thumbnail

AI – what it is and how it works

AI is more than the sum of its parts

AI (“Artificial Intelligence”) is a broad concept referring to a computer-based system which behaves intelligently. AI powers machines which can perform tasks that require human intervention or decision-making. AI definitely sounds like something that could make your life easier. So what is AI about? 

Is AI the answer to our human biological and mental limits?  

Yes and No. AI can excel and surpass humans in various ways:

  • Faster and more powerful computation 
  • More efficient processing of data   
  • More reliable insight   
  • More accurate decisions 
  • Higher and smarter automation   
  • Increased connectivity 

Yet, there are areas where AI cannot replace human cognition: 

  • Intuition 
  • Spontaneous answers 
  • Creativity 
  • Sentient thinking 

Is AI simply a juxtaposition of 2 components (“artificial” + “intelligence”)? 

poster

Intelligent behavior is difficult to capture and define because it involves many skills from many walks of life (Mathematics, IT, Linguistics, Neuroscience, Economics, Computer Engineering). Developing intelligent behavior and teaching it to a machine is even more difficult. An AI is as good as the algorithm and data behind it. AI is nurtured and built, and the way and proficiency this is done with will impact how adept the AI is at handling human tasks. 

AI is based on algorithms which grow and adapt based on: 

  • Curated input data sets 
  • Experience gathered over time 
  • Constraints

A lot of work goes into building an AI. Fortunately, there are AI-based solutions you can resort to, so that the work required on your part is limited to providing just the data. 

AI puts the pedal to the metal

AI is the way forward, as it can help science, businesses, individuals live and experience richer lives. AI is meant to transform our lives:

ASSIST with human tasks: translation, recognition, classification, computing, automation, tracking, monitoring, expert advice

PREDICT next best actions, risks, mitigation plans, estimations 

ENFORCE COMPLIANCE with regulations by spotting errors, gaps, inconsistencies

AI-enablement is not a far-fetched idea, but a current reality. AI already makes our lives easier, more straightforward, and more manageable with less effort. 

AI comes in many shapes and sizes 

AI can be classified as either Weak AI (good at specific tasks) or Strong AI (good at high-level reasoning and judgment), the latter being harder to achieve. The majority of AI advancements today fall within the boundaries of Weak AI.

The Weak AI – Strong AI distinction should be seen as a continuum: 

Narrow AI and Machine Learning 

You may already be familiar with Siri, Alexa, Cortana, Google Search, autonomous driving and other smart assistance solutions (weather prediction, product recommendations etc). These systems usually solve problems and execute tasks really well in very specific or controlled contexts. 

General AI and Machine Intelligence 

This is still a theoretical concept. Its implementation would require interconnected AI systems that work together and simulate authentic neuronal activity in activities such as: language processing, image processing, computational functioning, or reasoning. 

Super AI and Machine Consciousness 

This is SkyNet, Sci-fi material where AI cognition is as good as human cognition. 

AI is about learning

AI becomes intelligent over time by learning, by abstracting rules, by identifying correlations a human could not. Building good AI is a mix of continuous learning, knowledge reconfiguration, model building, and deep adjustment. 

How does learning happen?

For an AI to learn, you need to have:     

  1. Algorithms and constraints (define how processing should be done)     
  2. Data sets for training (show the AI what it’s supposed to process)  
  3. Data sets for testing (check whether AI works)

The way your AI learns can vary. Here are some of the most common patterns. 

Supervised learning 

The training data is prepared, marked, labelled, so that the AI system can learn on this basis and develop strong pattern recognition capabilities. Labelling the information can be time consuming or may require many resources. 

Unsupervised learning 

The training data is not prepared, and the system searches for patterns on its own. A great deal of trial and error is involved, but in the end the AI can reveal correlations which would otherwise remain hidden from human thinking and observation. 

Reinforcement learning 

Reinforcement learning is based on trial and error, but the AI is rewarded or penalized based on its results. The AI learns based on the ratio score it abstracts. Reinforcement learning has already been applied to game playing (Chess or Poker). 

Transfer learning 

The skills learned by an AI are applied to a new but similar or related problem.

