology across their operations while others are still formulating their AI strategies. From news issued along 2017 we briefly present AI initiatives in key CSP players as Telefonica, AT&T and SK Telecom .

http://telecoms.com/485584/sk-telecom-extends-its-ai-management-system-to-mobile-network/


Altice Positioning: AI initiatives in the DSP context 05.

In what concerns Telefonica we can refer two major initiatives announced this year (2017) on the use of AI techniques in its operations. A first one is related to the application of AI to the network management domain, namely the use of AI predictive analytics in the context of Service Operations Centres to deliver more insight into how its mobile networks are being used, to anticipate problems as the “silent churners” and identify new ways to improve user experience. http://www.computerweekly.com/news/450417040/... The other AI initiative is the Telefonica´s Digital Assistant AURA, a customer interaction platform based on cognitive intelligence. “AURA will enable users to manage their digital experiences with the company and control the data generated using Telefonica´s products and services in a transparent and secure manner”. https://www.telefonica.com/en/web/press-office/-/telefonica-presents....

http://www.computerweekly.com/news/450417040/

https://www.telefonica.com/en/web/press-office/-/telefonica-presents-aura-a-pioneering-way-in-the-industry-to-interact-with-customers-based-on-cognitive-intelligence








Altice Positioning: AI initiatives in the DSP context 05.

AT&T has been investing in AI for a while now namely in service assurance areas as to continuously predict failures and solve network degradation via automation tools, or the application of machine learning to help prevent, detect and mitigate cyber attacks. https://www.business.att.com/content/productbrochures/cyber.. In line with its previous open source initiative, namely ONAP, www.onap.org, the “operating system for virtualized networks”, AT&T, in partnership with Tech Mahindra, announced, last October 2017, the Acumos platform, that intends to bring AI to the mainstream creating a marketplace for accessing, using, developing and deploying AI applications. The Acumos platform is hosted as an open source project by the Linux Foundation, which will give the developer community the capability to edit, integrate, compose, package, train and deploy AI microservices. “Our goal with open sourcing the Acumos platform is to make building and deploying AI applications as easy as creating a website” (Mazin Gilbert, VP of Advanced Technologies at AT&T Labs). http://about.att.com/story/building_open_source_ai_marketplace.html

https://www.business.att.com/content/productbrochures/cybersecurity-threat-tech-white-paper.pdf

http://www.onap.org/

http://about.att.com/story/building_open_source_ai_marketplace.html

6. Altice Labs Positioning


Altice Labs Positioning 06. Altice Labs has been exploring, in the recent years, the application of AI tools and techniques, in the context of the functional evolution of some of its own product lines.

In order to reach the necessary critical mass in AI application and at the right pace, Altice Labs has been working with its academic partners with long term expertise in the area of Machine Learning, Artificial Intelligence and Analytics. Relevant examples of this are the H2020 SELFNET projects, where Altice Labs partners with important European AI experts and the recently contracted project with the University of Minho (Portugal), in the area of Self Organizing Networks .

From these work lines, we briefly detail results of AI use cases in the following areas:

•Marketing Campaign Engines for telecom services •Autonomous Network Management •Automatic digital content recommendation


Altice Labs Positioning: AI & Marketing Campaigns 06.

DESIGNER REAL TIME EXECUTION

ENGINE

OPERATION & ADMIN. CONSOLE

PERFORMANCE TRACKING

• ACM (Active Campaign Manager) is an Altice Labs product which allows the CSPs to design and then execute campaigns and promotions in Real Time

• By using customer behavior information and customer context, the CSP can target customers with personalized and accurate campaigns, giving him the best benefit at the right moment, leading to significantly better results

• AI techniques greatly improved the targeting process and the choice of the best campaign for each customer within the target group


Altice Labs Positioning: AI & Autonomous Network Management 06.

• Sensors in the network (probes) identify anomalous network behaviors (Monitoring framework)

• AI engine analyzes, diagnoses and autonomously decides which action to take on the network

• OSS (fulfillment framework) introduces the necessary changes into the network

• Work performed in the scope of EU H2020 project - SELFNET


Altice Labs Positioning: AI & Content Recommendation 06.

• Digital content offered today to customers as part of a 4P service package is overwhelming, with most options never actually used

• AI based content recommender systems help the service provider in providing personalized recommendations, by trying to predict what the most suitable products or services are, based on the past user’s preferences and constraints.

• Work performed in the scope of EU H2020 project - STREAMLINE

7. Conclusions


Conclusions 07. After going through ups and downs over the past few decades, it looks like AI came to stay!

If it is true that the creation of machines that exhibit intelligent behavior, at least as advanced as a person across the full range of cognitive tasks, seems to be decades away, the narrow field application of AI technology systems to solve specific business problems (narrow AI) is demonstrating a remarkable success in areas such as playing strategic games, language translation, self-driving vehicles or image processing. The current expectations are on the disruptive potential of AI across different business domains, significantly reducing costs, increasing revenue and better engaging enterprises with its customers, current and potential.

