Re:Data Brokers: Online Game Revenue Models with AI
by pdkl95 on Sunday January 28, 2018 @05:01PM
(#16253843)
A lot of these criticisms are based on the sloppy use of terminology and "pop" misunderstandings of machine learning ("AI"). However, this is clearly some sort of sales-pitch to a business (i.e. non-technical) audience. That kind of presentation is often filled with pop buzzwords with abused definitions and exaggerated claims.
Re:Data Brokers: Online Game Revenue Models with AI
by frisco on Sunday January 28, 2018 @05:33PM
(#16253984)
I think the purported backstory of how the internet came into possession of these slides via 4chan is at least as suspicious as the technical points.
This would definitely not be the first time sketchily photographed fake “leaks” have appeared there. Who is this company? Are they real? Who gave this presentation and to what audience?
Re:Data Brokers: Online Game Revenue Models with AI
by make3 on Sunday January 28, 2018 @04:52PM
(#16253794)
Yeah, fake or not, you can definitely use ML to optimize user acquisition, to find ways to increase per user spending, to find the people most likely become spending users, to cause addiction, etc, to break people's lives.
All of these things are numerical metrics that are reasonably easy to optimize for (nothing like, complicated things like generating context appropriate speech, for which it is really hard to find an accurate metric for), and you just need to give the model as much data from as many good features as possible, and it'll end up spitting out something that works for sure
Re:Data Brokers: Online Game Revenue Models with AI
by FridgeSeal on Sunday January 28, 2018 @09:32PM
(#16254933)
Yeah some of the stuff they brag about doing is advanced to the point of difficult-to-believe.
Whilst you can do some really impressive stuff with ML currently (and that list is only going to grow), they make it sound more like an AI you'd see in a Hollywood movie.
Re:Data Brokers: Online Game Revenue Models with AI
by JamesLeonis on Sunday January 28, 2018 @09:47PM
(#16254986)
Left4Dead (1&2) featured a persistent "Director" AI that altered the player experience (sound, number of enemies, health, etc) based on what it observed the player does. You don't need a particularly advanced AI/ML to get the experience this slide deck describes.
> This monitoring system keeps specialized information contained within the device so that our company and your company is not able to take in any raw or identifiable data. This means there is no legal issues across borders.
Aka, intent to bypass GDPR and similar laws by "cooking" the data (e.g. hashing) into what DJB calls[2] "magic crypto pixie dust" that "takes personally identifiable information and makes it incomprehensible to the marketing department" or anybody else that doesn't understand how hashes can be reversed, correlated, tracked, etc.
> The software takes in things like acceleration XYZ, geoposition, SMS send/receiving timings ..., GSM strength and estimated XYZ location, nearby audio, nearby MAC address/wifi names, and nearby cell phones discovered through a completely passive side channel attack
The rest of "Schedule 'Z'" shows some of the uses for that data. After reading that, does anybody still think giving easy access to sensors is a good idea? You cannot determine the behavior of Turing complete software without running it, and you cannot enumerate badnesss.
Re:Data Brokers: Online Game Revenue Models with AI
by anonytrary on Sunday January 28, 2018 @06:19PM
(#16254206)
Not to mention the pictures taken as if they were riding a bull. The ideas in their presentation are interesting, nonetheless. Fuck professionalism! /s
Re:Data Brokers: Online Game Revenue Models with AI
by iaw on Sunday January 28, 2018 @04:04PM
(#16253473)
This is a gross deformation of what a game should be.
We've hit the point where many applications that claim to be games are more like gambling for electronic rewards, this presentation unabashedly pushes things farther away from the historic value proposition offered by games (enjoyment) and closer to the present value proposition offered by gambling (rewarding compulsivity).
Re:Data Brokers: Online Game Revenue Models with AI
by anonytrary on Sunday January 28, 2018 @06:44PM
(#16254315)
TLDR: There are two ways to justify progress in a game: proof-of-play and proof-of-pay. Stay away from games that allow the latter if you like your money.
---
Yeah this is the problem with pay-to-win games. Pay-to-win often becomes pay-to-resell-equips-for-a-profit, which is just gambling on the future value of those in-game digital tokens, items, and currency. Without a good game underneath, it eventually becomes one big valueless scheme.
Pay-to-play is often implemented in such a way that it is actually pay-to-win. A really good bullshit test on any video game is to first ask which category it falls under:
1. Pay-to-play (sufficiently constant join-fee, this is fine.)
2. Pay-to-win (DANGER!)
3. Free-to-play (often implies paying for cosmetics, this is fine.)
One should carefully understand all value streams associated with a particular game, specifically when real currency goes into the system, what comes out? Some red flags that indicate a game may be pay-to-win:
1. You can buy EXP
2. You can buy better gear
3. You can buy scrolls that allow you to progress much faster (exp boosts, spawn increases, item drop rate increases, etc.)
4. You can skip levels with money
Put simply, if Kim Jong-un can walk into the game and be the best player in under 72 hours, then the game is probably a waste of time. MMO games need to be done right, there can't be sudden phase-change events like a new player suddenly becoming the best player because they bought $84,000 in items from the cash shop (twinking is fine, since it can be backed by proof-of-play).
Re:Data Brokers: Online Game Revenue Models with AI
by TeMPOraL on Monday January 29, 2018 @01:20AM
(#16255676)
Games used to be made by people thinking of them as art/entertainment; people who wanted to give players the most fun they can, and expected some compensation in return for hard work. The market has matured now, and (AAA) games are made as a business. They are a proxy for your wallet, and the "art/entertainment" part is now not the core value, but an incidental flavour, a different makeup on a pig.
