An Approach to learn more about AI Technology

Artificial intelligence (AI) is one of the most talked-about topics in information technology today, but frequently it is a concept that generates fear and creates a perception that it is a field only to be travelled by the elite. This article is to provide some of my views that I felt helped me on my continued journey towards learning more about AI.

I am sure there are many today that has the same question; “How do I start learning more about AI?”. Like most people, I started by searching the term ‘Artificial Intelligence and the search results very quickly became very hard to follow. The complexity started when I read about all the levels and types of AI; weak AI, general/narrow AI, super AI, supervised, unsupervised and reinforcement. Then, to try and make sense of the AI types/levels I learned that AI is developed through various machine learning (ML) techniques, which includes statistical models, traditional ML, deep learning, and artificial neural networks etc. I was trying to get a grip on the types, terms and techniques in AI, but I soon realised that I need to change my approach; instead of trying to understand the nuts and bolts to develop AI, I rather attempted figuring out where this new technology should actually be used.

My second attempt lead me to learn more about the patterns in AI, i.e. the fields/areas AI can be divided into and how it contributes to existing software developments. Focusing on these patterns made me understand the potential rather than starting from the technical theory. The patterns from various sources included the following fields;

1.     Autonomous

• Machines performing tasks without any human interaction.
• Examples include; self-driving cars, business process automation, audio file transcriptions, knowledge mining from documents.

2.     Computer Vision

• Machines with the ability to ‘see/read’ like humans
• Examples include; Facial recognition from videos or images, emotion detection, text extraction from documents.

3.     Conversational AI

• Virtual agents – creating a digital agent that can engage with humans through text or speech.
• Examples include; Natural language processing (natural language understanding, natural language generation), text analytics, text to speech

4.     Decision support (Predictions, patterns and anomalies)

• Augmenting human decision making with machine derived intelligence.
• Examples include; anomaly detection, fraud detection, time-series forecasting, product recommendations.

5.     Ambient AI:

• Bringing AI to our everyday environments - Virtual and Augmented reality – overlaying existing reality with digital components.
• Examples include; Adding digital metrics to a factory floor (which are visible through a HoloLens or other AR devices/apps), indicating any possible failure points or alerts that require attention.

Understanding these patterns made it easier for me to start exploring AI in more detail. Each of the patterns could include types of AI mentioned earlier (supervised, unsupervised, reinforcement learning) and could use any technique to build a solution (statistical, traditional ML, deep learning or Neural Networks). Through this I created an approach towards how I engage AI learning and projects; I would first identify the AI pattern rather than jump into types/techniques, and this makes it easier to then explore more complex techniques that might be available for your specific requirements. Focusing on all the patterns and AI techniques/technology available helped me grow a strong foundational understanding and an ability to match AI technology to a specific requirement.

The need to develop and solve a business requirement leads into the next learning phase to gain more practical experience, and in my journey, I found two possible approaches to gain practical experience; each approach requires different skills:

1. Build a machine learning model using open source frameworks. Research ML methodologies and learn how to build AI/ML solutions through frameworks that have been made available by leading open source organisations.
2. Build using managed AI services offered by leading Big Tech platforms; Amazon, Microsoft and Google offer various AI services that are available for developers and AI explorers to test out without requiring deep technical/statistical knowledge.

I have found that connecting to cloud-managed AI services provides a great interface to see what AI could do with your data; below are links to test out Microsoft and AWS managed AI services:

1. Azure AI Demo’s: https://aidemos.microsoft.com/
2. AWS AI Service Demo’s: https://ai-service-demos.go-aws.com/

Seeing AI in action on a small data sample helps understand how it could be implemented.

I hope this small write up helps you to structure a learning approach towards your own AI journey. In summary, my advice would be to first create a strong foundation about AI technology and where it could be used; reading more about AI patterns and case studies will assist with this. Once you are comfortable with how and where AI could be used, you can start exploring the technology available in the market. My view is that leading big tech platforms, like Azure and AWS, packages AI technology in a way that simplifies initial development and exploration. The last phase would be to start understanding the types of AI development approaches available to create custom ML models (Supervised, Unsupervised, or Reinforcement learning).

Below are some valuable links that helped me along the way:

1. AI The Next Digital Frontier: by McKinsey & Company
2. What is Machine Learning by Karen Hao
3. Emerging AI Patterns by Steve Guggenheimer
4. AI Journey by Steve Guggenheimer

Write-ups coming next:

• Prebuilt vs Custom AI.
• Catalysts to accelerate AI awareness.
• ML Frameworks
• AI Maturity Assessments