Artificial Intelligence (AI) is the ability of a computer program or machine to mimic human-like behavior. For example, to mimic visual senses, speech recognition, decision-making, natural language understanding, and so on. It’s not a technology of itself, but rather a goal set by technologists to imitate human intelligence.

Generative AI is a subset of AI. AI can be used to predict outcomes, detect entities, or classify documents, among others. However, generative AI, also known as GenAI, creates content, such as images, videos, code, or text. The goal is that this AI-generated content should be as useful as any created by humans. This approach is made possible by language models, which are complex AI models that can be used for a broad range of use cases.

For example, you may use generative AI to develop follow-up questions to a meeting, create an image from text, or explain the punch line of a joke, even if the joke is in a video.

Foundations of AI

Modern AI is built on a foundation of data science and machine learning. The primary goal of AI is to use machines for capabilities that are usually associated with humans. Let’s see data science concepts support the foundation of AI.

What is data science?

Data science is an interdisciplinary field whose aim is to achieve AI. It primarily uses machine learning and statistics techniques. In most cases, data scientists are the experts in charge of solving AI problems.

What is machine learning?

Machine learning is a technique where a machine sifts through numerous amounts of data to find patterns. This technique is frequently used for AI purposes. Machine learning uses algorithms that train a machine to learn patterns based on differentiating features about the data. The more training data, the more accurate the predictions.

Here are some examples:

• Email spam detection – Machine learning could look for patterns where email has words like “free” or “guarantee”, the email address domain is on a blocked list, or a link displayed in text doesn’t match the URL behind it.
• Credit card fraud detection – Machine learning could look for patterns like the spending in a zip code the owner doesn’t usually visit, buying an expensive item, or a sudden shopping spree.

What is deep learning?

Deep learning is a subset of machine learning. Deep learning is imitating how a human brain processes information, as a connected artificial neural network. Unlike machine learning, deep learning can discover complex patterns and differentiating features about the data on its own. It normally works with unstructured data like images, text, and audio. It requires enormous amounts of data for better analysis and massive computing power for speed.

For instance, deep learning can be used to detect cancerous cells in medical images. Deep learning scans every pixel in the image as input to the neural nodes. The nodes analyze each pixel to filter out features that look cancerous. Each layer of nodes pushes findings of potential cancerous cells to the next layer of nodes to repeat the process and eventually aggregate all of the findings to classify the image. For example, the image might be classified as a healthy image or an image with cancerous features.

Microsoft AI approach

AI is disrupting every industry and every business. For the last decade, AI has enabled companies of all sizes to achieve better business results. There’s already a mainstream business use of AI thanks to these three trends:

• Access to massive amounts of data.
• Access to massive computing power through cloud.
• Access to AI algorithms.

However, AI is now experiencing major breakthroughs. A new generation of LLMs enables new use cases that weren’t possible a few years ago, such as those based on high-quality generative AI. Based on these technologies, organizations will experience a second wave of AI-powered transformation. However, businesses need an easy way to access the latest AI if they want to take full advantage of it.

Microsoft is working to democratize AI use. For this, it has designed a wide range of solutions and services to bring AI to everyone, irrespective of their level of AI expertise. There are four approaches, ranged from the level of AI and coding expertise required.

Microsoft Copilot

Microsoft has embedded AI in everyday applications, so business users can benefit from it, even if they don’t have coding or data science expertise. In this approach, AI is delivered as a Software as a Service (SaaS) and becomes transparent, that is, it’s fully integrated within the provided service without users having to worry about it. For example, Microsoft Copilot for Microsoft 365 incorporates the latest generative AI in the shape of a virtual assistant that performs tasks for you in Microsoft 365 apps.

Microsoft Power Platform

A suite of low-code products that help you build different pieces of applications. These products have a layer of AI, but it’s transparent as well and you can benefit from it without handling it directly.

Azure AI Services

These are the solutions for users who want to deliver an AI project but have little data science expertise. They offer pretrained AI models for you to reuse or customize.

Azure Machine Learning

All machine learning tasks can be handled from this service. It helps data science teams in setting, automating, and enabling machine learning best practices.

Use AI embedded in everyday copilot applications

To truly realize the potential of AI, it’s essential to bring AI to every employee in ways that are relevant and meaningful to their work. Microsoft makes this possible by embedding AI in the applications people use in their everyday routine. No code or data science expertise is required because AI is delivered as just another feature of a SaaS product. The result is a wide range of intelligent applications for business users.

