microsoft ai innovate

Microsoft AI Innovate Build | Scale | Transform

© M i c r o s o f t 2 0 2 1

2

Microsoft AI Innovate

3

Microsoft AI Innovate

At Microsoft, we ask ourselves not just what technology can do but what it should do. That is why we built Microsoft AI principles:

Fairness AI systems should treat all people fairly

Reliability & Safety AI systems should perform reliably and safely

Privacy & Security AI systems should be secure and respect privacy

Inclusiveness AI systems should empower everyone and engage people

Transparency AI systems should be understandable

Accountability People should be accountable for AI systems

All of us need to support the AI ecosystem with the right programs, tools, and skilling. That is how Microsoft AI Innovate was born. It is a stepping-stone to AI-based innovation in the country, spurring the larger tech ecosystem to adopt AI-led innovation.

At Microsoft, we want to help reimagine what’s possible. Help every organization in every industry turn meaningful innovation into actionable results and give the power of AI to everyone. By doing so, we empower every person and every organization to achieve more.

t’s exciting to see how mainstream AI is, and how it touches almost every aspect of our lives. It has the potential to transform and solve some of the

most pressing issues for the planet.

India has the third-largest AI startup ecosystem in the world. AI adoption can add $90 billion to the Indian economy by 2025 and $15.7 trillion to the global GDP by 2030. It’s easy to be optimistic about the role of technology in shaping a better future for all. But the real question is how do we maximize AI’s potential and mitigate its risks? How do we need to develop it in a way that is responsible and fosters trust?

AI amplifies human ingenuity. It should be designed to benefit everyone. As creators, users, and advocates of technology, it is up to us to make careful choices so that technology translates into benefits and opportunities for all.

Anant MaheshwariPresident, Microsoft India

I

Building a better world with AI

C o n t e n t s04

11

22

06

14

25

28

32

08

20

17

Every startup is an AI startup

Age of intelligence in India

Why AI maturity is important

AI for developers anddata scientists

Challenges and approaches to productizing research

Developing AI skills

Intelligent platforms, apps,and agents

Building a successful AI culture

AI for Good

Responsible AI

Knowledge mining

54

Microsoft AI InnovateMicrosoft AI Innovate

Every startup is an AI startup: Accelerating startups’ growth with Artificial Intelligence

ndia is the world’s third largest tech startup ecosystem which, despite the pandemic,

continues to grow steadily at 8-10 percent annually, especially in the technology domain. Over 1600 startups and a record number of 12 tech unicorns emerged in India in 2020.1 DeepTech startups base in India has expanded at >40% CAGR over the past four years, with Artificial Intelligence (AI) and Internet of Things (IoT) accounting for two-thirds of all DeepTech startups.2

AI is an enabler of innovation and disruptionWe are seeing Indian startups leverage AI to become the pole-bearers of innovation, solving problems, disrupting industry

norms, or enhancing capabilities in diverse areas ranging from dairy to defense. AI is the enabler, the tool, the means to create solutions; it is not always the end-goal. Every startup, irrespective of the domain it focuses on, leverages AI capabilities in some form or the other, and many of their business functions and operations are powered by AI capabilities. Even if the end-solution is seemingly as “non-technical” as a Human Resource (HR) platform, it still has many components that leverage the power of AI, data, and analytics. When viewed in this light, it becomes evident that every startup is a tech startup, and, increasingly, an AI startup.

Rohini SrivathsaNational Technology Officer (NTO), Microsoft India

AI is now democratized, and increasingly, ubiquitousThere are many ways in which we, as consumers, use AI in our day-to-day activities without realizing it – unlocking our phone with face ID; seeing our emails automatically categorized; receiving content suggestions on websites; navigating while driving; converting speech to text in the form of subtitles or transcripts; interacting with chatbots and smart personal assistants, and more.

AI is now ubiquitous, and indeed democratized, such that every startup needs to use AI at the heart of its technology stack to offer its customers the experience they have come to expect. AI helps startups discover information, learn from data, and drive intelligent decisions. It makes operations more efficient; enhances product and service innovation; and enables higher employee effectiveness and creativity.

A three-pronged approach to AI in the digital ageMicrosoft believes that the approach of any organization to AI, including startups, should be based

I

DeepTech startups base in India has expanded at >40% CAGR in the last 4 years

on three key pillars: Meaningful Innovation; Empowering People; and Responsibility.

Innovation with Agility at Scale: Innovation can be greatly accelerated with AI capabilities that are on par with those of humans in areas such as object recognition, speech recognition, reading comprehension, and language translation. Microsoft, with many firsts to its credit in the AI domain, including support for 12 Indian languages, is enabling Indian startups to innovate in India and for India.

Empowering Everyone: AI capabilities are no longer restricted to data scientists, but rather accessible by all skill levels. Citizen developers to bot designers, application developers to engineers with no-code or code-first experiences can leverage cognitive services, knowledge mining and machine learning capabilities today to develop AI-powered applications and solutions. Microsoft is democratizing AI, so that anyone can use our AI solutions to build a business.

Responsible AI: “Businesses and users are going to embrace technology only if they can trust it” – Satya Nadella.

Beyond espousing our principles for Responsible AI – fairness, security & privacy, safety & reliability, inclusion, accountability, and transparency – Microsoft is putting our principles to practice. We are building our products with

overall economic potential.

A 2020 NASSCOM report estimated AI’s contribution to India’s GDP at $500 billion by 2025, with impact across a wide range of sectors. Startups are key to unlocking this potential for India through innovative solutions for untapped market opportunities, customer needs, socio-economic challenges. Microsoft’s mission is to empower every entrepreneur and every startup in India to innovate, scale and grow with the power of AI.

66% of DeepTech startups are driven by Artificial Intelligence (AI) and Internet of Things (IoT)

Responsible AI by design, as well as helping our customers and partners with tools and techniques to build AI systems responsibly.

AI has tremendous potential for India’s economyOver the past decade, AI has emerged as one of the most transformational technologies of the digital age.However, we are still only scratching the surface when it comes to the technology’s

Innovation with agility at scale

Empowering everyone

Responsible AI

1https://nasscom.in/knowledge-center/publications/indian-tech-start-ecosystem-%E2%80%93-march-trillion-dollar-digital-economy2https://nasscom.in/knowledge-center/publications/indias-deeptech-start-ups-next-big-opportunity

76


e are living through an unprecedented moment in human history and embarking upon

an age of intelligence driven by the ubiquitous digitization and breakthrough innovations around emerging technologies.

Artificial Intelligence (AI) has made an impressive progress piggybacking on the exponential increase in computing power and the availability of vast amounts of data. Last year, Gartner1 identified the democratization of AI as a top 10 strategic technology trend. In India, the rapid adoption of AI is aided and enabled by the widespread access to affordable computing, pervasive internet connectivity, and declining data costs.

This pervasive intelligence is not only radically transforming the products and services we use, but also how we communicate, collaborate, and function while evolving the society at large.

The infusion of AI to every business function in transformative ways is helping organizations deliver real business value, resilience, and differentiation while also helping create new business models, foster innovation, and potentially disrupt industries. AI is helping startups create compelling products, deliver highly personalized services, optimize operations, and enhance customer engagement across industry verticals.