These learning methods are part and parcel of the VIP AI methodologies that usually make the headlines: Machine Learning (learn from experience), Deep Learning (self-learning machines) or Neural Networks (drawing associations). The three AI methodologies use a mixture of learning methods, which make some of their models more accurate or reliable than others.

An AI system becomes good at what it does progressively by identifying patterns, inferring meaning from one or multiple layers of information, evaluating data, making predictions, and even optimizing itself. 

This seems like a lot of work. Are there solutions which already have AI embedded?

Yes. Microsoft for instance has AI embedded all over its suite.

Are there AI models that I can already apply to my business?

Yes. Microsoft offers you already built models you can use. Your job is to give the model data and look at the results. Your challenge in this case is give the pre-trained model good data it can work with, potentially even curate the data. 

Microsoft and its AI implementations

poster

Microsoft technologies are developed with AI in mind or are enablers for custom AI applications. Here are just a couple examples, by no means an exhaustive list. 

Microsoft 365 for any digital office

Microsoft 365 is a cloud platform with pervasive AI capabilities meant to streamline the processes of a digital office. Here are a few areas powered by AI: 

  • MyAnalytics

Employees do not work in the same way, but they all want to improve the way they work. MyAnalytics analyzes the work patterns of each employee and provides them with a report on the tasks performed, the amount of time spent on tasks, the time spent on collaboration, and a plan to automatically have time allotted to focus or follow-up. Each employee can make informed decisions about how to adapt and improve themselves. 

  • PowerBI

PowerBI is a smart assistant program that provides data analytics and insights about your business progress and outcomes. When you are asked to provide a report, you no longer have to access several data bases, open parallel Excel files, extract and compile the data manually, and find insight if any. With PowerBI, you can provide a report faster and more accurately while also having rich dashboards you can twist and bend to your will. 

  • Cortana

Use Cortana to engage with Microsoft 365 via voice. Talk to Cortana for various requests: access/read your emails, send messages, check agenda, join meetings, or add tasks. 

  • Scheduler

Scheduler is an AI-driven system which integrates with Cortana and helps you delegate several tasks easily: request scheduling, email dispatch to negotiate future meetings when several stakeholders are involved, find available rooms, and send invites. 

Dynamics 365 for any digital business 

Microsoft Dynamics 365 is a set of cloud business applications meant to facilitate a full-circle digital transformation for any business (retail, transportation, healthcare, industrial manufacturing, banking and finances, commerce and marketplace).  

The key element behind the success of any Dynamics 365 adopter is how your business infrastructure (industry warehouses, healthcare wards, retail supply chain, transportation routing) gets connected to: 

  • Your customer data   
  • Your business processes 

With Dynamics 365 the insights you get are based on living data processed by built-in AI.

Dynamics 365 helps you build:  

  • Personalization (engaged buyer experience, employee-company interaction, product-user information, empowered employees)   
  • Cross-channel touchpoints 
  • Automated workflows 
  • Remote issue detection and prevention   
  • Enriched performance   
  • Linked data for your asset inventory and its usage (sales, finances, shipping) 

Microsoft Azure 

Microsoft Azure includes a powerful AI platform meant for developers who need to expand or build AI-driven capabilities for businesses ranging from: 

  • Image recognition, face detection, computer vision 
  • Speech recognition, automated transcription 
  • Sentiment analysis, intent recognition 
  • Risk assessment 
  • Fraud detection 
  • Behavior-based recommendations  

There are several options to choose from: 

  • Adopt an open-source AI framework: PyTorch, TensorFlow, scikit-leanr 
  • Use the pre-trained Cognitive Services and customize them with your data   
  • Deploy Machine Learning models with: Azure Machine Learning, Azure Databircks. ONNX Runtime. 
  • Implement Cognitive Search and unlock hidden patterns from across your content (files, media, images) 
  • Use the template-based Bot service and train it with your data 

Do I need AI? 

AI is a domain you can benefit from today: 

  • Develop your own AI-based system if you have specialized personnel and you need something which is highly customized to your field. 
  • Train your own AI-based system if there are pretrained models on the market which fit your needs.  
  • Use software which already has built-in AI if the solution is a good match for your business goals.