Copilot refers to AI embedded into applications. Microsoft Copilot provides a transformed experience across business functions and everyday routines.

Some AI solutions are specialized in helping solve problems and gain insight in some specific horizontal functions and sectors. These intelligent business applications weave relevant AI capabilities into their existing workflows. For example, Microsoft Dynamics 365 helps workers from specific business lines and functions automate and improve certain tasks. Microsoft 365 does the same by addressing a more general audience.

These solutions are often delivered as SaaS AI solutions, which deliver fast and cost-effective results. With powerful intelligence in their existing workflows, business users can be more proactive and effective in their core competencies.

Business function	Example scenario
Commerce	Commerce users can use AI insights to help them more effectively manage cashflows using payment recommendations, intelligent budget proposals, and cashflow forecasting. They can even use AI to better protect their e-commerce business—and their customers—against fraud.
Customer service	Customer service users can gain insights to address increasing volumes and manage efficient agent distribution. They can also create virtual agents that identify and resolve customer issues quickly—all without having to write code.
Finance	Analysts are provided a range of AI-powered tools for real-time reporting, embedded analytics, and insights. For example, AI can predict when or whether their customers will pay their invoices.
Human Resources	Workforce data can be transformed into actionable insights and next-best-action guidance. AI can also be used to automate HR tasks for employees, making procedures more agile.
Marketing	AI-powered customer insights give marketing users a single view of their customers to optimize engagements and discover insights that drive personalized and meaningful experiences.
Project management	Embedded analytics can provide insights based on project sales and financial data. The solution proposes an AI-powered scheduling to anticipate needs. Operations users gain insights into how their customers use their products and services.
Sales	Sellers can sell smarter with embedded AI-powered insights fueled by customer data.
Supply Chain	Business users can use AI for predictive maintenance in factories. AI is also helpful to optimize inventory.

With business applications that use AI as a core ingredient, users can bring together relationships, processes, and data across applications to gain increased visibility and control.

Everyday AI

There are also numerous AI capabilities that are already included in the applications everyone uses in their everyday routine, since they’re integrated into almost every job and function. Anyone can use them to address the realities of the modern workplace like virtual communication and the overwhelming amount of information.

For years, Microsoft has been putting AI to work in the Microsoft 365 apps that people use every day—like Microsoft Teams, Outlook, and Office. With these intelligent productivity experiences, employees can collaborate and conduct meetings more effectively, focus their time on value-added work, and uncover timely insights to improve their work.

Microsoft 365 Copilot adds another layer of AI. Business users can ask this virtual assistant to perform certain tasks just by using natural language. The assistant uses the latest generative AI technology to understand the request and do what is asked.

These solutions can improve your routine by boosting your remote work, your focus, your productivity, and your search power.

Everyday AI for remote work

Virtual meetings are becoming increasingly critical in most of our lives. While there’s no true replacement for in-person collaboration, there are new AI tools that can decrease pain points, increase human connection, and make virtual work more engaging.

For example, intelligent experiences in Microsoft Teams like background blur and custom backgrounds can help meeting participants minimize the chances of disturbances appearing on their screen. Live captions help improve accessibility for meeting participants who are hard of hearing or have hearing loss, non-native English speakers, or people with a sleeping baby nearby. Business users can even leverage real-time noise suppression to reduce distractions such as loud typing or a barking dog.

When you’re not speaking in person, some nuances are missing and misunderstandings can occur. Copilot can help business users find the right tone for their emails in Outlook to help address such issues.

Everyday AI for focus

Nowadays, workers’ routines are too often interrupted by distractions, calls, and multitasking. AI can also help cope with this problem and enable employees to focus their time and attention on what matters most.

For instance, Microsoft 365 Copilot includes features for focus to make sure users don’t forget any important issues. In OneNote, for example, it identifies unanswered questions all across existing notes and grouping them in one centralized location. In Teams, Copilot can extract action points from the conversation in real time.

Everyday AI for productivity

Breakthroughs in AI technologies have also enabled the transformation of personal productivity in apps people use every day, apps like PowerPoint, Word, and Excel.

