Artificial Intelligence (AI) as a whole can often be perceived as a complex, dangerous, or inaccessible tool for the general public. How can we present this tool to a broader audience? How far can AI go in our world in 2024? How can we simply use it in the Microsoft Azure cloud? Below is a simplified and quick presentation of this tool, which is a source of excitement but also of fears and mistrust.
Firstly, when we hear about “AI,” we might immediately associate this technology with an intelligence that could create its own algorithms and adapt to its surroundings… or worse, turn against its own creator! In reality, thankfully, this is not the case! It’s only in movies that you see such scenarios… In truth, AI and its general functioning are quite simple to understand. AI is primarily defined as software that exhibits human-like capabilities. This means it can perform tasks that a human would normally be able to do. However, these capabilities are limited to well-defined tasks and can only be initiated by a human. AI will never perform these tasks unless instructed to do so. It is like a button you press, essentially a highly advanced calculator (to summarise very quickly). In other words, AI is a program that does what it is asked… nothing more. The key is to understand what we can ask it to do, whether it is accessible to everyone, and if there are any dangers, how we can define and avoid them!
We will develop this demonstration by giving a brief overview of AI, its functionalities in Azure, and its ethical guidelines. Then, we will show that any IT engineer can integrate this solution without necessarily being a coding enthusiast!
Overview of Different Artificial Intelligences
1. AI Accessible to the General Public
The general public today knows several AI tools such as ChatGPT and Microsoft Copilot, which are the most well-known. However, many other AI tools are also accessible to the general public.
2. Key AI-Related Terms
Several terms are used by the general public to talk about AI. Here’s a quick overview to help differentiate them:
- Data Science: “Data science” involves collecting and analysing data to discover trends and relationships. For example, a data scientist might study data on endangered animal populations and correlate it with information on industrialization and the economy of a region. By using statistical techniques, they can understand the impact of human activities on wildlife and propose strategies to balance economic development with the conservation of endangered species.
- Machine Learning: Machine Learning is a branch of data science focused on creating predictive models. For instance, a data scientist might use data on nesting sites, protected areas, human population, and road traffic to predict the growth or decline of a species. This model can help evaluate the impact of new development projects on local wildlife.
- Artificial Intelligence: Artificial Intelligence (AI) often uses machine learning to simulate aspects of human intelligence. For example, to monitor endangered species without disturbing their habitat, a predictive model can analyse photos taken by motion-sensor cameras to identify animals. This allows tracking animal populations over large areas and determining which areas should be protected.
The Different Services
As previously mentioned, AI is defined as a tool performing operations that a human could perform. But what operations is it capable of? Here are the four main pillars of operations that can be performed in Azure:
- Visual Perception: AI can “see” and understand images and videos. It uses advanced technologies to analyse and interpret what it sees, such as recognizing objects or faces in photos.
- Text and Conversation Analysis: AI can read and understand written texts. It is capable of responding realistically and understanding the meaning of words and sentences. This allows it to hold written conversations with people, similar to messaging.
- Speech Recognition and Synthesis: AI can understand spoken messages and respond by speaking. Combining this with its ability to understand text, it can converse with users like a person. For example, you can ask a question to a virtual assistant and receive a spoken response.Decision Making: AI can use its experience and what it has learned to evaluate situations and make decisions. For instance, it can detect issues in machines by analysing data and act to prevent breakdowns or damage.
1. Does Artificial Intelligence Have Morality?
To answer the question quickly… Yes! Artificial intelligence is surrounded by moral and ethical rules that an engineer must take into account:
- Fairness: AI systems must be fair. For example, a machine learning model for a bank loan application must predict approval without biases related to gender or ethnicity.
- Reliability and Safety: AI systems must be reliable and secure. For example, an AI system for an autonomous vehicle or a machine learning model for diagnosing patients must be reliable to avoid risks to human life. Like all software, AI-based applications must undergo rigorous testing and management processes to ensure their functionality before production deployment.
- Inclusivity: AI systems must involve and benefit everyone. AI should bring advantages to society as a whole without discriminating against physical abilities, gender, sexual orientation, ethnicity, or other factors.
- Transparency: AI systems must be understandable. Users should know the purpose of the system, how it works, and its limitations. For example, for an AI system based on a machine learning model, inform users of the factors influencing prediction accuracy, such as the number of training cases or the most influential features. Also, share the confidence index of the predictions. For applications using personal data, such as facial recognition, specify how the data is used, stored, and who has access to it.
- Accountability: The responsibility for AI systems must be attributable to individuals. Although AI systems may seem to operate autonomously, the developers who trained and validated the models, and defined the logic of decisions based on predictions, are responsible for them. To ensure this, designers and developers must work within a governance framework and organisational principles to guarantee that the solution complies with defined ethical and legal standards.
2. Is a “Private” AI Possible?
Today, cloud hosting providers are increasingly participating in an AI race that seems unstoppable. At Microsoft, for example, it is now possible to implement artificial intelligence in your infrastructure and make it work with your software as an additional component, bringing efficiency, analysis, and thus a considerable improvement in your capabilities.
3. How?
Microsoft Azure indeed offers several different services that allow implementing a solution that can meet simple or more complex needs when the service is implemented within an existing application. They can be installed manually or automatically and integrated with each other. Azure offers a service encompassing all AI functionalities called Azure AI Services.
Other services exist and offer more specific functionalities. Here are the most well-known:
- Azure AI Search: Azure AI Search (formerly Azure Cognitive Search) offers a secure solution to retrieve information on a large scale in traditional and generative search applications. It is commonly used for catalogue or document search, data exploration, and conversational applications based on proprietary data. One of its key features is the internal search engine for companies. When integrated with Azure OpenAI, this service can provide search results as constructed sentences.
- Azure AI Vision: Azure AI Vision is a Microsoft service that allows applications to analyse images and videos to extract useful information. It uses advanced technologies such as optical recognition to read text in images and AI to detect objects, faces, and even analyse people’s movements in real-time.
- Speech: The Speech service provides speech recognition and synthesis capabilities with a Speech resource. You can transcribe speech to text with high accuracy, produce speech synthesis voices with natural tone, translate spoken content, and use speaker recognition during conversations.
- Azure OpenAI: Azure OpenAI Service provides access via the REST API to OpenAI’s powerful language models, including GPT-4, GPT-4 Turbo with Vision, GPT-3.5-Turbo, and embedding models. The new GPT-4 and GPT-3.5-Turbo models are now generally available. These models can be easily adapted to specific tasks such as content generation, summarization, and more.
- Azure Language Service: This service is a suite of tools that allows developers to integrate natural language understanding capabilities into their applications. It offers features such as sentiment analysis, entity extraction, automatic translation, and speech recognition.
Many other AI services are of course available on Azure. The services presented represent only a small part of what AI has to offer. It is possible to work with these services without being a developer or an expert in this solution. The reverse is also possible when you want to customise your tool from A to Z.
What interests us here is the ability to implement this service without being an expert in the field. In a future article, we will see how to implement OpenAI in a basic and secure way in your infrastructure without having to know any programming language compatible with Microsoft certifications to become an AI engineer.
4. Need Support?
Devoteam can support your AI projects to define use cases and implement them with the help of our AI, Cloud, and Data experts. In some cases, we can offer funding programs to carry out POCs.