Re:Data Brokers: Online Game Revenue Models with AI
by BatFastard on Sunday January 28, 2018 @09:48PM
(#16254993)
There are enough real techniques in here that it has some credibility. Looks like some of the things I have seen marketing teams who really don't understand the tech put together.
Re:Data Brokers: Online Game Revenue Models with AI
by pimmen on Sunday January 28, 2018 @04:40PM
(#16253710)
When these statisticians, engineers and developers stand face to face with one of their beloved "whales" who lost his wife, children and home to his addiction will they say "I was just doing my job, could you ever forgive me?" or will they say what they were thinking when they implemented this shit:
"Thank god you're so predictable, sir! You're not an idiot, really, we're just really good at our job! But, now you're spent."
Re:Data Brokers: Online Game Revenue Models with AI
by make3 on Sunday January 28, 2018 @04:47PM
(#16253764)
I mean, the cigarette people for example have done much worse for much longer. Not all humans are humanists. That's why governments need to regulate what is legal and what is not, like they did in many jurisdictions with cigarettes.
Re:Data Brokers: Online Game Revenue Models with AI
by pimmen on Sunday January 28, 2018 @04:52PM
(#16253791)
And we stopped the cigarette people from advertising to children and required that they become completely transparent about their habit forming product.
As you say, we can't make everyone a humanist but we can demand that everyone be transparent when they're selling a potentially life destroying product.
Re:Data Brokers: Online Game Revenue Models with AI
by IntronExon on Sunday January 28, 2018 @05:40PM
(#16254024)
There will always be sociopaths, the desperate, the greedy, and the emotionally stunted who are willing to harm their fellow humans for a good payday. Software hasn’t really changed that, it just widened the scope.
(Regardless of whether it's fake or not, it definitely brings up a lot of interesting questions as a lot of stuff described here is not impossible...)
This would definitely not be the first time sketchily photographed fake “leaks” have appeared there. Who is this company? Are they real? Who gave this presentation and to what audience?
All of these things are numerical metrics that are reasonably easy to optimize for (nothing like, complicated things like generating context appropriate speech, for which it is really hard to find an accurate metric for), and you just need to give the model as much data from as many good features as possible, and it'll end up spitting out something that works for sure
Whilst you can do some really impressive stuff with ML currently (and that list is only going to grow), they make it sound more like an AI you'd see in a Hollywood movie.
That's a very clear confession of mens rea[1].
> Clever Indicators - Intro
> This monitoring system keeps specialized information contained within the device so that our company and your company is not able to take in any raw or identifiable data. This means there is no legal issues across borders.
Aka, intent to bypass GDPR and similar laws by "cooking" the data (e.g. hashing) into what DJB calls[2] "magic crypto pixie dust" that "takes personally identifiable information and makes it incomprehensible to the marketing department" or anybody else that doesn't understand how hashes can be reversed, correlated, tracked, etc.
> The software takes in things like acceleration XYZ, geoposition, SMS send/receiving timings ..., GSM strength and estimated XYZ location, nearby audio, nearby MAC address/wifi names, and nearby cell phones discovered through a completely passive side channel attack
The rest of "Schedule 'Z'" shows some of the uses for that data. After reading that, does anybody still think giving easy access to sensors is a good idea? You cannot determine the behavior of Turing complete software without running it, and you cannot enumerate badnesss.
[1] https://en.wikipedia.org/wiki/Mens_rea
[2] https://projectbullrun.org/surveillance/2015/video-2015.html...
This entire presentation feels dirty.
We've hit the point where many applications that claim to be games are more like gambling for electronic rewards, this presentation unabashedly pushes things farther away from the historic value proposition offered by games (enjoyment) and closer to the present value proposition offered by gambling (rewarding compulsivity).
---
Yeah this is the problem with pay-to-win games. Pay-to-win often becomes pay-to-resell-equips-for-a-profit, which is just gambling on the future value of those in-game digital tokens, items, and currency. Without a good game underneath, it eventually becomes one big valueless scheme.
Pay-to-play is often implemented in such a way that it is actually pay-to-win. A really good bullshit test on any video game is to first ask which category it falls under:
One should carefully understand all value streams associated with a particular game, specifically when real currency goes into the system, what comes out? Some red flags that indicate a game may be pay-to-win: Put simply, if Kim Jong-un can walk into the game and be the best player in under 72 hours, then the game is probably a waste of time. MMO games need to be done right, there can't be sudden phase-change events like a new player suddenly becoming the best player because they bought $84,000 in items from the cash shop (twinking is fine, since it can be backed by proof-of-play)."We are also in the process of contacting the author of http://jmlr.csail.mit.edu/papers/volume14/stowell13a/stowell... which with their help in applying their algortithm listed in the paper will help us to pinpoint certain variables we require."
"Thank god you're so predictable, sir! You're not an idiot, really, we're just really good at our job! But, now you're spent."
As you say, we can't make everyone a humanist but we can demand that everyone be transparent when they're selling a potentially life destroying product.
> The stories and information posted here are artistic works of fiction and falsehood.
> Only a fool would take anything posted here as fact.
Breadth: Ultra-Wide Depth: Ulta-Deep Noise: Very Low Problem Set: [omitted] Goal Tree: [omitted] Residual Block Weight: [omitted] Server Origin: Testing setup room past [censored]'s Gate Conference Room.
That is hilarious, I'd really enjoy it if Breadth: Ultra-Wide became a meme, it's so nonsensical.