Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

[Link] pplapi is a virtual database of the entire human population.

1 Post author: morganism 12 January 2017 02:33AM

Comments (9)

Comment author: iandennismiller 12 January 2017 03:39:35PM 6 points [-]

This is my research. It's an ongoing project that I will be discussing as part of my PhD dissertation, which I plan to defend in a few months. I had hoped to finish my PhD before this project got attention, but the website is already live and I can't really complain. In fact, I'm really happy with the interest and I want to engage.

Here's a few important facts about pplapi:

  • the agents are synthetic, not real people
  • the numbers are based on population parameters, so in aggregate, pplapi approaches reality
  • at the individual level, these agents exist in the "uncanny valley" of kindof-sortof-implausible demographics
  • some dimensions have more noise (error) than others; income is the easiest example
  • the research is ongoing, so there many opportunities to improve
  • one way to "use" pplapi is with an agent-based modelling environment, like NetLogo or MASON

This work is not merely a technical demonstration; it serves as a proof-of-concept "simulacrum" of humanity, which is axiomatic in my dissertation. I believe this work also raises a host of moral and ethical questions that were previously abstract but are now easier to comprehend. I do not believe pplapi itself has crossed any ethical boundaries, but I think the line of questioning is now clearer.

Comment author: morganism 12 January 2017 02:35:44AM 1 point [-]

"pplapi (pronounced “people API”) is a virtual database of the entire living human population. The current database contains 7,171,922,938 agents and is approximately 6.8 TB in size. "

"Currently, Agent Space contains a simulated entry for each of the ~7 billion humans alive in 2014, who are all addressable with pplapi."

Comment author: morganism 02 April 2017 05:53:34PM 0 points [-]

athena 6.3.3: Athena Regional Stability Simulation

http://htmlpreview.github.io/?https://github.com/AthenaModel/athena/blob/master/docs/developer.html

This software is subject to U.S. export control laws and regulations and has been classified as EAR99.

A modeling program for integrating social, military, and gov interactions in conflict areas

Comment author: morganism 12 January 2017 11:55:44PM 0 points [-]

It would be interesting to see if you could get a subset of your agents to set up a group here, and see if you can devise a way to make them influence the model

http://unu.ai/

"UNU unleashes the brainpower of groups, combining their knowledge, opinions, and instincts into a Swarm Intelligence that is often smarter and more insightful than the individuals would be alone."

Comment author: Viliam 12 January 2017 12:56:18PM 0 points [-]

Interesting.

http://pplapi.com/4869779458.html

Age: 0 (unborn)
Language: Persian
Religion: Muslim
Income: $7909 USD
Openness [i]: above average
Conscientiousness [i]: average
Extraversion [i]: below average
Agreeableness [i]: barely above average
Neuroticism [i]: average

How exactly did they determine language, religion, income, and neuroticism of a person who wasn't even born yet?

Comment author: iandennismiller 12 January 2017 04:16:34PM *  1 point [-]

Personality characteristics are partially heritable (i.e. a small amount of variance in these dimensions is attributable to genetics) and the global distribution is not uniform. pplapi implements a simplistic model of this general system, but the point is that many of these demographic and psychometric dimensions - including language and religion - are indicated prior to birth. While there will always be exceptions, pplapi models the typical case.

From the agent you selected, income is obviously implausible but all the other dimensions are probably in the ballpark. If you went to Iran (the country of that agent), then odds are, you can find a real person with those characteristics (excepting income, of course). Even better, if you sampled from Iran on pplapi and if you sampled real people in Iran, then odds are the distributions will be similar.

I know several ways to violate the assumptions of pplapi in order to generate garbage analyses, so there's a lot of nuance here. However, I also know several interesting questions that can be answered with pplapi, so I think pplapi is nevertheless an interesting contribution.

Comment author: ChristianKl 12 January 2017 09:12:51PM 1 point [-]

However, I also know several interesting questions that can be answered with pplapi

Can you give examples of such questions?

Comment author: ChristianKl 12 January 2017 09:13:53PM 0 points [-]

What is the income of a 0 year old supposed to be? Shouldn't income be very dependent on age?

Comment author: ingive 12 January 2017 01:26:50PM *  1 point [-]

It's fake/virtual. I don't fully understand the use yet.