16 comments

[ 0.62 ms ] story [ 2614 ms ] thread
(comment deleted)
We just built Juji based on AI technologies. Like the “precog” Agatha live online, Juji can “read” you by automatically inferring your faction(s) and strengths, and your likelihood to career success and a long-lasting marriage, all from your own data (e.g., Facebook posts and tweets).

Do you believe a computer system like Juji can actually know you better than you know yourself? Do you believe it knows your faction and the faction of your friends but without asking you or your friends to go through the tests that Tris did in Divergent?

Check it out yourself to see what Juji knows about you (Juji Me) and others you want to (Juji People) at https://juji.io !

Michelle (CEO of Juji)

CTO here. Here are some more information, we did the project in Clojure, and we are very happy with this choice :-). The other parts of our tech stack include Kafka, Postgres and Aerospike, hosted on AWS. Please feel free to ask questions or give suggestions. Thanks.
You know me more than myself! How is it accurate?
If by "knowing", it means knows how one's traits compare with others, yes, we can know it better than oneself. Because that's exactly what the science of psychometrics does, and we are now applying it at a larger scale. Please look at http://juji.io/resources for a list of references this project builds on.
huahaiy, are you familiar with the work of William D. Wells - he created the psychometric Lifestyle Survey run for many years by the ad agency DDB Needham? I had the opportunity to study with Bill after he 'retired' to academia.

Anyway it was fun in looking through your site to see you referencing psychometrics, the big 5 and NLP - it was like I was back in grad school.

That's cool. I am aware of the Lifestyle Survey. We are leveraging knowledge on psychometrics, but we also try to go beyond. One problem with self-reporting psychometric surveys or tests is that the takers know that they are being assessed, and may adjust their answers accordingly, or in other words, "fake it". I know there are many ways to combat the issues in test design and statistic analysis, but we took a very different approach. We don't even ask people to take a test, but instead analyze people's existing social media data, so as to mitigate some of the problems.
I had read far enough to know that you are using what people say via facebook and twitter and email archives as input signals rather than surveys or other traditional psychometric instruments. Are all your word -> personality trait relationships taken from academic research or are you doing something independent to build out those relationships?
A lot of them come from published academic research, but some of them come from our own empirical studies.
Why is this different from Mindreader (IBM program from 1985?)
Are you talking about a word processor produced in 1985 that had a type ahead feature? This is very different. Not a word processor, but a psychometric tool.
If we are talking about the same piece of software, the MindReader that I know lets a user to describe him/herself by choosing a bunch of words. It then synthesizes a report describing your personality traits. In that sense, it does not really read your mind, instead of putting together the words you chosen to describe you. Juji on the other hand automatically derives your traits based on your word use in your own informal communication content. For example, in MindReader, you have to tell it you are "ambitious" by choosing the word, while Juji infers whether you are ambitious or not by various word use patterns even if you have never mentioned the word "ambitious" or its synonyms.
apparently I have not written more than 6000 words on facebook. I feel strangely proud
Currently we only have access to the Facebook statuses updates. So your comments and other content are not included.