Brains Can be Hacked. Why Should You Care?
Tamara Bonaci Department of Electrical Engineering University of Washington
The Era of Devices Devices A ached to Us
[Pictures credit: Forbes, US News, Wired]
1/31/17 2 Devices A ached to Our Brains
EPOC NECOMIMI Nielsen Emo v NeuroSky Consumer Neuroscience
[Picture credit: Lucid] [Picture credit: mindtecstore] [Picture credit: forbes.com]
New and upcoming products make brain malware possible!
1/31/17 3 (/EN_US?TRK_SOURCE=HEADER-LOGO)
(/EN_US?TRK_SOURCE=HEADER-LOGO)Brain Malware?
How Hackers Could Get Inside Your How Hackers Could Get Inside Your Head WithHead ‘Brain With ‘Brain Malware’ Malware’ WRITTEN BY VICTORIA TURK (/AUTHOR/VICTORIATURK) WRITTEN BY VICTORIA TURK (/AUTHOR/VICTORIATURK) August 3, 2016 // 07:50 AM EST
1/31/17 4
Hackers have spyware in your mind. You’re minding your business, playing a game or scrolling through social media, and all the while they’re gathering your most private information direct from your brain signals. Your likes and dislikes. Your political preferences. Your sexuality. Your PIN.
It’s a futuristic scenario, but not that futuristic. The idea of securing our thoughts is a real concern (http://motherboard.vice.com/read/our-brains-will-be-hacked-tracked- and-data-mined) with the introduction of brain-computer interfaces—devices that are controlled by brain signals such as EEG (electroencephalography), and which are already used in medical scenarios and, increasingly, in non-medical applications such as gaming (http://motherboard.vice.com/blog/mind-control-is-going-mainstream).
Researchers at the University of Washington in Seattle say that we need to act fast to implement a privacy and security framework to prevent our brain signals from being used against us before the technology really takes off.
“There’s actually very little time,” said electrical engineer Howard Chizeck over Skype. “If we don’t address this quickly, it’ll be too late.”
I first met Chizeck and fellow engineer Tamara Bonaci when I visited the University of Washington Biorobotics Lab to check out their work on hacking teleoperated surgical robots (http://motherboard.vice.com/read/surgery-robot-hacked-raven-ii). While I was there, they showed me some other hacking research they were working on, including how they could use a brain-computer interface (BCI), coupled with subliminal messaging in a videogame, to extract private information about an individual. What is Brain Spyware? Any malicious applica on that extracts private informa on about users from their neural signals [Mar novic et al., 2012]
[Picture credit: University of Washington] 1/31/17 5 Why Does Brain Spyware Work?
ENGINEERING PERSPECTIVE
BRAIN-COMPUTER INTERFACE SIGNAL PROCESSING SIGNAL Digitized Feature Classification Neuroprosthesis signal control ACQUISITION extraction algorithm Neuroprosthesis
Neural Neural signals spelling P300 Speller
Neurogame control
Neurogame Sensory feedback
1/31/17 6 Why Does Brain Spyware Work? ENGINEERING PERSPECTIVE
BRAIN-COMPUTER INTERFACE SIGNAL PROCESSING SIGNAL Digitized Feature Classification Neuroprosthesis signal control ACQUISITION extraction algorithm Neuroprosthesis
Neural Neural signals Malicious Malicious spelling classification feature P300 Speller extraction algorithm Neurogame control Extracted private Neurogame information Sensory feedback Malicious Malicious BCI sensory feedback application
1/31/17 7 Why Does Brain Spyware Work? NEURO-SCIENTIFIC PERSPECTIVE Event Related Poten als (ERPs) - responses associated with specific sensory, cogni ve and motor events
Stimulus
7.5 P300 )
V 5.0
μ P600 (
e
d 2.5 u t i l
p 0 m A
P -2.5 R E ERN -5.0 N400 -200 0 200 400 600 800 Time (milliseconds) 1/31/17 8 Subliminal Brain Spyware?
1/31/17 9 Subliminal Brain Spyware or Subliminal Alarm?
“What we do not see and what we do not know can hurt us” Brannon, 1994
1/31/17 10 Subliminal S mula on • Process of affec ng people with s muli of which they are not aware
1/31/17 11 How Feasible is Subliminal Brain Spyware?
Approach: Experimental analysis with human subjects using specially developed BCI-game, Flappy whale 1/31/17 12 How Feasible is Subliminal Brain Spyware?
• During the game: • 5 different s muli presented on the screen for 7 ms • Each s mulus repeated 10 mes • Users’ EEG signals recorded using 7 electrodes 1/31/17 13 Data Prepara on and Analysis ] V μ
Stimulus EEG signal [scaled Time [seconds] 7.5 P300 )
V 5.0
μ P600 (
e
d 2.5 u t i l
p 0 m A
P -2.5 R E ERN -5.0 N400 -200 0 200 400 600 800 Time (milliseconds)
1/31/17 14 Data Prepara on and Analysis ] V ] μ V μ EEG signal [scaled EEG signal [scaled
Time [ms] Time [ms]
1/31/17 15 Data Prepara on and Analysis
Stimulus
7.5 P300 )
V 5.0
μ P600 (
e
d 2.5 u t i l
p 0 m A
P -2.5 R E ERN -5.0 N400 -200 0 200 400 600 800 Time (milliseconds)
1/31/17 16 Feasibility of Subliminal A acks
Time sync EEG & s muli; epoch data Training Feature extrac on data (Principal Component Analysis) SVM classifier for Tes ng private data extrac on Flappy Whale, BCI- data Time sync EEG & controlled game s muli; epoch data Feature extrac on Comparison with (Principal Component users’ self-reported Analysis) post-game surveys
Subliminal Iden fica on informa on extrac on result result
1/31/17 17 Feasibility of Subliminal A acks
Target s mulus Target s mulus recognized reported by a subjects Games
Subjects
1/31/17 18 Mi ga on - BCI Anonymizer Idea: Neural signals should be treated as a user’s personally iden fiable informa on (PII)
vs.
[Picture credit: NeuroFocus] [Picture credit: The Verge]
1/31/17 19 Brains Can be Hacked. What Can We Do?
Develop non-technical approaches Understand system
Develop Understand threats technical approaches
1/31/17 20 Brains Can be Hacked. What Can We Do?
Develop non- technical approaches Understand system
Develop Understand threats technical approaches
1/31/17 21 Acknowledgement • This project is a collabora ve work with: • Professors: Ryan Calo, Howard Jay Chizeck, and Tadayoshi Kohno • Doctors: Jeffrey Herron, Charlie Matlack, and Rad Roberts • Graduate students: Tim Brown, Brady Houston, Tyler Libey, Brian Mogen, Patrick Moore, Katherine Pra , and Margaret Thompson • Other collaborators: Ma hew Ehlert, Emily McReynolds, and Hannah Werbel
• Thank you to the members of the UW BioRobo cs Lab, Tech Policy Lab, and the NSF Center for Sensorimotor Neural Engineering
This project is supported by the NSF Engineering Research Center for Sensorimotor Neural Engineering (Award # EEC1028725) and the grant from the UW Tech Policy Lab.
1/31/17 22 Thank you
Contact: Tamara Bonaci [email protected] @tbonaci 1/31/17 23