Sunday, December 15, 2019

Global AI Boot-camp 2019, Accra - Hands on experience with AI tools

Global AI Boot-camp 2019, Accra - Hands on experience with AI tools
Global AI Boot-camp 2019, Accra Ghana


The Global AI Bootcamp is a free one-day event organized by local communities all over the world that are passionate about Artificial Intelligence on the Microsoft stack. At the start of the year, I had reservations about this AI technology, there were so many unanswered questions and it looked like  I was against the technology. Could it be because I didn't understand it better and was quick to talk ill about it with very little knowledge? 10 months on, I found myself at an AI Bootcamp at NewEdge Infotech, Accra GH and I knew it was a great opportunity for me.


Building Machine Learning & AI Models

The Bootcamp kicked off with presentations from the crew which comprised of very reputable names. Mr Samuel Adranyi, Senior Software Engineer, Andela who gave an in-depth presentation on building machine learning and AI models. In building AI you have to be careful hence focus on specific solutions. During the presentation, I was asking myself three questions - What is AI? Why is AI model necessary, and how can I implement/train an AI model? I did have an idea about the first two. I couldn't hold my excitement when Mr Adranyi answered my last question in the course of his presentation. There are basically two classifications when it comes to training an AI  model, it's either by supervised learning or unsupervised learning.




Supervised Learning - this answers the question of when to use machine learning. In supervised learning two techniques are applied:

  • Regression - how much and how many data will be needed
  • Classification - the class the data set belongs to.


In Unsupervised Learning, clustering, anomaly detection and recommendation techniques are used.
Machine learning is becoming a trend and AI models have the ability to learn themselves. This is basically the summary of how the model building process works:
  1.  Prepare data
  2. Train the model
  3. Test the model 
  4. Deploy.
AI models can be implemented in a lot of scenarios but wait till we get to the hands-on experience with the tools provided by Microsoft.

Cognitive Service

The next speaker was Mr Francis Arkhurst Odoom who took us through Cognitive Services. The idea of building intelligent and smart applications have always been something I have been aiming to do. Cognitive service is a platform that was built by Microsoft so developers can workaround to make their apps to behave like humans. Imagine building an app that can detect human expressions. Cognitive services use vision, facial recognition, speech and knowledge-based techniques to achieve this. You can detect if a customer is sad or happy so you know how to offer a particular service to them. Processing text that can be read out to a user is also by means of cognitive service.



Frank demonstrating how Elaine works


The big question at the end of his presentation was how to implement this? Frank showed a demo of an app he created, Elaine which implements the tools of cognitive service. Elaine is an app the reads images. There is a lot of mathematics behind the use and implementation of cognitive services as well as other deep learning models. As a developer, you have to make use of this service since Microsoft has invested a lot in it. You start by registering with Microsoft Cognitive Service, after which you will be provided with some keys to be working with.

Why you should use Cognitive Service?

It is very flexible, easy to use and will suit any platform of your choice. With cognitive services, it is driven by less code as you just call the API and work with it. One other feature with cognitive service is sentiment analysis that monitors the feedback from users to improve the quality of service. Imagine big brands like MTN or Vodafone might want to know how their users are interacting on a social media platform. If they want to sort out negative comments and offensive posts no need to employ someone to be going through the feed. You can use sentiment analysis by connecting the users to a workflow. Anytime someone interacts on a platform, it is passed through the sentiment analysis to achieve the goal.

Chatbots


Mr Emmanuel Asimadi presenting on Chatbots & NL via Skype

The presentation resumed after a short break. Mr Emmanuel Asimadi had already introduced the concepts of chatbots right before the break. He mentioned that chatbots evolved through binary instructions, command prompt and GUI. The topic of chatbox would not stop because of the recognition it has gained over the years. The downside is when people try to build complex chatbox rather than focusing on achieving specific goals. "Chatbox should not be designed to behave like humans", he said.

