Designer should read research papers?

Let’s revive the meme for a minute, shall we?

As a designer, I love constrains. I see them as a healthy challenge and a great way to make better decisions since we’re not dealing with the infinity of possibilities.

As an AI designer, contains are highly technical, and it can get tricky to design something we don’t get the ins and outs.

And there are two roadblocks you should cross if you’re in the same situation as me. First, AI technologies can get outdated pretty fast despite built on +10 years principles/frameworks. Then, non-specialized media are just full of clickbait trash.

Please don’t embarrass yourself by sharing a Big Think Facebook post with your R&D team.

The solution? Designers should read more research papers, less carrousel Instagram posts.

Find papers

My friend, you have to go to the source.

For AI scientists, it’s called arXiv (Pro tip: pronounce it Archive.)

It’s one of the largest open-source databases of scientific papers. Counting +1,6M documents since the late 1990s.

The website will remind you of pre-redesign Reddit. Keep the site into a browser tab for extra credibility points at your company.

It gathers a lot of scientific discipline, but we will restrain ourselves to the AI part under the Computer science category. Nuclear Experiment sounds fun tho.

Designers, you will also see that there’s an old friend in the tree structure: Human-Computer Interaction. Don’t hesitate to get a sneak peek.

Once you’re into the Artificial Intelligence category, you’ll find all the papers sorted by date. I recommend to filter papers that are not on AI you use; I mainly search for Natural Language Processing advancements.

How to read a paper

You don’t read a paper like you will read the letter your ex left on the counter. There’s a method.

There’s a paper on how to read a paper called « How to Read a Paper » by S. Keshav. Isn’t it so meta?

I’ve learned and experienced that reading a paper is time-consuming, and despite the scientific method and community controlling the quality, not all are equal.

Srinivasan Keshav has a three-pass approach, macro to micro, to optimize his time.

He will seek a birds-eye view before deep-diving into the detail unless he feels it’s not worth it.

If, like me, it’s not your field of expertise, the first pass will be enough since we don’t have the background to challenge the research and being able to reproduce it.

The first pass is all about the central concept, and nothing more. You should be able to speak about it after the first pass to your R&D team at a coffee break.

• Start with the title, abstract and author’s introduction
• Then readjust the titles and sub-titles learn the structure and have hints at the through process/method
• Finish with the conclusion

That’s it. You can’t also read the references as a bonus.

You should have enough data to challenge what you’ve just read and being able to « answer the five C’s,» as S. Keshav said.

1. Category: What type of paper is this? A measurement paper? An analysis of an existing system? A description of a research prototype?

2. Context: Which other papers is it related to? Which theoretical bases were used to analyze the problem?

3. Correctness: Do the assumptions appear to be valid?

4. Contributions: What are the paper’s main contributions?

5. Clarity: Is the paper well written?

Don’t invest more time if the subject doesn’t interest you, or you haven’t the background to pass the other two steps. Reading the content and trying to re-implement it.

What’s next?

• Read the MIT Tech Review article so you won’t have to read +16k papers since 1993.
• Check often new publications; you will be more connected to the AI state-of-the-art.
• Share what you’ve learned with your team or other teams. Organize book club on Papers. To give you a critical eye when crossing the research with an experienced researcher, you can have in-house.
• Join meme groups on deep learning and machine learning and that great feeling when understanding it. You’re in the tribe now.