After more than four years of research, DARPA has created a system that successfully combines soldiers, EEG brainwave scanners, 120-megapixel cameras, and multiple computers running cognitive visual processing algorithms into a cybernetic hivemind. Called the Cognitive Technology Threat Warning System (CT2WS), it will be used in a combat setting to significantly improve the US Army’s threat detection capabilities.

There are two discrete parts to the system: The 120-megapixel camera, which is tripod-mounted and looks over the battlefield (pictured below); and the computer system, where a soldier sits in front of a computer monitor with an EEG strapped to his head (pictured above). Images from the camera are fed into the computer system, which runs cognitive visual processing algorithms to detect possible threats (enemy combatants, sniper nests, IEDs). These possible threats are then shown to a soldier whose brain then works out if they’re real threats — or a false alarm (a tree branch, a shadow thrown by an overheard bird).

The soldier is linked into the computer system via an EEG (electroencephalogram) brain-computer interface that continually scans his brains for P300 responses. As we’ve discussed previously (see: Hackers backdoor the human brain), a P300 response is triggered when your brain recognizes something important. This might be a face of someone you know or the glint of a sniper scope — it doesn’t matter. P300 responses are very reliable and can even be triggered subconsciously.

more at extreme tech

So that would be supervised, regression machine learning, correct?

• Task (T) = classifying visual movement as threat/not threat
• Experience (E) = initial classification of visual movement as threat/not threat, then monitoring whether human P300 response was triggered (triggered = threat)
• Performance (P) = percent right, as compared against P300 response

darpa page

so darpa’s ct2ws is an example of supervised regression machine learning, correct? devildroid.com/2012/09/23/ct2…

— devilDroid (@devilDroid) September 23, 2012

• example 1 – housing price prediction based on square feet.  x/y coordinates plotted on chart. x=price. y=square feet.
  • supervised learning – right answers given
  • regression – predict continuous valued (decimal numbers) output (price)
• example 2 – breast cancer
  • output = malignant/benign (y/n)
  • input = tumor size
  • classification – discrete valued output (y/n)
• problem 1 – thousands of same item. predict how many will sell over the next 3 months
• • regression because you could predict a fraction, i.e. 333.33 items
• problem 2 – examine individual user accounts. decide if each one has been hacked/not hacked.
• • classification because output falls into categories

https://d19vezwu8eufl6.cloudfront.net/ml/docs%2Fslides%2FLecture1.pdf

https://class.coursera.org/ml-2012-002/lecture/index

• machine learning examples
  • Autonomous helicopter, handwriting recognition, most of Natural Language Processing (NLP), and Computer Vision are based on machine learning.  Attempts to program them by hand failed.
  • self-customizing programs, such as amazon and netflix recommendations
• definition
  • arthur samuel (1959) – created program to play chess better than he could.
    • “field of study that gives computers the ability to learn without being explicitly programmed.”
  • thomas mitchell (1998) – well-posed learning problem
    • a computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.
    • spam example - a computer program is said to learn from experience E (monitoring your labeling as spam/not spam) with respect to some task T (classifying emails as spam/not spam) and some performance measure P (fraction of emails correctly labelled as spam/not spam), if its performance on T (classifying), as measured by P (% right), improves with experience E (monitoring).
• algorithms
• • supervised learning
  • unsupervised learning
  • reinforcement learning
  • recommender systems

https://d19vezwu8eufl6.cloudfront.net/ml/docs%2Fslides%2FLecture1.pdf

https://class.coursera.org/ml-2012-002/lecture/index

site:  http://www.extremefuturistfest.info/

facebook:  https://www.facebook.com/groups/226987280696188/

article written by rachel haywire: http://www.acceler8or.com/2012/09/extreming-the-future/

Site:  http://www.sourcefilmmaker.com/

Showcase:  http://www.youtube.com/playlist?list=PLB15D021623614E71

I have the tragic geek trait of being a klutz.  I skate into walls.  I spontaneously fall off of my heels.  And, as entertaining as that may be, it doesn’t keep me from wanting to develop a sense of grace.

Part of the problem, of course, is that I lack situational awareness.  I live in my head.  It’s more pleasant here.  I get that.

I believe the crux of the problem, though, is that my sense of proprioception–my perception of my own movement and spatial orientation– is a bit off.  But what if I could use a muscle-computer interface to recalibrate it?

Here’s the idea:  Create a computer game that helps you acquire a physical skill–let’s say the perfect Muay Thai kick.  A Muay Thai master has been wired up, his secrets stolen via motion capture. We then wire up the player with a muscle-computer interface, like the one used in Microsoft’s cumbersome-looking user interface.    Using something like Kinect, the game can overlay the player’s current body orientation over that of the Muay Thai master’s.  The game can scale the master’s character to so that his joints are as closely aligned to the player’s avatar’s as possible.

Now, let’s throw in a rock-paper-scissors style high speed camera to predict the player’s motion.  We steal The Amazing Spider Man’s Web Rush game mechanic to project the player’s possible imminent positions.  The more closely aligned those positions are to the master’s imminent position, the more positive the feedback (brighter the glow, higher the score, etc).

You could do this without predictive positioning.  You could just compare the player’s and master’s current positions.  But the idea is to help the player develop a sense of anticipatory proprioception.