Age of intelligence in India

The IntersectionAI enables computing devices to assist with and solve problems in ways that are similar to humans by perceiving, learning, and reasoning. The past few years have seen a leap in practical innovations, including computer vision, natural language processing, sentiment analysis, speech recognition and synthesis, et al, to power applications across a broad range of business cases.

Big DataAt the center of AI is data. The increasing digitization is resulting in the proliferation of what is known as ‘big data.’ The growing amount of information created and consumed online, a slew of online services across the board, and the plethora of personal information gathered via sensors has made limitless data available for AI systems to learn and reason.

Cloud ComputingThe need for digital transformation across industries has catalyzed cloud computing. The cloud-hosted services allow for AI at scale and elastic computing resources to integrate, analyze, and learn from the heterogenous data available on a real-time basis.

Deep LearningThe explosion of use cases for AI has also driven enormous progress in AI algorithms allowing computers to learn deep concepts, relationships, and representations from vast amounts of data and perform intelligent tasks with accuracy comparable to humans.

W

The India storyIn a country like India, apart from achieving business efficiencies, the potential of AI can also help address key socioeconomic causes. According to a 2018 PwC survey2, 71% respondents believed that AI would help humans solve complex problems and help live more enriched lives.

AI adoption in India is at an inflection point. A range of AI-powered solutions have been disrupting the technology enterprises which also cater to multiple clients in diverse sectors. While for the BFSI industry, AI-powered solutions around automation, data analysis, and customer support have the highest impact on the business, the decision makers in manufacturing sector lean towards a mix of machine learning solutions, decision support systems, and automated communications, combined with industrial IoT platforms.

However, unlike large enterprises, Indian startups are born in a digital era-often with AI at the core of their product and operations. Predictive analytics using statistical analysis, machine learning, and data mining is helping startups optimize their services better and deliver highly personalized services to their customers.

Increasingly, AI systems like chatbots, digital assistants, and robots are deployed to enable differentiated, convenient, and seamless customer experience. In fact, AI assistants are now evolving

from being mere assistants to decision support systems that can effectively process information and help in decision making.

AI applications also help automate operations to augment productivity as well as to avoid repetitive tasks and any human errors that might creep in. As the fastest growing economy globally and the world’s third largest startup ecosystem after the United States and China, India has a significant stake in the AI revolution. The AI strategy for India published by NITI Aayog outlines sectors of the economy where AI can play a significant role and guides the government to undertake exploratory proof-of-concept projects for building a vibrant AI ecosystem.

ChallengesOne of the challenges involved in building such an ecosystem is the short supply of AI-ready talent in the country. The lack of technical expertise in core AI skills is one of the key barriers to the effective integration of AI in businesses. However, government and policymakers, academia, and enterprises are collaborating across multiple

initiatives to address this gap and create opportunities that will engage the next generation of AI professionals.

While reskilling existing technology professionals has been driven by technology vendors and the growing ed-tech ecosystem in the country, several technology companies have also partnered with diverse institutions and government agencies to train students and faculty at scale to drive the future of AI in the country.

The road aheadIn 1986, the Department of Electronics (now Ministry of Electronics and Information Technology, or MeitY) launched its ‘Knowledge Based Computer Systems’ project with the assistance of United Nations aiming towards research and capacity development in the domain. Thirty five years later, India has already registered its capability and competitive advantage in the global IT and ITES industry.

According to a 2019 study3, India is on its way to create up to a trillion dollars of economic value from the digital economy in 2025. The pandemic-led digital transformation is bound to accelerate the same. The country is, therefore, in an advantageous position for rapid adoption of AI and other allied technologies. AI-led and AI-powered startups will seize this moment in India’s economic evolution while also solving some of our most pressing social challenges.

1Top 10 Strategic Technology Trends for 2020 – Gartner2Artificial Intelligence in India – hype or reality – PwC (February 2018) 3India’s Trillion-dollar Digital Opportunity – MeitY (February 2019)

According to a 2019 study, India is on its way to create up to a trillion dollars of economic value from the digital economy in 2025

98


I-based systems operate in the realm of probabilities and must, therefore, be continuously

trained, monitored, and evaluated for performance. Maintaining the performance, predictability, and accuracy of AI is required for continued benefit from these systems. Organizations that don’t assess their own characteristics for creating, owning, and operating AI-based systems may be subject to various harmful consequences, ranging from inaccurate systems that silently decay over time to systems that may inadvertently harm organization employees or customers.

AWhy AI maturity is important

Dangers of disregarding maturityOrganizations that rush to adopt AI technologies without regarding their own maturity may face minor to disruptive challenges.

Fairness and bias harms Fairness can be a subjective concept and mean different things to various individuals, cultures, geographies, and countries around the world. However, the concept of fairness as it relates to AI technologies and experiences deals with mitigating bias and the effects that bias can have if it’s not identified or controlled. Users can experience unfair, discriminatory, or prejudicial experiences if bias is not mitigated when training an AI-based system. This may lead to harms of allocation where the system discriminates the access to resources equally for qualifying individuals (for example, a loan

creditworthiness AI-enabled system that discriminates based on biased data sets). Additional harms include classification and representational harms that can negatively impact system users.

Turning the system off Once the act of AI creation is successful, organizations may benefit from the system for a time. However, over time, if the organization is not equally adept and mature enough to handle operationalizing advanced AI technologies, the system may begin to decay. Some organizations have found that understanding the reasons why decay happens and then remediating these situations is beyond their current capabilities. A percentage of these organizations have even rolled back to non-AI-based systems and turned off the AI-enabled digital experience as a last resort to prevent harms but continue operations.

Mature organizations ask, “We know we CAN do things with AI, but SHOULD we?

Mistrust and withdrawal from AI

While the desire for AI-enabled systems is clearly to help achieve objectives and goals, each time a system is not maintained properly, and begins to decay and behave unpredictably, organizational leadership can lose confidence in AI in general. In some cases, this may mean a withdrawal or rejection of AI technologies based on unsuccessful attempts to adopt AI. Organizations or their customers may lose trust in AI as a technology, branding it as unpredictable or too hard to operate. This outcome will push true digital transformation further away and cause the organization to miss out on the powerful impact that AI can have on their business.

The AI maturity modelMicrosoft has worked to define an operational model that helps organizations assess their own attributes that contribute to the adoption of AI technologies. The AI Maturity Model is designed to help organizations gather information related to the core characteristics required for teams and organizations to own AI and help guide adoption of the right AI technologies at the right time. Additionally, Microsoft has compiled prescriptive guidance associated with adopting the right AI technologies for an organization’s current maturity level, while advising on how to increase maturity to embrace more advanced AI capabilities.

The diagram above is a representation of the AI Maturity

Model that describes the maturity levels and some characteristics associated with each.

1. Foundational

Organizations at this level of maturity strive to acquire systems and processes to help make data-driven decisions. These organizations often rely on the talent, instincts, and experience of leaders to make decisions. Within foundational organizations, historical analytical systems are in place, but may not be consulted in favor of an experienced leader’s recommendations. Foundational organizations need to invest in successful projects that focus on fast, iterative experimentation. Investments should also be made to understand AI and how it can further digital transformation.

