When my mentor assigned me to research as my first week's task, I heaved a sigh of relief thinking to myself "Hey that should be fairly easy" since I spend a fairly large amount of my time surfing the web, reading people's blogs, and stumbling over interesting things people are doing these days. However, unlike traditional academic research which I had some previous experience from my college days, for this task, I was supposed to research possible industries and find a problem specific to that industry that can be potentially solved using Machine learning.
Here are a few reasons why I found this task unusually hard:
Theoretical research vs practical applications
So how do people research stuff? Usually when you are asked to write a paper for your college/Ph.D. thesis your supervisor or you would have some topic or area of research at the back of your mind that you would use to navigate your entire work. Also, you can approach academic research in various ways e.g surveying the past and current state of a particular invention, improving the performance of a previous experiment, or inventing a novel way to solve a problem. That's how we are normally taught to do research and it works fine when you're in college or getting yourself published. However, as it seemed to me from over a year of reviewing various papers for different reasons, people at school don't necessarily publish to solve everyday problems or invent yet another new gadget. A large number of research done lack industry applications.
I struggled with the task at first because I approached it in the wrong way.
I needed to find a problem to investigate whereas I was looking for different solutions that exist nowadays. After two days of reading a bunch of papers where different people applied various ML solutions in a range of different scenarios, it sure inspired me but I was getting nowhere. I realized that while it was a phenomenal way to get hyped about ML but this approach
- did not educate me about the problems that it solved
- I did not know about the scale/intensity of the problem
- I could not comprehend why the ML approach was the best one instead of other methods (if they existed)
In the quest for actual problems
To solve problems you need to bring into the picture the recipients of those problems, and you guessed it right, yes it's people like you and me. We are the ones who deal with or a lot of times also create those problems. So understanding the activities of the people who face problems is crucial here as my mentor advised at first but I lacked understanding of the necessity of this until I experienced the pain first hand. Sometimes I feel as if I deliberately choose to suffer but I digress here. I want to give you a layout of doing these kinds of research if you ever have to and reduce your suffering in the process.
Pick an industry of your liking or feel that would have the most impact by using a particular technology. We chose to investigate the HCI industry and design-related fields e.g UI/UX design as these industries seemed comparatively least impacted by the ML revolution
Once you find a niche of your choice, find actual people to talk to. It is paramount to do so because you necessarily won't have experience in that industry. You may have assumptions about some problems that those people may face every day but you would realize that the theory is far from true. You can do this in a survey style or by raw interviewing people by asking them questions e.g Tell me about how you go about doing your job, what are some typical tasks that you do every day, what is your biggest headache in your job etc. and you will notice that by getting a glimpse of their daily job the problems people face in real life are quite different than what you read in scientific papers.
On that note of talking to people, it may appear extremely hard to even find people to begin with. I found it quite hard to do so as I have relatively fewer friends and I'm not extremely social. I sought some shortcuts to do the job, I utilized Twitter, LinkedIn, and some Facebook groups to find people. I searched with relevant hashtags, on both Twitter and LinkedIn e.g #designerproblems, #hciandml, etc as people frequently post a lot about their jobs and work on these platforms. Once I found people of that particular industry, I just sent them a direct message if they would be willing to spend a few minutes answering some of my questions regarding their job. I kid you not that people on the web even though they are strangers, are generally willing to help others. I heard back from a handful of wonderful people who were kind enough to open a window to their daily life and that helped me find a lot of interesting problems to brainstorm with my mentor the following week.
To summarize, researching for problems in the real world is different than researching in the academic world and a good way to do so is to get in someone else's shoes and understand the people, activities, and tasks carried out in that industry than surfing through google scholar for research papers related to that industry (well that's how it was in my case!)
That's all folks, thanks for making it this far.👋