To help prepare more engaging presentations, users can take advantage of intelligent suggestions for slide designs. Copilot also incorporates generative AI to create custom images for their slides. Users can rehearse the presentation and receive real-time feedback to improve pacing and limit filler words or culturally insensitive phrases.

Writers can take advantage of intelligent suggestions to not only correct spelling and grammar but also rephrase entire sentences for more effect or clarity. Copilot goes further and can write summaries, brainstorm ideas, organize ideas into key themes, or fully rewrite content.

Everyday AI for search

Harnessing information has become the key to almost everything—from improving productivity to understanding customers and much more. However, data is often siloed and hard to find.

AI-powered search experiences like Microsoft Search can help business users wade through this data to uncover more effective insights and make better data-driven decisions. Microsoft Search enables users to search for people, files, sites, and more across their organizational data and public web data—all from within the Microsoft 365 products they’re already working in. Results are even personalized to each user to ensure relevance. This feature is improved with Copilot.

Use Microsoft Power Platform to bring AI to your business

AI embedded in everyday applications may not be enough to power the business applications an organization needs. In these cases, Power Platform is the next step towards more customizable AI solutions. It provides a simple, low-code way to introduce AI in your business applications without having to create or manage the AI yourself.

What is Microsoft Power Platform?

Microsoft Power Platform provides low-code and no-code services designed to simplify the process of building technical solutions. It provides building blocks that help teams work faster. Even if Power Platform isn’t centered on AI, its services are often powered by AI and help you create smart solutions.

The Power Platform portfolio includes five different products: Power BI, Power Apps, Power Automate, Copilot Studio, and Power Pages. It also offers three additional tools: AI Builder, Microsoft Dataverse, and Connectors. Let’s see what each of them can do for you.

What can you do with Microsoft Power Platform?

All of the products contained in Power Platform are used to speed up business app development. Beside the specific AI functionalities included in them, they can be connected to Copilot. Thanks to this feature, users can leverage the Copilot generative AI to automatically create the report, workflow, app, website, or chatbot just by describing what they need.

Product	Description
Power BI	Power BI is a business analytics service. It provides insights on a customizable dashboard. It helps organizations be more data-driven and take better decisions based on data. This data-driven approach aligns with one of the core principles of AI, which emphasizes using data to gain valuable insights and make better choices.
Power Apps	Power Apps is a low-code development environment that enables businesses to easily create custom apps without extensive coding knowledge. With the inclusion of AI Builder, developers can seamlessly integrate prebuilt or custom AI models, optimizing business processes and enhancing the intelligence of their applications.
Power Automate	Power Automate is a powerful tool that allows businesses to automate repetitive tasks and streamline workflows without the need for extensive programming. With the integration of AI Builder, users can effortlessly incorporate prebuilt or custom AI models, enabling intelligent decision-making and driving efficiency in business processes.
Copilot Studio	Copilot Studio is a tool for building chatbots. It’s built over many AI models, mostly those enabling natural language understanding (NLU), so the bot can understand what is being said. However, its AI can also detect pieces of the bot that can be improved, and even automatically implement the improvements.
Power Pages	Power Pages is a low-code software-as-a-service (SaaS) platform for creating, hosting, and managing websites. Power Pages simplifies the website development process, making it accessible even to users with limited technical expertise.
Data connectors	Data connectors establish seamless connections between various components (apps, data, devices) and the cloud. These connectors ensure smooth integration and communication, creating a cohesive experience across the platform.
AI Builder	AI Builder empowers developers to incorporate AI capabilities into their applications and workflows without requiring data science expertise. With prebuilt and customizable AI models, AI Builder enhances Power Apps and Power Automate by enabling functionalities like sentiment analysis, category classification, entity detection, key phrase identification, and language analysis.
Dataverse	Dataverse acts as the storage solution in the Power Platform, enabling seamless integration with all its products. It serves as a central repository for data, allowing for efficient organization and accessibility.

In summary, Power Platform is a suite of powerful tools designed to help businesses create apps, analyze data, automate tasks, build chatbots, and manage websites. With Power BI, you can get valuable insights from your data and make better decisions. Power Apps lets you easily build custom apps without coding, and AI Builder adds intelligent features like language analysis and sentiment analysis. Power Automate helps you automate repetitive tasks and save time, and Copilot Studio allows you to create chatbots that understand and respond to users. Plus, Data connectors ensure smooth integration between different components, and Dataverse provides a central place to store and access your data. By using these tools together, you can enhance productivity and make your business more efficient.