Experimented andapplied AI● High digitalization ● Design new business models● Achieved a data culture

Emerging data science and operational capability● Understands model lifecycle and management● Building a foundational data architecture

Questioning what AI is and how to apply it● Wrong expectations or disappointment● Low digitalization● Basic analytical capabilities

Hopeful on AI and it’s promise● Digitalization under way● Looking to increase or optimize process● Cautions about disruption

FoundationalApproaching

AspirationalMature

The A I matur i ty mode l

1110


2. ApproachingApproaching organizations have demonstrated the ability to implement solutions using quick iterative sprints and value learning from those efforts. These organizations are poised to embrace rapid experimentation. Operationally, these organizations will invest more in understanding how to implement, monitor, and improve AI over time. Investments should continue in accountability protocols for AI governance; monitoring, orchestrating, and improving AI over time; and infusing ethical viewpoints when deploying AI-based systems.

3. AspirationalOrganizations at this maturity level are focused on shifting culture to empower employees. Employee empowerment increases collaboration, generates ideas for optimization, and helps to create new business models. These organizations are becoming increasingly comfortable with taking risks and are striving to transition away from sequential fixed projects to more iterative projects. Aspirational organizations can adopt configurable AI, which is AI hosted by technology companies like Microsoft. Aspirational organizations should invest in using advanced analytics (predictive and prescriptive) to drive decisions, shift culture toward experimentation, and even investigate custom AI to create new experiences.

4. MatureMature organizations continue to successfully curate AI creation

talent and understand how to apply these resources to several AI initiatives simultaneously. Additionally, the organizations understand how to create digital experiences that are impactful over time. Organizations at this level of maturity should continue to evaluate tool chains for

configurable and custom AI while maintaining operational vigilance when it comes to monitoring, retraining, and implementing AI-based systems. Maintaining AI talent, prioritizing new strategic initiatives, and continuing agile experimentation are required areas of focus for mature organizations.

Lab to land: Challenges and approaches to productizing research

oday, with the world being driven by deep technology and intellectual property, there is an increased

and immediate need for getting research evidence into policy and practice. The importance of getting research used by stakeholders in leveraging the new technology towards higher value, lowered costs, better products, and solving existing problems cannot be overstated, and yet we find that the challenges seem to be overwhelming the practice.

Collaboration between academia and industry is still quite limited, and there is no real-world model that is tried and tested, or widely used. Most of the problems are derived from academic angles and less often driven by industry problems. Post facto, many of the research works do have application opportunities in industry. However, even these don’t get realized as often as they could.

T

Typical industry-research engagement models Industry engages with academia typically for tactical projects or enabling talent acquisition. The project models are:

• Commissioned research projects: This helps reduce the risks of error, omission or misunderstanding and helps improve the quality of the final product.

• Consulting (Industry) projects: In this model, industry has access to the professors and other researchers in the university to discuss their problems and try discussing various modes of working and finding solutions. At the Nurtural Lab in IIIT-H, the industry pays an upfront amount to the university for access to the professors, but cannot influence the work that is being carried out by the researchers.

• Technology licensing: This is when a technology licensing agreement by the licensor authorizes the licensee to use the technology under certain agreed terms and conditions. It is, therefore, a contract freely entered into between two parties (in this case between researchers and companies/ industry/ startups), and contains terms and conditions so agreed.

1312


Low engagement Several independent analysis done over the years suggest that corporate-sponsored research is highly valuable for further innovation. But various factors reduce the probability of research being used in industry products and solutions.

Academic research relies on multiple sources of funding. Most of the funding, however, is from government sources like DST, MEITY and MHRD. A corollary of corporate funding being so low is that globally the research industry engagement is very low. The fact that most of research funding is received from government sources is an indicator of how minimally industry is engaged in research.

Mismatched expectationsThe expectations from industry and the state of research technology in the labs have a few critical gaps.

Conflicting demands from industry and facultyThe industry wants experts to guide them in solving their problems, and with the assumption that the researcher is already working on the same problem. This can be disruptive from the regular work of researchers. The consulting model also has problems owing to a lack of structures and mechanisms. Academia setting up collaborations with the industry also tend to be vigilant in their mission to generate and transfer knowledge.

Evolving scope in industry requirements

Many of them being application projects, they are not well defined. Most likely being defined from the existing business, they also require researchers to have a strong understanding of the domain. Problems like access to internal systems, APIs, and people, among others, are common problems that commission projects face.

Hard to discover what is licensable

Licensing starts with lack of information on technologies.

Academic institutions have a well-documented list of research works and research publications. While this is a great source of research insights, it does not provide an adoption view of the same.

Research works excites, but not ‘building’ productsA huge challenge is also that researchers scarcely want to start companies of their own, making the startup scene coming from researchers an almost non-existent option. This, of course, leads to the ever-increasing number of unused patents held by universities and public research institutions.

Where are the prototypes? Prototypes of research works are key in helping industry understand the possibilities. Typically, in the technology licensing mode, research labs concentrate only on the research work and not much on prototypes, although the industry looks for prototypes when trying to license/patent the research. Building prototypes can take up to six months, and while the researchers would prefer moving onto the next research, the industry also prefers reading the papers and building the prototype on their own. The industry too, at times, is not sure about the knowledge that can be received from academia. They are not sure what they can ask the labs and research faculty and how to get help from them.

Experiments at IIIT HyderabadExperiences of IIIT Hyderabad in increasing industry-academic engagements:

Successful collaboration between industry and academia could have manifold results. Academia will benefit from a better understanding of the real world. The industry also gains through new market opportunities explored by leveraging the state-of-art research. As the Indian tech industry actively seeks growth through value creation, innovation and new tech products are imperative. And academic collaboration is key to making this happen.

Authored by IIIT-Hyderabad

Technology Catalog, discovery of technologies and opportunities:• Simplify discovery of

technologies and enhance through prototypes research reports

• Create product prototypes from research work

• Enrich research lab experimental prototypes

• Create market prototypes of research output, through entrepreneur in residence

Product Labs, translating research into market facing prototypes: • Research teams starting

product startups

• Co-innovate new tech prototypes with industry (labs knowledge, industry’s engineers)

• Co-creation of products with startups

Deeptech startup advisory, adding research insights to startups: • Prototyping/hackathon

events, to identify products based on research works

• Bring your own project (BYOP)

• Student deeptech product ideas

• Invert it by adding research to startups

• Add research advisory to startup acceleration programs

1514


idely popular in China, Xiaoice has authored a collection of poems, released dozens of songs, and hosted

several television and radio programs.

Xiaoice is an AI-powered chatbot.

Even though digital assistants like Cortana, Siri, Google Assistant, and Alexa have long exhibited a high IQ in following instructions and providing information, Xiaoice combines that with a high emotional quotient. Which is why, millions of youngsters in China turn to Xiaoice to talk about their everyday lives and personal issues.

Intelligent platformsStartups can infuse intelligence into their applications and agents using in-house built machine learning and deep learning models. However, that approach is resource-intensive and can be often unwieldy for startups or SMBs. To avoid reinventing the wheel, they adopt AI services offered by cloud service providers with pre-trained AI models to augment their products and services or accelerate the operations.

