Most important thing - since you have a community which seems to be interested in AI, I would strong recommend that you organise paper reading sessions along the lines of Papers We Love. The domain of AI is evolving very rapidly and it gets overwhelming to keep up with the new research. Doing such paper reading and knowledge sharing sessions would keep you and the community updated about the field.

Now lets come to the more general question.

Before jumping into AI, get some hands-on experience with more traditional machine learning. For someone totally new to ML, Andrew NG's Machine Learning Course is a very good starting point. But make sure you complete all the homework exercises and not just watch the videos. The course would give you a good flavour of all the broad aspects of ML and prepares you for the next step.

For next step, I would suggest you go through the Deep Learning Book to get an idea of what broad areas are most relevant for understanding deep learning. Depending on your requirements you may or may not need to take an online course in say Linear Algebra. You should also check out the Convolutional Neural Networks for Visual Recognition Course from Stanford. Even though the course focuses on computer vision, it does introduce you to a lot of aspects of deep learning.

The next 2 paragraphs are taken from my previous answer in this session and I am adding them here for the sake of completeness.

Build the habit of reading - You need not read all the papers end to end. For starting, you may simply read the blog accompanying the paper. (check out Deepmin Blog, OpenAI Blog etc) or try out the demos and so on. The idea is that you start connecting with this research and understand what is possible with deep learning. Once you understand "what can be done" part, "how it can be done" part will become more appealing and natural to understand.

Do a lot of coding - Don't just read and watch videos. Implement the algorithm that you read about. Then read more papers/algorithms and implement them as well. Once you start implementing the algorithms, you would realise that there are various aspects of the algorithm that you did not fully understand up till this point even though it made perfect sense when you read it. Once this happens, you actually start to appreciate and understand the intricacies of deep learning.

Coming to the part about how to switch from application engineer to AI engineer - I believe it is easier to move to a team that does ML (within your current company) versus finding a new ML-based job in a new company. Also, do not expect to instantly make a complete switch over from one role to another, it would rather be a gradual change. In your new role, you would be doing more of applications part and less of AI but continue to learn and try to get more and more involved in the AI part of the business. Also, Kaggle (or other data science contests) could help you to showcase your talent and find a full-time AI engineer job.

I have tried to answer the question in a very general sense. If you have any specific questions, please reach out to me.