Power Virtual Agents (PVAs)

The next item was PVAs Power Virtual Agents by Mr Yaw Amoateng, who started his presentation with an interesting story about he lost his bird. He definitely couldn't hide his love for birds. The session was about how to create a chatbot application that users can interact with using another service powered by Microsoft. Fun fact, you can get a comprehensive chat box running with no code. For those lazy coders out there, this is a win for us.

You create the bot using a QnA and this can also be integrated into websites, your applications developed with Power Apps (which also supports creating applications with no code), Facebook, Telegram and other social media platforms. Most of these chatbots are integrated into School's websites, business pages and a good number of them have humans replying questions fed into the chatbox. With PVAs, you can automate this by allowing the bot to answer FAQs.

How to get started with PVA

You need to plan your conversation and outline the flow of the conversation. The environment used to define your flow is called the design canvass. Keywords are declared to kick off the robot or wake the robot up. The likely response for each wake-up call will follow up. After that, you publish it or integrate into the platform of your choice.

Mr Yaw Amoateng demoed a chatbox he created after the presentation. It was not so different from what Frank did with Elaine. The only difference is it took Frank about a week of hard coding to produce a chatbox that took Yaw just some few minutes to get it running. Another feature with PVAs is that they have engines that allow them to learn themselves and find similar words that are used to trigger the bot.


Everyday AI by Charles Wartemberg 

This was the shortest presentation but then, in my opinion, was the most practical. The other speakers did a great job but at some point, it felt like rocket science. Charles who is currently the Product Research & Marketing at Microsoft described himself as a non-coder who fell in love with AI.

Two years ago I was two years into Microsoft, I was working with Xbox doing supply chain. I was ready for my next opportunity and I was having a lot of conversations internally. One thing you realize when working in a big company is that working for your next job is as hard as working for a job in the outside. The only difference is you know the people you work with. I was then taking the bus to and from everyday. I asked myself can I use technology to solve this problem? Now looking for a job and not having enough time during the day. I put my head together and built a bot framework specifically using the Power Platform. And I used a mix of Cortana dynamics, office 365 drag and lastly, cognitive service API. I used the LUIS tool. What this allowed me to do was set up a smart chatbox. I am not  a traditional coder, he added

The point is simple, all these AI tools have made it easy to work around with. For example, Amazon makes it easy for you to build your own Alexa skill for free.

How AI is used in everyday life

 Microsoft office- Like the Microsoft analytics that is part of the office 360 sweet. It gives you analysis on the people you talk to a lot and predict the ones you should catch up with. Gives you more insights on how you spend your time, the people you email all to ensure your well being. The application of AI is endless. Google Pixel Portrait mode and the use of Open AI.

An important aspect is the ethics behind AI which include data privacy. Every technology can be a tool as well as a weapon. The debate with AI would go on forever. AI going to take our jobs and all the controversies surrounding it. Maybe AI is the change we need to embrace in a positive way. AI is here to make things better if we decide to understand it. It will take some time to fully transition into this new era but the future is here.

Resources

1. MS Learn
2. Github - frameworks
3. Documentation of  PVAs
4. AI Business School for Microsoft - a website meant for business leaders of professionals to teach about AI.
5. Google AI
6. Google TensorFlow
7. Open AI

Breakout session & hands-on

Now the part that everyone was waiting for. A chance to implement the chunk of information that was grabbed from all the presentations. We broke into a team of three: Building bots with cognitive service, PVAs and its resources and Machine learning. I wasn't sure which team to join for a while.
But after doing some deep thinking, I decided to go with Yaw who was the resources manager for the PVAs.



With Yaw guiding our group, we worked extensively to build basic music chatbox. You can read the full documentation of how to create your own bot here.  I will advise anyone reading this is to never overlook or underrate AI because of its capabilities. Take opportunities and time to study it, attend nearby events when you can and you will reap the results soon enough. See you next year Global AI Bootcamp!



More