These intelligent platforms bring AI within the reach of developers of all skill levels and to organizations of any size. These cognitive services offer a collection of domain-specific pre-trained AI models in vision, speech, language, search, or decision-making to solve common problems and make business applications smarter and more intuitive to build a competitive advantage.

Intelligent platforms, apps, and agents

• Vision allows apps and services to accurately identify and analyze content within images and videos.

• Speech services can convert spoken language into text or produce natural-sounding speech from text.

• Language services can understand the meaning of unstructured text or recognize the speaker’s intent.

• Knowledge services create rich knowledge resources that integrate into apps and services.

• Search enables apps and services to harness the power of a web-scale search engine to find information.

WFor example, Ola Cabs uses real-time data from rides to automatically detect irregular trip activity, including prolonged stops and unexpected route deviations, with its AI-based safety feature called ‘Guardian.’ The alerts from ‘Guardian’ are flagged off in real-time to the firm’s round-the-clock safety response team that reaches out to customers and drivers to confirm if they’re safe.

Then there are AI apps and agents. An AI app is simply a web or mobile application that is infused with AI capabilities, such as vision or language processing, while an AI agent is a machine program that uses AI capabilities to interact with a human user.

An intelligent agent can make decisions or perform a service based on its environment, user input, and experiences. These programs can be used to autonomously gather information on a regular, programmed schedule or when prompted by the user in real time. AI virtual assistants, like Alexa, Cortana, and Siri, are examples of intelligent agents as they use sensors to perceive a request made by the user and automatically collect data from the internet without user’s intervention.

Conversational AIThe growing adoption of chatbots and conversational AI underscores our essential nature as social beings. Instead of filling forms or navigating through multiple pages on a website, chatbots can offer a variety of services using speech or text with the ease of a natural conversation.

The chatbots owe their capabilities to the advances in natural-language processing (NLP). Since the 1950s, scientists have been working to solve the problem of processing and analysing the complexities of human language, which is often chaotic and vague. However, the recent advances in machine learning and the availability of vast amounts of conversational data have aided the immense progress in NLP.

The defining characteristic of a chatbot is its conversational interface. Instead of relying on traditional graphical interfaces, conversational AI allows users to communicate with applications more naturally-through typing or speaking-via chatbots. A bot, or an AI agent, is a computer program designed to mimic the actions of a person - execute commands, reply to messages,

or perform routine tasks, either automatically or with minimal human intervention. A chatbot is, of course, an umbrella term for several variations, including virtual assistants and virtual agents.

A chatbot can be deployed across diverse channels - from web pages to consumer-focused messaging services like Skype to business communication services like Microsoft Teams or internal portals and mobile apps.

AskDISHA, the AI-powered chatbot of Indian Railway Catering and Tourism Corporation (IRCTC), uses AI, ML, and NLP to answer passenger queries and has helped improve satisfaction of customer interactions by 70%. Developed by conversational AI platform CoRover on Microsoft Azure, the virtual assistant now provides customers with correct instant responses to repetitive queries with zero wait time instead of the lengthy process involved via phone, email, or interactive voice response. Since its inception in October 2018, the AskDISHA chatbot has handled over 10 billion interactions and customer queries across other channels like social media, phone calls, and emails have been reduced by 70%.

AskDISHA, the AI-powered chatbot of Indian Railway Catering and Tourism Corporation (IRCTC), uses AI, ML, and NLP to answer passenger queries and has helped improve satisfaction of customer interactions by 70%.

1716


While basic chatbots recognize and respond to trigger phrases, they can also be created directly from existing web pages, such as FAQs, to offer information that is already available in a more convenient manner. Chatbots can also connect to backend systems as needed or escalate to a live human agent if it doesn’t recognize any trigger phrases.

Chatbots today offer a sophisticated, multiturn conversation experience with seamless interruption handling, cancellations, and context switching. They support both text and speech and also speak multiple languages. The chatbots can benefit from reusable conversational components and can be extended with prebuilt components, dialogues, and language models. They can also integrate with existing products and services via prebuilt connectors and can be used to build custom workflows or complex scenarios.

Businesses can also observe detailed reporting on the bot’s usage and performance with real-time reporting around successful interactions, escalations to a human agent, the topics that are executed most often, and the failures or abandonments. They can also capture insights about new trigger phrases as well as new topics to be added to better address the user needs.

Democratizing intelligenceDue to physical restrictions necessitated by the COVID-19 pandemic, Paisabazaar, one of India’ largest marketplaces for

financial products, tried to push presence-less lending enabled by end-to-end digital processes.

By using cognitive services, the KYC module of the Paisabazaar stack helps verify the identities, locations, and liveness of the applicants. The income of the applicant and the employment too is validated digitally by the stack and the bank statements are analyzed digitally using AI to determine financial health of the applicant which helps the Paisabazaar’s lending partners in faster decision making with disbursals taking place within 24 hours.

The intelligent platforms and chatbots help startups democratize AI. Startups can accelerate their digital transformation by applying AI capabilities to their business and using knowledge mining and machine learning to make better sense of and utilize the available data. Cognitive services allow developers to easily apply pre-

trained AI models in their own applications without requiring machine-learning expertise.

Chatbots can offer conversational interfaces for various transactional scenarios in banking, travel, ecommerce, and other sectors to process customer requests using text and voice better contextually. There are also information bots that can answer questions defined in a knowledge set or frequently asked questions as well as more open-ended questions via web search.

Large, modern enterprises also build powerful productivity bots to streamline common work activities by integrating line-of-business apps and external systems. Not only does this help reduce costs and improve customer satisfaction by providing automated responses to common requests, they also free up time of human agents to handle higher-value concerns or those that can’t be automated.

Yellow.ai, one of the world’s leading conversational AI platforms, leverages Azure Cognitive Services to transform its voice automation solution and NLP tools. This allows Yellow.ai o increase the accuracy of its voice bot solutions and offer a more human-like voice assistant platform that is capable of understanding end-user intent better. Such voice-based virtual assistants enable brands to increase their sophistication and help enterprises across sectors in enhancing consumer experience automation.

Knowledge miningriven by advances in cloud, data, and artificial intelligence (AI), digital transformation is

fueling unprecedented change across organizations. For most organizations, a lack of data is no longer a primary challenge and the expanding volume of data generated compounds this predicament. In fact, one of the primary difficulties faced by businesses today is how to extract actionable information and business insights from this massive influx of unstructured data.

Such unstructured data – in a wide variety of content types – does not have a predefined data model, making it more challenging to search and analyze. The harnessing and refining of information from this unstructured data, across all industries, can help realize significant benefits. However, transforming all this information into insights requires extensive time, resources, and data science expertise.

This is where knowledge mining comes into the picture. Knowledge mining is an emerging category in AI, which refers to the orchestration of a series of AI services for comprehending,

perceiving, calculating, organizing, and reasoning to uncover latent insights in vast amounts of data. It allows organizations to consume and process all types of content across multiple locations and formats and extract new insights and knowledge from the content by using deeply integrated prebuilt AI technologies across vision, speech, and language. The end goal is, of course, to surface this knowledge through a variety of tools such as search, data visualization, and business applications to make it available to end users.

