"The Massively Multiplying Mini Me" is a fascinating post by Alvis Brigis, a social futurist and entrepreneur, who notes the work by innovators like Steven Wolfram to create a quantified self based on the personal usage data gleaned from your computers, devices, and social networks. Imagine if everything you said, expressed that you liked or disliked, all your travels and social interactions, were aggregated into a single profile. You could create an AI based on you. Here's where things get head-spinning, because Alvis has a brilliant if somewhat discomfiting idea:
Imagine taking all of your Gmail, Facebook, computer and sensor data, then putting it into a simple 3d avatar that looks like you and sits in a window on your Facebook page or anywhere else on the web. Perhaps this Pet You is displayed on a wall in your house. You can interact with this Pet You - ask it questions, push it around, introduce unexpected elements... You yourself, family, friends or other users can then rate each interaction to guide the development of this Mini You, essentially raising it by rewarding the most desired behavior...
After interacting with this AI long enough, it'll probably start to evolve, or at least get quirky. And here's it gets really cool, or at least a little eerie:
Over time, some of these growing Mini Yous are bound to get interesting. These are the ones that will literally survive and thrive by being shared on different people's Facebook pages, via email, in public settings and... this is where it can get really interesting ... as Non Player Characters (NPCs) in bigger virtual environments like Second Life, single player big world video games like Skyrim & Assassin's Creed, as cleverer A.I. in first person multiplayer games like Call of Duty & Halo...
Read much more from Alvis here. And the thing is, while it might seem like wild speculation, I've know of many projects in development now that are pushing into this realm of AI. So if you're not ready yet, prepare yourself for the rise of Mini Mes (and Mini Yous).
Given that Facebook owns your data, the phone company owns your dialing habits, google owns your web habits...
I have to wonder, who will own the AI you? If it's not your computer code running on your computer, why should you believe it belongs to you?
You think your data is yours? Tell the phone company to delete your phone records then -- see how well that goes.
Posted by: shockwave yareach | Tuesday, April 10, 2012 at 11:16 AM
I would love this. I actually already started programming a cardboard cutout of myself in Second life that when someone in local chat says certain words it triggers a response from the cutout.
Posted by: Seymore Steamweaver | Tuesday, April 10, 2012 at 11:41 AM
@shockwave - right on, Facebook is sitting on a TREASURE TROVE of social data. it takes thought experiments like this to realize just how much value could be generated from that data + the benefits/risks. I'm just about done with an Open Foresight project on Facebook futureoffacebook.com in which some of these futures are discussed. Check it out! Especially the Society video.
Posted by: A Facebook User | Tuesday, April 10, 2012 at 02:48 PM
@Seymore Steamweaver - love it. "Sorry, I am not home right now." got any screengrabs of your Mini Me or Mini Seymmore in action?
Posted by: A Facebook User | Tuesday, April 10, 2012 at 02:50 PM
Sure, Fbook User. Here's a quick one with a front and backside shot of it. https://lh3.googleusercontent.com/-VUayttalXlM/T4T9OFLxm8I/AAAAAAAAAEw/pRkXUTkoCSg/s728/Seymore_talking.png
Posted by: Seymore Steamweaver | Tuesday, April 10, 2012 at 08:51 PM
I can see this technology being used for criminal profiling as well as advanced forms of identity theft / impersonation. But hey, it's being introduced as a cool toy to play with. Never mind that it will cast a very long shadow across our lives.
Posted by: Alazarin | Wednesday, April 11, 2012 at 12:43 AM
Sadly, I'm afraid any AI trying to compute "what would Arcadia say?" is headed for a digital breakdown.
Posted by: Arcadia Codesmith | Wednesday, April 11, 2012 at 12:05 PM
Personal usage data ? Up to this day Amazon recommends me guns and ammo stuff because once i gifted it to my uncle. My kids share my Netflix which gives the Netflix "top 10 movies for Renmiri" completely useless.. and kinda comic. The poor software tries to make sense of someone who loves Anime, kids movies and current news and fails on all 3 categories!
Until our "usage data" collection devices get smarter this is what you will get: a "mini me" with multiple personality disorder
Posted by: Renmiri Writer | Friday, April 13, 2012 at 10:32 PM
Look, sorry, this is nonsense.
First off, this guy did not come up with it... it was the central plot of a flopped series, Caprica.
But we will set thin aside for a moment.
If you simply aggregate all of the random junk data that a person creates, you will find that it is essentially random. You could simply create a chart of all of the aggregated data for people everywhere, and be capable of generating data for fictitious people that looks like a real person's data.
Such models are excellent for marketing and research, but there are limitations. Saying Person A likes vanilla better than chocolate might be interesting, but it provides no understanding as to WHY they prefer it.
Unfortunately, the data is not enough in and of itself... you need to provide the model with understanding in order to pretend to create intelligence. The nature of that understanding *IS* creating Intelligence.
What this essentially means is that you are simply draping a personality over an already created AI... the personality does not make the model any more or any less intelligent.
Then you jump to claim how these intelligences will evolve to fill roles in computer games? For real? Like really real?
What makes you think someone's shopping habits would miraculously make an AI better at playing Halo?
Since we would have to write the AI from scratch anyway, and simply drape the personality across it, then the only gain we have made, here is personality... not mad skillz, not hyper intelligence... just your sense of style for example.
Which leads us to the final problem.
If I am writing a game, especially one with a well crafted story line, then the characters need to be well crafted as well.
Just like real life actors, their personality becomes secondary to the personality of the character they are playing.
So why wouldn't I simply custom create a personality created for specifically the role that it is being used for, rather than one off the rack, which would be suppressed for the role anyway?
This entire idea is suitable for nothing more than it was already used for... the plot of poorly written scifi.
Posted by: Steve | Thursday, July 05, 2012 at 10:18 AM
@Steve - Sure, Caprica hit on the idea of feeding personality traits into an avatar of the self that then went on to become runaway Hard AI. I'm sure that's been written up in scifi prior to that as well. That's not the central concept I proposed here.
The key concept is 1) pockets of personal social media data can pool together, 2)these can be fed into a basic Q&A or poke-and-react model that gets rated by human users, 3) these ratings filter interesting reactions / behaviors to the top of the queue , 4) "Over time, some of these growing Mini Yous are bound to get interesting.", 5) through a market process, these traits be applied to NPCs not to (at first) fully govern their behavior, but to enrich it appropriately by guiding conversation, adding physical behaviors, or such. I could see such "attributes packages" becoming useful to character designers / coders who are looking to quickly add nuance to some very general characters without spending a lot of time or programming / writing hours. Eventually, the richness of the personality / behavior data, the sorting market for interesting things and the fundamental character A.I. code will probably evolve to fit together nicely.
I agree that the max extreme scenario won't come into being overnight. But the gradual development of personality attributes, bottom up review of those, and then language and systems that allow those attributes to be applied quickly seems to be underway already, per James' closing. My bet is that A.I. won't all come from top-down code all at once.
Posted by: plus.google.com/112868036618386137011 | Thursday, December 26, 2013 at 12:32 PM