Krieg & KI

Präsentationsfolie zum Download: Krieg-und-KI-in-Ramberg.pdf

Von welchen Technologien sprechen wir, sprechen wir von Kriegstechnologien?

Wer spricht?

wie sprechen wir über kulturelle konsequenzen konkreter einsätze neuer technologien?

start bei 32:49:

der psychologische Effekt des Automation Bias:

der besagt, dass Menschen dazu neigen, einer automatisierten Entscheidung übermäßig viel zu vertrauen


Transfer Learning, wo bestehende Modelle angepasst werden, um eine neue Aufgabe zu lösen.






ramstein


machtkaputt-video

dual use

von welchen technologien sprechen wir, sprechen wir von kriegstechnologien in der alogrithmischen Kriegsführung?

nicht nur die us- streitkräfte source out wo es nur geht (projekt maven)

The DoD’s Project Maven initiative is using open-source code currently available to Google Cloud users to help identify objects in the military’s drone footage. The employees’ primary concern is that this project could eventually grow into using the AI to control and even deploy weapons from drones, despite company representatives stating otherwise.
https://9to5google.com/2018/04/04/should-google-develop-ai-military-drones-poll/

technologien, 
gesichtserkennung google wo?
steuerung << ??

auch die bundeswehr ... (amazon << rekognition << what adversarial??

battle management systems << bml << NLP << siri << bochum adversarial attack

bml:
Such initiatives offer the potential for C4I systems to be able to easily switch between interacting with real
people and systems, including robotics and autonomous systems, and simulated people and systems.

C4I << Command, Control, Communications, Computers, and
Intelligence 

privat << atos, siemens, bosch, fraunhofer, max planck, EU-Forschungsprogramme FP7 und Horizon2020


import boto3

BUCKET = "amazon-rekognition"
KEY = "test.jpg"
FEATURES_BLACKLIST = ("Landmarks", "Emotions", "Pose", "Quality", "BoundingBox", "Confidence")

def detect_faces(bucket, key, attributes=['ALL'], region="eu-west-1"):
	rekognition = boto3.client("rekognition", region)
	response = rekognition.detect_faces(
	    Image={
			"S3Object": {
				"Bucket": bucket,
				"Name": key,
			}
		},
	    Attributes=attributes,
	)
	return response['FaceDetails']

for face in detect_faces(BUCKET, KEY):
	print "Face ({Confidence}%)".format(**face)
	# emotions
	for emotion in face['Emotions']:
		print "  {Type} : {Confidence}%".format(**emotion)
	# quality
	for quality, value in face['Quality'].iteritems():
		print "  {quality} : {value}".format(quality=quality, value=value)
	# facial features
	for feature, data in face.iteritems():
		if feature not in FEATURES_BLACKLIST:
			print "  {feature}({data[Value]}) : {data[Confidence]}%".format(feature=feature, data=data)

"""
	Expected output:

	Face (99.945602417%)
	  SAD : 14.6038293839%
	  HAPPY : 12.3668470383%
	  DISGUSTED : 3.81404161453%
	  Sharpness : 10.0
	  Brightness : 31.4071826935
	  Eyeglasses(False) : 99.990234375%
	  Sunglasses(False) : 99.9500656128%
	  Gender(Male) : 99.9291687012%
	  EyesOpen(True) : 99.9609146118%
	  Smile(False) : 99.8329467773%
	  MouthOpen(False) : 98.3746566772%
	  Mustache(False) : 98.7549591064%
	  Beard(False) : 92.758682251%

"""