Knowledge mining represents an enormous opportunity for companies to gain actionable business insights and is heralding a new wave of AI-powered digital transformation with knowledge mining at its heart. It is fundamentally changing how organizations make sense of real-world information and hence are seeing tremendous impact resulting in increased business performance and profitability. More than two-thirds (68%) of respondents to a 2019 Harvard Business Review Analytic Services survey believe knowledge mining is important to achieving their companies’ strategic goals.

D Challenges Organizations frequently use unstructured data for business performance analysis, knowledge management, and routine business processes and workflows. However, many organizations struggle with the growing volume of unstructured information-information that is critical to business functions but not readily visible or available to process.

The growing footprint of unstructured information in the workplace also creates scalability challenges as well as makes it difficult to explore and understand in a timely manner. The methods to extract and understand unstructured information too are outdated and insufficient. Additionally, the manual, time-intensive methods are also error-prone and expensive. Per the earlier cited Harvard Business Review Analytic Services survey, 77% of respondents are using manual methods to handle unstructured information, and those methods will quickly be outpaced by the growth of data and potential use cases in which this information could provide great value.

The issue is exacerbated by the fact that the value from unstructured information remains trapped in traditional formats - in spreadsheets, PDFs, presentations, and documents that are the most common forms of semi-structured and unstructured information. Then, there’s also industry-specific formats such as computer-assisted design (CAD) files.

According to a Harvard Business Review Analytic Services survey, 77% of respondents are using manual methods to handle unstructured information.

1918


Benefits of knowledge miningThere is a definite need to automate the understanding of data from the unstructured information as well as use it for analytics, make it searchable, or to visualize it in a dashboard.

Knowledge mining can provide the ability to automatically categorize and curate vast streams of content that would be far too labor intensive to handle manually. By using a variety of AI technologies in combination, knowledge mining can transform data into a comprehensible structure which can expand visibility into business processes and create new knowledge and actionable insights. With knowledge mining, it is now possible to train a system to recognize the key data to extract from a data dump consistently as well as extract metadata from scanned documents that was never available in a searchable format before.

Amway, the American multi-level marketing company that sells health, beauty, and home care products, has built a cognitive search driven platform to power their applications and allow sellers to quickly extract product information from their expansive pool of sources. The search solution ingests and applies powerful machine learning algorithms before indexing information for search helping customers with accurate, expedient, and holistic information about the products they want to buy.

While knowledge mining can bring both immediate and recurring cost savings, as the trend evolves, more organizations will be able to differentiate themselves based on their external and internal data and automation in workflows. Knowledge mining can also surface more of that relevant information and highlight risks and anomalies that might otherwise be lost buried in the data and therefore enable more effective decision making and more efficient work processes.

How it worksIn layman terms, knowledge mining is the process of applying a series of AI services to extract information and context from structured and unstructured data. It involves the ingestion and enrichment of data followed by the exploration and analysis of this newly enriched, structured data.

IngestKnowledge mining begins with the process of aggregating raw data, whether structured or unstructured, from various siloed sources and locations into a persistent, centralized data store. The ingested data is typically given a standard structure based on the information extracted via ‘document cracking’, the process of extracting or creating text content from non-text sources.

EnrichOnce the raw data has been ingested and cracked, the next step is to enrich the data using AI to identify patterns, obtain information, and gain understanding from the

unstructured data. These AI enrichment models can be either pre-trained or custom models around natural language processing and computer vision services.

Explore & AnalyzeThe final step is to expose the newly enriched, structured documents, so they are accessible for exploration and analysis. While exploration is the process of reviewing the added enrichments to learn more about the data, analysis usually refers to the application of analytics tools to gain a deeper understanding of the enriched data.

Icertis recently enhanced its contract management platform for enterprises businesses with cognitive search and now uses AI-infused search to easily uncover hidden insights in contracts. This not only reduces time in contract negotiations, but also lowers risk, improves contract compliance, and increases revenue.

Knowledge mining works by orchestrating this overall enrichment pipeline. By ingesting a variety of unstructured data from distributed data stores, the use of cognitive functions and machine learning can present that data in a structured format that can be surfaced, analyzed, and illuminated to solve real business problems.

To develop its AI-based business English proficiency test, iMocha decided to do a mix and match using its own models for few evaluation points and use Azure Cognitive Services like Text Analytics, Speech Services, Custom Text, Azure Video Indexer, et al

to arrive at a reasonably accurate score of a candidate’s English proficiency within reasonable time.

The way forwardKnowledge mining is the next wave of AI, generating a dynamic understanding of relationships and patterns in a corpus of information. It enables businesses to discover patterns and relationships among previously disparate data points in a variety of channels. The extracted insights help businesses make better-informed decisions, automate business processes, and identify risks and opportunities.

Skit, an Indian SaaS voice automation startup, offers a next-gen multilingual voice assistant that enables enterprises to automate their contact centre operations. It uses a conversation monitoring technology that detects any instances of bad conversations in a user-bot interaction allowing for remedial action before it can make a significant impact on the customer experience.

Knowledge mining ensures that information is an asset and not a burden and helps solve three of the key challenges of handling unstructured information: time, scale, and insights. Microsoft estimates that 80% of business data is unstructured, making knowledge mining capabilities necessary for digital transformation. It provides organizations an opportunity to effectively, efficiently, and contextually understand all their information, enabling them to better engage with customers, transform products and services, and optimize operations.

2120


AI for developers and data scientists

ata scientists and developers are key to turning AI ambitions into reality. Armed

with the right set of tools, they have an opportunity to play an integral role in growing their organisation’s AI capabilities.

Microsoft empowers developers and data scientists to be AI leaders with a comprehensive set of AI services and learning resources. Building on nearly three decades of investment, Microsoft Research continues to achieve breakthroughs across vision, speech, and language. Those breakthroughs are built into our Azure AI services, so developers and data scientists can build on the cutting edge while still using familiar coding languages and frameworks.

Azure AI offers a range of technologies suited for different business use cases and skill levels, from cognitive APIs and drag-and-drop model training to advanced data science tools. We are committed to an open, flexible approach where users can work with the tools and frameworks of their choice and deploy AI across the cloud and the edge. Our vision is to help our customers invent with purpose, with productive, mission-critical, and responsible solutions.

D

Building custom machine learning modelsAzure Machine Learning is ideal for creating custom machine learning (ML) models. It’s an enterprise-grade platform that streamlines the end-to-end machine learning lifecycle, including data preparation, training, testing, deployment, and ongoing model monitoring and management. It empowers developers and data scientists with experiences for all skill levels, from a no-code drag-and-drop interface (Azure ML Designer) to code first experiences with best-in-class support for open-source frameworks and languages.

Azure Machine Learning helps data scientists accelerate model development with automated ML, which automatically creates models in a few steps. But this doesn’t mean it’s a black box; it includes model interpretability that helps users evaluate and understand why a particular model was recommended. Users also get access to industry-leading MLOps capabilities, or DevOps for machine learning, which help accelerate and scale the ML lifecycle and make it easier for data scientists, developers, and IT teams to collaborate and stay on the same page.

