Building a discovery assistant

Digital assistants, Chatbots, and Artificial intelligence have sort of become buzzwords these days, haven’t they? Whether that’s a good thing or a bit of an overkill, that’s up for debate, but one thing none of us can ignore is the amount of impact AI is going to have in our lives in near future. Unlike the ‘AI will unleash doom-and-gloom’ club, I’m from the optimistic club advocating the positive impact on our lives. In this post today, I wanted to pen down my thoughts on a problem which I personally face and that I think can be solved better. Do let me know whether it resonates with you or not.

CURIOSITY to keep ourselves updated on a topic of interest and knowing more and more about it, is  a fundamental need we all share to varying degrees. The way we do it today goes something like this. Let’s say I am interested in ‘artificial intelligence’ as a topic. There are two parts to my curiosity. One being getting to know about AI right at the moment (active, pull). Second being keeping myself updated about the latest happenings in AI (passive, pushed content).

Google is obviously the de-facto place we all start our research with, and is a product fit for the ‘active curiosity’ part of the problem. Reddit is another good product already solving much of the ‘staying updated on topic’ problem. Part of it is solved by Twitter. Part by Quora and similarly part by Youtube, StackExchange, Google News, Google Alerts, etc.

For the active research part, the way we do it is google a bit, read Wikipedia, read past research, read some books and follow any links to blog posts, videos, etc that we come across while we’re on it.

For the passive part, we use much of the output of the active research part and follow relevant Quora topic that we find, Twitter handle, Reddit topics, YouTube channel or some influential person from the domain and follow them so that we remain updated on a constant basis. Sometimes some of these, at other times multiple of these depending on which of these platforms we’re more used to, and depending on the level of interest in the topic. Other than that, we can also start following a person/page which we came across while not actively looking for it, when, what we came across sounds interesting enough.

Now when I think about it, I see 10 years down the line this won’t be the way we’ll discover stuff. The users won’t need to submit stuff (to Reddit). They won’t need to go through a lot of low SNR content everywhere. They won’t create and manage lists. They won’t have to mix knowledge discovery with keeping in touch with their friends and family on world’s biggest social network. They won’t have to think through a right query to search for and then try finding a needle in the haystack with the results being non-personalized. With exponentially increasing amount of user-generated content that is being added to the internet every day, the haystack is getting bigger and bigger.

One solution would be to borrow best ideas from many of the above-mentioned products. So instead of users submitting links (to Reddit), it can very well be bots crawling (like Google does). To infer interests, it would bootstrap starting from picking up info from Facebook and Twitter profile, pages followed and posts/retweets. That gives it a fair idea about user’s interests while still allowing the power users to be in full control to manage the interests. To begin with, it won’t have to crawl everywhere like Google, but can pick up content becoming popular elsewhere, for example, on Twitter, FB, and Reddit. Instead of making it a community (again like Reddit, FB, Twitter) it could very well let the original host of the content be the place for the discussions while still picking up explicit like/dislike signals from the user to improve personalization. Based on the initial profiling and user’s activity, it would infer user’s comfort level with the topic and use that to provide right personalization instead of throwing everything at everyone.


Solving this problem right will obviously require solving multiple challenges to build something which really works well. What do you think? Are you completely satisfied with the current way of discovering stuff? Think it could be better?

Leave a comment

Your email address will not be published. Required fields are marked *