So, today in our little community of Hacker Pack India, I asked a group of college students what questions they are facing as they are starting with AI. Here are the questions they asked and my best attempt at answering them:
Pratikshit Baruah asks “What is the best source to learn AI? With a limited knowledge of AI what problems should we attempt to solve using AI to hone our skills?”
In my earlier article, I’ve discussed the best sources to learning AI. The best way is to starting with Andrew Ng’s awesome Machine Learning online course. The second part is more interesting to answer. To get better at AI, you should think about the problems that AI is solving today. I would suggest pick one of the problems and try to make limited prototype of it by yourself.
Understanding statistics and probability also gives a clearer picture of the opportunities that you can tackle with AI. One of the ideas I am getting is getting the 500 Greatest Songs according to Rolling Stone , get the exact metadata about those songs with a song detail api and build a classifier that tells me if the song I’m listening to right now is in the class of 500 greatest songs or not.
If someone does build it, do tell me all about it at firstname.lastname@example.org
Monseej Purkayastha asks “What aspects of AI should we focus on to make ourselves industry-worthy?“
There are a lot of aspects of AI but solely in my opinion industry is more focused today on building classifiers and recommenders, still in supervised learning for majority of the applications. Looking to really develop your skillset and if you understand supervised learning well, you can also learn unsupervised learning, solving problems with neural networks and anything to do with images.
Solve as many problems as possible. AI and ML, like any other skill, gets better with practice and continuous experience.
Rohit Upadhyaya asks “How can cyber security be applied to AI?“
Great question. There is a lot of work being done in cyber security with AI. Most production systems in large enterprise and cloud software have thorough logging and all the traces are kept in most places for a definite period. These logs contain data about hardware, uptime, application, and cyber security breach attempts as well on all layers of the architecture. An AI system should be able to train itself from this data and predict vulnerabilities in system in whenever new software is written and deployed.
I’m not an expert, but that is how I imagine it would go down. There’s another interesting resource on this which explores this topic in depth.
Udipta Sharma Bharadwaj asks “What aspects of AI needs to be focused on to make a virtual helper or butler to carry on monotonous daily tasks? I am also interested in their physical implementation using electronics.”
This is basically the problem of a conversational AI. There has to be an interface that understands your natural language and then converts it into machine instructions to call an api, or do some calculations, or perform a task like uploading something to a cloud. When you go towards the IoT angle of these things, you can start with home automation — like an AI for turning on the lights and opening the door.
It’s interesting as there can be a lot of automation that can happen in our daily life. Personally for me, I’d want to automate the menial decision making I do during the day like choosing what shirt to wear, booking a cab to go to office, and buying groceries. Mark Zuckerberg did a lot of JARVIS work recently in his hobby project which was kind of cool. You can read the whole explanation and understand how he made his own virtual butler.
And that’s a small sneak peek on how people are approaching AI! If you are also starting up with AI, get in touch with me.
Also, please subscribe to my blog. Your comments keep me motivated to write more content like this. You can also follow me on twitter (@braveproductguy) or shoot me an email at email@example.com