Knowledge miningBusinesses collect a staggering amount of data every day, largely in unstructured formats like PDFs, images, videos, audio files, and Office files. Using Azure Cognitive Search,

developers can ingest disparate structured and unstructured content then use AI services to uncover insights. In addition to built-in AI capabilities, developers have the flexibility to integrate their own custom models to identify information specific to their business, like legal clauses, industrial parts, or pharmaceutical terms. The AI output can then be used for a number of end-user applications like search, business applications, or analytics. Ultimately, Azure Cognitive Search helps stakeholders find essential needles in haystacks of files, make better-informed decisions, identify risks and opportunities, and much more.

Adding intelligent experiences to appsUsed by over 1 million developers, Azure Cognitive Services make it easy to build intelligence into apps. Cognitive Services include over 25 pre-built AI models in categories like vision, speech, language, search, and decision. Without machine learning expertise, developers can deploy AI for various use cases using familiar programming languages.

All it takes is an API call to give an app the ability to communicate with users in natural language, identify content in images, detect anomalies, translate speech or text, and much more.

One of the most common use cases for AI in the enterprise is chatbots or virtual agents. These solutions can provide customers with on-demand, personalized service across channels or help employees find information more quickly. With Azure Bot Service, developers can create anything from a simple Q&A bot to a branded virtual assistant for sophisticated scenarios.

https://azure.microsoft.com/en-us/overview/ai-platform/

https://azure.microsoft.com/en-us/services/machine-learning/

https://azure.microsoft.com/en-us/services/machine-learning/automatedml/

https://azure.microsoft.com/en-us/services/machine-learning/automatedml/

https://azure.microsoft.com/en-us/services/machine-learning/mlops/

https://azure.microsoft.com/en-us/services/search/

https://azure.microsoft.com/en-us/services/cognitive-services/

https://azure.microsoft.com/en-us/services/bot-services/

https://azure.microsoft.com/en-us/services/bot-services/

2322


Developing AI skills

ased on their level of AI maturity, organizations can be classified as AI Beginners (where

AI has not been explored or used in any way), AI Intermediates (that are exploring or experimenting with AI in a limited capacity), and AI Leaders (that have AI at the core of their overall strategy).

AI-leading organizations have a distinct advantage over laggards when it comes to using AI to make operations more efficient. However, when it comes to improving customer experience or innovating through the delivery of new services, AI leaders are twice as likely as the least mature segment to benefit from applying these breakthrough technologies.

AI-leading firms seem to have fostered a virtuous cycle between AI investment and upskilling, which allows them to cement their lead over slower-moving

Mitra Azizirad Corporate Vice President, Microsoft AI & Innovation

firms. Their leadership in AI results in successful AI projects, leading to demand from across the organization to use more AI technologies. The increasing skill levels in employees creates a more motivated and engaged workforce, keen to use more AI tools and continue to strengthen those skills. This fosters a more vibrant culture; and creates a

B The true potential of AI is only realized when AI experiences are fully democratized for every single employee. No matter where you are on your AI journey, every employee can begin using AI solutions and tools today, in ways that enable them to extract more insights relevant to their roles and enhance their productivity.

willingness to broaden these new approaches across the organization, leading to more success.

According to Microsoft’s global research, as many as 84.4% of AI-leading firms claim to get value from AI, compared to 58.9% of businesses at an early stage in their AI journey.

Companies generating the most business value from AI are prioritizing skills as much as technologySuccess with AI implementation is not simply about introducing more and more AI. Instead, AI-leading businesses are remarkably focused on cultivating the skills of their people. The clear majority (93.8%) of senior executives at AI-leading organizations indicate they are actively building the skills of their workforce or have developed plans to do so. Their employees corroborate this, with 70.2% indicating that they are confident their employer is preparing them for a world in which AI is ubiquitous, and 64.6% confirming that they have already taken part in reskilling initiatives.

Employers are prioritizing a wide range of skills. Technology-centered skills are a large focus, with employers looking to bolster skills such as advanced programming and advanced data analysis. Business leaders are also keen to cultivate softer skills such as advanced communications and negotiations, as well as leadership and management.

Insights for startup leaders

• Benchmark your business: What does your skills mix look like in comparison to businesses at a similar stage of AI maturity, and those further ahead?

• Have confidence in the return on AI and skills: When assessing business outcomes and value, consider the contribution of technology, skills, and culture. Apply those insights to both optimizing existing AI programs, as well as informing future investments in skills and technology

2524


Leading businesses are using AI to augment human ingenuityCompanies seeing the most value from AI are not just generating value from automation and operational efficiency-they are leading on augmentation that unlocks the ingenuity of their employees. Among the businesses seeing the most value from AI, 84.7% of senior executives and 62.1% of employees consider themselves to be an ‘augmented worker’-where AI supports them in their job, from helping manage simple tasks and processes through to providing insights and aiding decision-making. This focus on human ingenuity is making a major contribution to the business value that AI-leading firms are gaining from AI.

However, AI-leading firms are driving deeper value from greater innovation empowered by augmentation – through things like new product and service development and improving the customer experience. As AI continues playing a larger, strategic role in every business, senior executives expect an increase in demand for people who can work with AI-powered technology effectively.

Importance of a “learn-it-all” workplace cultureAs companies become more advanced in their use of AI, a cultural change takes place in parallel as people embrace a learn-it-all mindset. The overwhelming majority of workers (91.7%) indicate that they are highly motivated to acquire or deepen AI-relevant skills. There also seems to be a correlation between AI maturity and a company culture where new ideas are welcomed and supported. Senior executives and employees at AI-leading firms are three times more likely to report working in an innovative culture where the idea of continuous change is embraced.

Actionable insights for startup leaders

• Develop a holistic skilling strategy to benefit from both AI automation and augmentation

• Actively seek out and learn from examples of how others are unlocking human ingenuity with AI

• Understand where teams want to use time freed up by AI and help channel how they reinvest it

Actionable insights for startup leaders

• Your teams are highly motivated to learn. Consider a flexible, continuous learning program

• Communicate effectively about your plans to equip employees with the key skills they need in a world of AI

• Leverage free resources: there’s a wealth of free educational content available online

ew disruptive business models are springing up across multiple industries, and they have AI at the core. Early adopters of AI at the

strategic level are already leveraging it for business and competitive advantage. A successful AI strategy must consider cultural issues as well as business issues. Becoming an AI-ready organization requires a fundamental transformation in how you do things, how employees relate to each other, what skills they have, and what processes and principles guide your behaviors. This transformation goes to the core of an organization’s culture, and it’s vital for organizations to tackle such transformation with a holistic approach.

Being a data-driven organization

Empowering people to participate in the AI transformation, and creating an inclusive environment that allows cross-functional, multidisciplinary collaboration

Creating a responsible approach to AI that addresses the challenging questions AI presents

Share data across your organization

Adopt rigorous data practices

Fostering an AI-ready culture requires:

Building a successful AI culture

N

Being data-drivenThe first step is to ensure that you have the best and most complete data, and that you can reason over your entire data estate. Due to data ownership or storage issues, most organizations generate, organize, and use data in a siloed manner. While each department may have a good view of the data coming from their own

processes, they may lack other information that could be relevant to their operations.