For CSP/DSP it is time to engage with AI and build/consolidate their AI competences by working with AI experts, namely from academia and technological startups, when there is a lack of in-house expertise. A key aspect on this AI positioning is the existence of a consolidated Data strategy in the organization. Data has long been understood to be the source for developing insights about business´s operations, its customers and its prospects.

Envisaged opportunities for AI in the context of CSP/DSP value chain span from improving customer experience and support (via Virtual Agents assisted customer support), to network operation optimization (evolving towards the vision of autonomous self organized networks), to Home services and entertainment (via natural language processing agents and intelligent content recommenders) or to marketing engagement (based on personalization at scale, on optimized campaign management or on automatically fine-tune product catalogue/marketing offerings).

Aware of AI's strategic importance, Altice Labs is already addressing some key areas of AI application and expects to be investing more in this area in the near future, recognizing the “game-changing” nature of AI for DSPs.

8. References


References 08. • J. Pias, P. Neves, L. Cortesão, J. Zemitis, C. Analide, “AI: Overview and Applications”, Innovaction #2, Altice Labs, Aveiro, 2017 • Pilkington, Ace G. “Science Fiction and Futurism: Their Terms and Ideas”, McFarland, 2017. ISBN: 978-0786498567 • A. Uzon, F. Vatansever “Ismail Al Jaziri Machines and New Technologies”, acta mechanica et automatica, vol.2 no.3 2008 • M. Moreira, F. Carvalho, J. Neves, “Sistema Pericial Genérico para Diagnóstico de Placas de Hardware”, 1º Encontro Nacional do Colégio de

Engenharia Electrotécnica da Ordem dos Engenheiros, Lisboa, 1994 • Cruz, João Pedro “Conclusões sobre o trabalho desenvolvido para o “Sistema Diagnóstico para a ATC 200/800” – CET internal report, 1993 • L. da Silva , T. Belleli, S.P. Harris “ A Decision Support System for Planning GSM Radio Coverage”, IEE Colloquium on GSM and PCN Enhanced

Mobile Services , 1991 • L. da Silva, F. Carvalho, “A KBS for Mobile Cell Configuration”, 4th RACE TMN Conference, Dublin ,1990 • http://searchcio.techtarget.com/definition/AI • https://theconversation.com/understanding-the-four-types-of-ai-from-reactive-robots-to-self-aware-beings-67616 • http://telecoms.com/485584/sk-telecom-extends-its-ai-management-system-to-mobile-network/ • http://www.computerweekly.com/news/450417040/Telefonica-rolls-out-AI-based-network-management • https://www.telefonica.com/en/web/press-office/-/telefonica-presents-aura-a-pioneering-way-in-the-industry-to-interact-with-customers-

based-on-cognitive-intelligence • https://www.business.att.com/content/productbrochures/cybersecurity-threat-tech-white-paper.pdf • http://about.att.com/story/building_open_source_ai_marketplace.html • https://www.forbes.com/sites/gilpress/2017/01/23/top-10-hot-artificial-intelligence-ai-technologies/#79fb0f751928 • https://docs.microsoft.com/en-us/azure/machine-learning/studio/algorithm-choice • http://docs.aws.amazon.com/machine-learning/latest/dg/types-of-ml-models.html • Executive Office of the President of the United States, NSTCCT, “Preparing the Future Of Artificial Intelligence”, October 2016 • https://www.uipath.com/platform • https://www.tcs.com/robotic-process-automation-for-telecom-carriers • https://en.wikipedia.org/wiki/Robotic_process_automation • https://www.pega.com/marketforce-future-of-work

http://searchcio.techtarget.com/definition/AI

https://theconversation.com/understanding-the-four-types-of-ai-from-reactive-robots-to-self-aware-beings-67616

http://telecoms.com/485584/sk-telecom-extends-its-ai-management-system-to-mobile-network/

http://www.computerweekly.com/news/450417040/Telefonica-rolls-out-AI-based-network-management



https://www.business.att.com/content/productbrochures/cybersecurity-threat-tech-white-paper.pdf

http://about.att.com/story/building_open_source_ai_marketplace.html

https://www.forbes.com/sites/gilpress/2017/01/23/top-10-hot-artificial-intelligence-ai-technologies/


http://docs.aws.amazon.com/machine-learning/latest/dg/types-of-ml-models.html

https://www.uipath.com/platform

https://www.tcs.com/robotic-process-automation-for-telecom-carriers

https://en.wikipedia.org/wiki/Robotic_process_automation

https://www.pega.com/marketforce-future-of-work

Artificial Intelligence December 2017

Documents

ECS - Visão de Produto - Altice · focus on the set of AI technology systems that solve business problems, namely: robotics and autonomous vehicles, computer vision, language, virtual