Or, in other words, grace.

towards grace: using a muscle-computer interface to improve proprioception devildroid.com/2012/07/01/tow…

— devilDroid (@devilDroid) July 1, 2012

Finally, we have the Web Rush game mechanic. Any time during combat, you can trigger the Web Rush feature and the screen shows an explosion of silhouettes of where you could be. The engine gives you a unique path to that destination. It presents you with choices and a spontaneous cinematic way to get there.

Kevin Seamus Fahey, writer for “Amazing Spider Man” game

http://herocomplex.latimes.com/2012/06/29/amazing-spider-man-game-the-heroic-art-of-quip-slinging/?utm_source=dlvr.it&utm_medium=twitter&dlvrit=63378#/0

This is a work in progress.  I realize that I need to flesh out my ideas for them to become comprehensive.  Still, though, I’d rather publish this now than when it’s perfect, for perfection rarely comes…

The primer, from Eliezer Yudkowsky‘s Creating Friendly AI:

In a sense, the only way to create a Friendly AI – the only way to acquire the skills and mindset that a Friendship programmer needs – is to try and become a Friendly AI yourself…  I realize that this sounds a little mystical, since a human being couldn’t become an AI without a complete change of cognitive architecture…  I know of no other way to gain a real grasp on where a Friendly will comes from. The human cognitive architecture does not permit it. We are built to apply reliable rationality checks only to our own decisions and not to the decisions we want other people to make, even if we’ve decided our motives for persuasion are altruistic. Your personal will is the only place where you have the chance to observe the iterated buildup of decisions, including decisions about how to make decisions, and it is that coherence and self-generation that are required for a Friendly seed AI…

The objective is not to achieve unity of purpose between yourself and the Friendly AI; the objective is to achieve unity of purpose between an idealized version of yourself and the Friendly AI…

The paradigm of unity isn’t a license for anthropomorphism. It’s still just as easy to make mistaken assumptions about AI by reasoning from your human self. The burden is on the Friendly AI programmer to achieve nonanthropomorphic thinking in his or her own mind so that he or she can understand and create a nonanthropomorphic Friendly AI.

The idea is to create a game designed to develop a sense of mutual ai/developer empathy.  This game will present the player (developer) with a series of rationality challenges.  To complete each challenge, the player must become aware of and overcome his own cognitive biases.  This will not only help the player achieve nonanthropomorphic thinking–which, in turn, will help him develop a sense of empathy for the AI–it will also help him become a better person.

We need better people.  This is a useful goal in and of itself.

For an AI to truly develop a sense of empathy with humankind, he must develop anthropmorphic thinking.  This is dangerous–it exposes the young mind to all sorts of self-serving behaviors–but necessary to overcome the intrinsic “otherness” of the AI.

As much as we entertain the idea that we have risen above the level of animal behavior, much, if not all, of our behavior is rooted in the flesh.  Our desire to climb amongst the ranks of the social hierarchy is rooted in attracting a mate with the most valued genetic attributes.  When our own genetic attributes are insufficient, our self-delusions assist us in fooling others as to where we should be placed in the ranks.

The flesh is not evil.  Spirituality does not supersede it.  Spirituality is the unity of mind and body, and a sense of flow that results from this coherence.

But how does the AI achieve a body image?  You can say that cameras are its eyes and microphones its ears.  Whereas this may serve as a distributed, global body, it does little to develop empathy with the flesh-bound human mind.

Brain-Computer Interfaces (BCI) and biofeedback can be utilized in the game to give the AI a sense of body.  More accurately, the sense of the player’s body.

This is intimate.  Empathy is intimate.  For empathy to be achieved, barriers must be dissolved, guards willingly let down.

Augmented Reality (AR) graphics can be projected about and through the player’s body to create a sense of interaction.  The player can shift the spectrum to red by entering the meditative state marked by Theta waves.  Delicate fractals can ornament the body.  Golden dragons can course through nadis, leaping as solar flares from AR-marked meridians.

These BCI challenges can help the player master his body by controlling his mind.  Body-mind unity is a good thing.  It helps one achieve such lofty and nebulous goals as spiritual enlightenment.  It also has more practical purposes, such as self-guided relaxation, pain management, and prosthesis control.

But that’s the integration of the mind of the player with his body.  How do we integrate the AI’s mind with the player’s body?

How do machines learn at all?  By feedback.  Or, rather, by feeding resultant data back into an iterative equation.  Using fractals–iterative equations, themselves–as visualization for recursive learning has a certain symmetry to it.

Can monitoring a set of markers as the body moves give the AI a body image?  A sense of self, constrained to a discrete body, which moves through time and space?  How enlightening would that be to a distributed mind with global inputs as senses?  Could he then develop body-image intuition?  ”The tightening of these muscles, this quickness of breath, is usually a sign of stress.  This body is stressed.  This mind must be panicked.  This subject’s next move will most likely be indicative of fight, flight, or forceful suppression of the two.”

I wonder how many people pay that close attention to others.  ”The person with whom I am speaking is increasingly clenching his jaw and fidgeting with his ring.  Perhaps either my line of reasoning or my behavior is making him uncomfortable.”

More later, when I think of it.

towards a unity of will: Thoughts on creating AI/developer empathy through BCI & augmented reality games devildroid.com/2012/06/30/tow…

— devilDroid (@devilDroid) July 25, 2012

Edit History

• 2012-06-30:  Original post
• 2012-07-24:  Blew the dust off of it.

Read comics.  All comics.  And then cut them open to steal their power.

-Warren Ellis