By sharing data across the organization, the sum becomes greater than the parts. It’s no longer each piece of data that matters, but what that data adds up to: a unified view of the customer. With that unified view, you can make better decisions, act more effectively, and provide a better customer experience.

The quality of the data is key to creating next-level experiences for customers and to successful AI. An AI model is only as good and complete as the data it can operate on and learn from.

All this needs to be supported by strong, inspirational leadership and clear ethical standards and governance.

2726


Being empowering and inclusiveFostering an AI-ready culture means empowering people to be part of the AI transformation. Fundamental to empowerment is enablement: giving people the space, resources, security, and support to improve what they do with AI. Empowerment also requires allowing room for errors, encouraging experimentation and continuous improvement, helping people get the knowledge and the skills they need, and of course celebrating and acknowledging success.

It also means creating an inclusive environment – one that is predicated on the willingness and

ability of employees to work in cross-functional teams that cut across organizational boundaries. Furthermore, it means making those who best understand the business a central piece of your transformation process.

Data scientists working in isolation often create models that lack the business knowledge, purpose, or value that would make them an effective AI resource. Similarly, business people working in isolation lack the technical knowledge to understand what can be done from a data science perspective. But by enabling cross-functional teams that include both data scientists and the business employees closest to the business need, you can create powerful and effective AI solutions.

Provide resources

Create a culture of sharing and collaboration

Make those that know your business a central pieace of your transformation process

Being responsibleThe third key element of an AI-ready culture is fostering a responsible approach to AI. As AI continues to evolve, it has the potential to drive considerable changes to our lives, raising complex and challenging questions about what future we want to see. Organizations need to ask themselves: How do we design, build, and use AI systems to create a positive impact on individuals and society? How can we ensure that AI systems treat everyone fairly? How can we best prepare the workforce for the new AI era?

Responsible AIFor AI to benefit everyone, it must be developed and used in ways which warrant people’s trust. Over the past few years, principles around developing AI responsibly have proliferated and, for the most part, there is overwhelming agreement on the need to prioritize them. While principles are necessary, having them alone is not enough. The hard and essential work begins when you endeavor to turn those principles into practices.

Microsoft has put forth six Responsible AI principles that are reasonably comprehensive, mutually exclusive, and very powerful. These principles are used to contextualize every single AI use-case.

Responsible AssessmentThis includes identifying high priority areas within your AI development, building a way to track and review the process, and securing approvals.

Responsible DevelopmentThis includes everything from data collection and handling, to ensuring fairness in performance and representation.

Responsible DeploymentThis includes reinforcing practices and empowering people to use AI responsibly through documentation, gating, scenario attestation, and more.

2928


AI for Good Microsoft’s AI for Good initiatives combine Microsoft’s technology and resources in AI and data science with the talent and expertise of groups around the world in areas such as environmental science, humanitarian issues, accessibility, health, and cultural heritage. Microsoft works deeply with selected non-governmental organizations, and humanitarian organizations through financial grants, technology investments, and partnerships.

From counting penguins in Antarctica to identifying elephants in Africa, AI tools can help in biodiversity conservation, as companies like Gramener demonstrate.

Penguins inhabit one of the most secluded parts of the planet, yet human activity is threatening their existence. There is very little data available, and manned missions are difficult. Oxford University’s Penguin Watch Project set up time-lapse cameras to monitor penguin colonies and click a photo every hour over many years. Using this data, Gramener trained their AI

and ML-based model to identify a penguin from different angles and perspectives. Gramener used a Convolutional Neural Network model to come up with a solution that can identify and count penguins with a high degree of accuracy. The team is now working on the classification, identification and counting of other species using similar deep learning techniques.

We need to rapidly prototype AI applications for conservation because we don’t have time to wait for humans to annotate millions of images before we can answer wildlife population questions.

AI for EarthAI and cloud software can help those working to solve global environmental challenges. Microsoft supports organizations that are applying AI to environmental challenges, by helping them harness the full power of cloud computing.

AI can help in studying climate change and air pollution to predict TB outbreaks.

Dr. Nupur Giri, a computer science professor at an engineering college in Mumbai, is working with her students to use AI to predict how airborne diseases can propagate based on climate conditions, air quality, and population density in an area. “The initial results are good, but we are currently testing

multiple machine learning and neural network models to improve the accuracy of their prediction for every district in India,” Dr. Giri says. Once done, they plan to provide a data visualization dashboard to help those working on eradicating TB make more informed decisions. Predicting the hotspots for TB is just the beginning. The team is hopeful that once they crack the model, they will be able to employ it for other diseases too.

AI for HealthAI can empower nonprofits, researchers, and organizations tackling some of the toughest challenges in global health. Microsoft provides access to AI and expertise to accelerate medical research to advance the prevention, diagnoses, and treatment of diseases; increase our shared understanding of health and longevity to protect against global health crises; reduce health inequity and improve access to care for underserved populations; and support fundamental research capabilities, including data collaboratives and differential privacy.

AI for Humanitarian ActionMicrosoft supports nonprofits and humanitarian organizations working for disaster response; refugees and displaced people; human rights; and the wellbeing and safety of women and children around the world.

The current data available on childhood undernourishment around the globe is inaccurate as manual weight measurement scales often lack standardization. Moreover, it is difficult for the human eye to detect if a child is suffering from malnutrition.

In 2019, Welthungerhilfe, one of the largest private aid organizations in Germany, launched a project in India to help address malnutrition

in children with “Child Growth Monitor” – a cloud-based, smartphone application powered by Microsoft Azure and AI services.The application could detect malnutrition and enable health workers identify and provide care to children struggling from chronic undernourishment.

Welthungerhilfe’s pilot project using Microsoft AI to tackle malnutrition in India.

3130


“Seeing AI” is a free app for visually impaired people that narrates the world around them. It was launched in English in 2017 and is available in 70 countries. Microsoft has expanded its support to seven other languages: Italian, Turkish, Dutch, German, French, Japanese and Spanish. The app leverages AI technology, and has helped people with more than 20 million tasks so far. A new feature was recently added in response to user requests – the ability to explore photos by touch and hear descriptions of images. This feature makes it very easy to know the information contained in photos.

Florian Beijers, a web developer in the Netherlands, who was born blind, believes that “Seeing AI” in Dutch will be a “game changer” for making daily tasks easier in a country where accessibility is not as good as it should be. He says that menus in restaurants often aren’t accessible. “There’s not always a web-based version of it you can look up on your phone.” Beijers says that, for him, Seeing AI’s Document channel, which provides audio guidance to capture a printed page and recognizes the text, along with its original formatting, is very useful, especially for a menu.

Saqib Shaikh, Microsoft project lead and co-founder of Seeing AI, is also blind. “More than the technology itself, the thing that has really touched me is the way that people have taken the features of Seeing AI and incorporated it into their personal lives,” Shaikh says. “I get the most joy hearing from customers about how they’re using Seeing AI.”

AI for AccessibilityAI expands access to learning materials, language development, and assistive technology, making education more engaging for everyone.

By supporting the development of professional skills and evolving inclusive hiring and economic recovery, technology can help solve unemployment challenges.