What is the business value of Microsoft Power Platform?

There are two main ways in which Power Platform creates business value for organizations:

• Reducing development costs: It provides the building blocks for teams to create custom solutions in much less time than required when starting by scratch. Teams can build custom apps in just a matter of days or weeks.
• Enabling more agile, scalable development: The low-code philosophy is central to Power Platform. It allows for faster, more agile solution development. It empowers citizen developers, that is, employees with less coding expertise, to provide working solutions to end users. Professional developers can iterate on this version for further improvement. This collaborative development approach implies solutions are available to end users at an earlier stage and are less costly. This structure is easy to escalate by adding custom functionality.

Develop AI solutions with Azure AI Services

When considering adopting AI into your business, you should consider prebuilt AI services first. Azure AI Services is a Microsoft product that delivers AI as SaaS. It includes pretrained models developed by Microsoft global researchers and data scientists to solve common problems. To avoid reinventing the wheel, businesses can leverage prebuilt services to achieve quality and accelerate delivery of technology solutions.

It’s better to use the Azure AI Services that offer prebuilt AI services in vision, speech, language, search, or generative AI to solve common scenarios. This brings AI within reach of every developer and organization without requiring machine learning expertise. As a result, it enables developers of all skill levels to easily add intelligence to new or existing business applications.

Using Azure AI Services can:

• Save costs: Since AI Services is serverless, they’re usually less costly than developing and training custom models from scratch internally.
• Give deployment flexibility: You can export AI Services models and run them wherever you need, in the cloud, on-premises, or on the edge.
• Provide enterprise-level security: AI services provide a layered security model, including authentication with Microsoft Entra credentials, a valid resource key, and Azure Virtual Networks.
• Connect to an ecosystem of products: AI services are part of a broad ecosystem that includes automation and integration tools, deployment options, Docker containers for secure access, and tools for big data scenarios.

Azure AI Services capabilities

Azure AI capabilities include: vision, language, speech, document intelligence, search, and generative AI. You can build solutions with these capabilities using a suite of Azure AI services, including:

• Azure AI Vision: includes models that analyze images and videos. Beside more generic models, there are specific ones for extracting text from images (optical character recognition or OCR), for recognizing human faces. Another option is Azure Custom Vision, which lets users build their own AI models to recognize objects or classify images. Keep in mind that face recognition services are highly restricted due to Microsoft responsible AI policies.
• Azure AI Language: focuses on processing and analyzing text. They’re trained to understand natural language and extract insights. For example, models recognize language, intent, entities, and sentiment in a text. Besides, they can find answers to the questions put to them.
• Azure AI Speech: provide models that deal with oral conversation. They can transform speech to text and vice-versa. It’s also possible to translate what the speaker says and identifying each speaker. Models can even suggest pronunciation corrections to the speakers.
• Azure AI Document Intelligence: incorporates OCR and text analytics models to extract data from invoices, receipts, and other documents. Document intelligence relies on machine learning models that are trained to recognize data in text.
• Azure AI Search: provides secure information retrieval at scale over user-owned content in traditional and generative AI search applications. Azure AI Search can index unstructured, typed, image-based, or hand-written media. The indexes can be used for internal only use, or to enable searchable content on public-facing internet assets.
• Azure OpenAI Service: enables users to leverage generative AI models via Azure AI Services. In other words, it allows you to access OpenAI models directly from Azure, instead of the public API. Keep in mind that Azure OpenAI Service isn’t the only Microsoft product delivering this kind of models to users. In previous units, we’ve already discussed generative AI included in Microsoft Copilot for Microsoft 365 and Copilot in Power Platform. These copilot features are powered by GPT, an OpenAI model for text generation.
• Azure AI Foundry: a Microsoft cloud platform that brings together multiple Azure AI-related services into a single, unified development environment. Developers can use these services to build end-to-end AI solutions. Specifically, Azure AI Foundry combines:
  • The model catalog and prompt flow development capabilities of Azure Machine Learning service.
  • The generative AI model deployment, testing, and custom data integration capabilities of Azure OpenAI service.
  • Integration with Azure AI Services for speech, vision, language, document intelligence, and content safety.