Disability can affect any of us, or those we love, at any time. Technology can be a powerful tool to connect people.

AI makes devices smarter, allowing people with disabilities to use appliances, healthcare, and transportation more easily and safely.

Microsoft is helping preserve Inuktitut, an Inuit language spoken in Nunavut, Canada.

Nunavut – which means “Our land” in Inuktitut, the primary dialect among the Inuit languages – has been home to an indigenous population for more than 4,000 years. Today, more than 80% of the population is of Inuit descent. While the territory is huge, the population is not: 39,000 people.

Working together toward a common goal is important in Nunavut and part of Inuit culture. Taking care of each other and respect for elders are key. Inuktitut, the first language of 70% of the residents of Nunavut, is integral to the teamwork and the traditional passing of knowledge.

Microsoft worked closely with the Government of Nunavut and added Inuktitut text translation to Microsoft Translator. Using Microsoft Teams, the Government of Nunavut can reach subject matter experts and consultants outside of the territory more easily than before, paving the way for expanded possibilities in health, and for students to get tutoring and school courses.

AI for Cultural HeritageAI is empowering people and organizations dedicated to the preservation and enrichment of cultural heritage. Microsoft supports projects that celebrate the people who have made significant impact throughout history; use digital tools to preserve important monuments and sites; engage with communities around the world for language preservation; and create ways for collections and archives to be more easily accessed and enjoyed.

3332


icrosoft has put forth six Responsible AI principles that are reasonably

comprehensive, mutually exclusive, and very powerful. These principles are used to contextualize every single AI use-case.

FairnessSociety is unfair and biased in many, different ways. We need to ensure that the AI systems that we develop and deploy reduce unfairness in society rather than keeping it at the same level or making it worse. Fairness relates not just to the technical component of the AI system but also to the societal context in which the system is deployed. It is thus a sociotechnical challenge. We, therefore, need greater diversity among the very people who develop and deploy AI

systems. Moreover, we need to take care at every stage of AI development and deployment that that the assumptions or decisions made by teams do not introduce biases.

Reliability and SafetyAt Microsoft, we make sure that the AI systems we develop are consistent with our values, principles, and design ideas. AI systems and models should not create any harm in the world. And if there are situations where they could make mistakes, the risks and harms should be well understood and quantified and shared with the end-users. This is our guiding principle with respect to reliability and safety, and it applies to every AI product we have. Small mistakes can pile up when a system gets used many times across large groups of people. Hence, reliability and safety of AI and Machine Learning models,

Fairness

Reliability and Safety

Privacy and Security

Inclusiveness

Accountability

Transparency

especially when they concern human lives, should be taken very seriously.

Privacy and SecurityPrivacy is a fundamental right. AI and Machine Learning add new complexity to systems and increase the reliance on data to develop and train those systems. We need to protect this data; ensure that it’s not leaked or disclosed. One of the ways to do this is by not removing the data from a customer’s device and running the AI/ML models locally on the device. When we think about the security of AI systems, we have to think about where and how the data is coming from; whether it is user-submitted data or a public data source that’s being used; how to prevent it from being corrupted; and having systems for detecting changes in the data or any indications that someone is trying to influence the results of the system.

Responsible AIFor AI to benefit everyone, it must be developed and used in ways which warrant people’s trust. Over the past few years, principles around developing AI responsibly have proliferated and, for the most part, there is overwhelming agreement on the need to prioritize them. While principles are necessary, having them alone is not enough. The hard and essential work begins when you endeavor to turn those principles into practices.

MInclusivenessMicrosoft seeks to empower and engage the full spectrum of communities around the world, leaving no one out. We make sure that we are intentionally inclusive and intentionally diverse with our approach to AI.

If we think deeply about how we can design for the underserved three percent, we can usually meet the needs of the remaining 97 percent at the same time.

We need to ensure that underserved minority communities are involved at every stage of the AI journey, from design to deployment, to make sure that we are not takingan ableist perspective for the population we intend to serve.

AccountabilityDespite the complexities of AI and ML models and the fact that new technology can be unpredictable and hard to interpret, we are still accountable for how the technology we develop impacts the world. Accountability is not only about consistently enacting one’s principles but also helping customers and partners be accountable too. Microsoft, for instance, has a set of principles that guides how we develop and sell facial recognition, and advocate for its regulation. For every practice and process that we follow across the company, we take high-level principles, such as our commitment to fairness and privacy, and think about how our development teams can view every stage of the product

development lifecycle in the context of those principles.

TransparencyTransparency and intelligibility can help us achieve a diverse range of goals – mitigating unfairness in Machine Learning systems; helping developers debug their AI systems; gaining the trust of our users, and much more. There are two sides to transparency. In part, transparency means that the people who create AI systems should be open about how and why they are using AI and about the limitations of their systems. Transparency also means that people should be able to understand the behavior of AI systems – the reason why it is often referred to as interpretability or intelligibility.

3534


How Microsoft turns AI principles into practice

Governance as a foundation for compliance

Our responsible AI governance approach borrows the hub-and-spoke model that has worked successfully to integrate privacy, security and accessibility into our products and services. The hub comprises three groups of people who work together to set a consistent bar for responsible AI across the company and they empower our “spokes” to drive initiatives and be accountable for them. The spokes are Responsible AI Champs in engineering and sales teams, who raise awareness about Microsoft’s approach to responsible AI and the tools and processes available.

Developing rules to enact our principles

In the fall of 2019, we published internally the first version of our Responsible AI Standard, a set of rules for how we enact our responsible AI principles underpinned by Microsoft’s corporate policy. Today, we are previewing version two of the Responsible AI Standard with our employees. For each requirement in the Responsible AI Standard, we will build out a set of implementation methods that teams can draw upon.

Drawing red lines and working through the grey areasWe have created create a process for ongoing review and oversight of high-impact cases and rising issues and questions. Our sensitive uses process requires that use cases that meet our review criteria are reported to our Office of Responsible AI for triage and review. This review process has helped us navigate the grey areas that are inevitably encountered and, leads in some cases to new red lines.

Evolving our mindset and asking hard questionsWe have developed training and practices to help our teams build the muscle of asking ground-zero questions, such as “Why are we building this AI system?” and “Is the AI technology at the core of this system ready for this application?” We have seen teams wonder whether being responsible will be limiting, only to realize later that a human-centered approach to AI results in not just a responsible product, but a better product overall.

Pioneering new engineering practicesIntegrated systems and tools help drive consistency and ensure that responsible AI is part of the everyday way in which our

engineering teams work. We are therefore embarking on an initiative to build out the “paved road” for responsible AI at Microsoft – a set of tools, patterns and practices that help teams easily integrate responsible AI requirements into their everyday development practices.

Scaling our efforts to develop AI responsiblyAs we look ahead, we’ll focus on three things – consistently and systematically enacting our principles through the continued rollout of our Responsible AI Standard; advancing the state of the art of responsible AI through research-to-practice incubations and new engineering systems and tools; and continuing to build a culture of responsible AI across the company.

Integrated systems and tools help drive consistency and ensure that responsible AI is part of the everyday way in which our engineering teams work

microsoft ai innovate

Documents