Create custom AI models with Azure Machine Learning

The availability of sophisticated AI models can help organizations reduce significantly the intimidating amount of resources a data science project can require. Let’s see how organizations can tackle machine learning challenges and operations with Azure Machine Learning.

Machine learning challenges and need of machine learning operations

Maintaining AI solutions typically requires machine learning lifecycle management to document and manage data, code, model environments, and the machine learning models themselves. You need to establish processes for developing, packaging, and deploying models, as well as monitoring their performance and occasionally retraining them. And most organizations are managing multiple models in production at the same time, adding to the complexity.

To cope effectively with this complexity, some best practices are required. They focus on cross-team collaboration, automating and standardizing processes, and ensuring models can be easily audited, explained, and reused. To get this done, data science teams rely on the machine learning operations approach. This methodology is inspired by DevOps (development and operations), the industry standard for managing operations for an application development cycle, since the struggles of developers and data scientists are similar.

Azure Machine Learning

Data scientists can manage and execute machine learning DevOps from Azure Machine Learning, a platform by Microsoft to make machine learning lifecycle management and operations practices easier. Such tools help teams collaborate in a shared, auditable, and safe environment where many processes can be optimized via automation.

Machine learning lifecycle management

Azure Machine Learning supports end-to-end machine learning lifecycle management of pretrained and custom models. The typical lifecycle includes the following steps: data preparation, model training, model packaging, model validation, model deployment, model monitoring and retraining.

The classic approach covers all the usual steps of a data science project.

1. Prepare dataset. AI starts at data. First, data scientists need to prepare data with which to train the model. Data preparation is often the biggest time commitment in the lifecycle. This task involves finding or building your own dataset and cleaning it so it’s easily readable by machines. You want to make sure the data is a representative sample, that your variables are pertinent for your goal, and so on.
2. Train and test. Next, data scientists apply algorithms to the data to train a machine learning model. Then they test it with new data to see how accurate its predictions are.
3. Package. A model can’t be directly put into an app. It needs to be containerized, so it can run with all the tools and frameworks its built on.
4. Validate. At this point, the team evaluates how model performance compares to their business goals. Testing may return good enough metrics, but still the model may not work as expected when used in a real business scenario.
   • Repeat steps 1-4. It can take hundreds of training hours to find a satisfactory model. The development team may train many versions of the model by adjusting training data, tuning algorithm hyperparameters, or trying different algorithms. Ideally the model improves with each round of adjustment. Ultimately, it’s the development team’s role to determine which version of the model best fits the business use case.
5. Deploy. Finally, they deploy the model. Options for deployment include: in the cloud, on an on-premises server, and on devices like cameras, IoT gateways, or machinery.
6. Monitor and retrain. Even if a model works well at first, it needs to be continually monitored and retrained to stay relevant and accurate.

Machine learning operations

Machine learning operations (MLOps) apply the methodology of DevOps (development and operations) to manage the machine learning lifecycle more efficiently. It enables a more agile, productive collaboration in AI teams among all stakeholders. These collaborations involve data scientists, AI engineers, app developers, and other IT teams.

MLOps processes and tools help those teams collaborate and provide visibility through shared, auditable documentation. MLOps technologies enable users to save and track changes to all resources, like data, code, models, and other tools. These technologies can also create efficiencies and accelerate the lifecycle with automation, repeatable workflows, and reusable assets. All these practices make AI projects more agile and efficient.

Azure Machine Learning supports the following MLOps practices:

• Model reproducibility: means different team members can run models on the same dataset and get similar results. Reproducibility is critical for making results of models in production reliable. Azure Machine Learning supports model reproducibility with centrally manage assets like environments, code, datasets, models, and machine learning pipelines.
• Model validation: before a model is deployed, it’s critical to validate its performance metrics. You may have several metrics that are used to indicate the “best” model. Validating performance metrics in ways relevant to the business use case is critical. Azure Machine Learning supports model validation with many tools to evaluate model metrics, such as loss functions and confusion matrixes.
• Model deployment: when a model is deployed, it’s important to have data scientists and AI engineers work together to determine the best deployment option. These options include, cloud, on-premises, and edge devices (cameras, drones, machinery).
• Model retraining: models need to be monitored and periodically retrained to correct performance issues and take advantage of newer training data. Azure Machine Learning supports a systematic and iterative process to continually refine and ensure the accuracy of the model.