Self-driving Cars & Data Collection · time it encountered a self-driving car. Uncomfortable...
Transcript of Self-driving Cars & Data Collection · time it encountered a self-driving car. Uncomfortable...
1
Self-driving Cars & Data Collection
Privacy Perceptions of
Networked Autonomous Vehicles
___________________________
Cara Bloom, Joshua Tan, Javed Ramjohn, Lujo Bauer
2
3
4
5
Why networked autonomous vehicle (AV) privacy?
6
Why networked AV privacy? 1. Data collection
capabilities
7
Why networked AV privacy? 1. Data collection capabilities
2. Operated by a private company
8
Why networked AV privacy? 1. Data collection capabilities 2. Operated by a private
company
3. Collection of physical information in public
9
Research Goal Discover what is ‘reasonable’ data collection and use for autonomous vehicle (AV) fleets
10
Research Goal 1. What do people think
AV fleets are capable of?
11
Research Goal 1. What do people think AV
fleets are capable of?
2. How comfortable are people with AV fleet capabilities?
12
Research Goal 1. What do people think AV
fleets are capable of? 2. How comfortable are
people with AV fleet capabilities?
3. How much effort would people expend to opt out?
13
Up Next 1. Study Design 2. Findings 3. Policy Applications
14
Study Design - Exploratory online
survey
15
Study Design - Exploratory online
survey
- Privacy primed & unprimed groups
Privacy Priming Scenarios
Secondary Uses
Perceiving People
Perceiving Vehicles
Primary Uses
Image Capture
Aggregation &
Storage
Specific Incident Analysis
Continuous Analysis
16
Recognition Identification Tracking
Recognition Identification Tracking
Privacy Priming Scenarios
Secondary Uses
Perceiving People
Perceiving Vehicles
Primary Uses
Image Capture
Aggregation &
Storage
Specific Incident Analysis
Continuous Analysis
17
Recognition Identification Tracking
Recognition Identification Tracking
Necessary for autonomous navigation
Privacy Priming Scenarios
Secondary Uses
Perceiving People
Perceiving Vehicles
Primary Uses
Image Capture
Aggregation &
Storage
Specific Incident Analysis
Continuous Analysis
18
Recognition Identification Tracking
Recognition Identification Tracking
Privacy Priming Scenarios
Secondary Uses
Perceiving People
Perceiving Vehicles
Primary Uses
Image Capture
Aggregation &
Storage
Specific Incident Analysis
Continuous Analysis
19
Recognition Identification Tracking
Recognition Identification Tracking
Non-necessary. Can be achieved with same sensors.
Privacy Priming Scenarios
Secondary Uses
Perceiving People
Perceiving Vehicles
Primary Uses
Image Capture
Aggregation &
Storage
Specific Incident Analysis
Continuous Analysis
20
Recognition Identification Tracking
Recognition Identification Tracking
Privacy Priming Scenarios
Secondary Uses
Perceiving People
Perceiving Vehicles
Primary Uses
Image Capture
Aggregation &
Storage
Specific Incident Analysis
Continuous Analysis
21
Recognition Identification Tracking
Recognition Identification Tracking
Survey Questionnaire
22
Likelihood of currently occurring
Comfort if currently occurring
Survey Questionnaire
23
Likelihood of currently occurring
Comfort if currently occurring
General AV questions
Effort to opt out
Bias against Uber &
demographics
Survey Questionnaire
24
Likelihood of currently occurring
Comfort if currently occurring
General AV questions
Effort to opt out
Bias against Uber &
demographics
Seen by Primed Group
Survey Questionnaire
Likelihood of currently occurring
Comfort if currently occurring
General AV questions
Effort to opt out
Bias against Uber &
demographics
Seen by Primed Group
Seen by Unprimed Group
25
26
Study Design - Exploratory online
survey - Privacy primed &
unprimed groups
- Pittsburgh & four similar cities
27
Recruitment - Ads on Craigslist - Posts on city
Subreddits - Posters Pittsburgh only
28
302 Participants - 60% male - 25% in tech fields - Avg. age 34 [18, 79]
29
Question: What do people think networked fleets of autonomous vehicles are capable of?
Privacy Questions
Secondary Uses
Perceiving People
Perceiving Vehicles
Primary Uses
Image Capture
Aggregation &
Storage
Specific Incident Analysis
Continuous Analysis
30
Recognition Identification Tracking
Recognition Identification Tracking
31
Likelihood of Technical Capability
Primary Uses
Secondary Uses
32
Likelihood of Technical Capability
How likely do you think this scenario is to be happening now?
33
Likelihood of Technical Capability
How likely do you think this scenario is to be happening now?
Q13. A self-‐driving car recognizes a vehicle that has been seen by another self-‐driving car in the fleet
34
Likelihood of Technical Capability
How likely do you think this scenario is to be happening now?
Q13. A self-‐driving car recognizes a vehicle that has been seen by another self-‐driving car in the fleet
For example: Uber knows that different self-‐driving cars encountered the same vehicle on different days, but does not know who owns the vehicle
35
Likelihood of Technical Capability
36
Likelihood of Technical Capability
37
Difference between primary & secondary uses
Likelihood of Technical Capability
38
Likelihood of Technical Capability
22%
46%
39
ParTcipant misconcepTons
Likelihood of Technical Capability
40
Likelihood of Technical Capability
ParTcipant misconcepTons
41
Question: How comfortable are people with these potential capabilities?
Secondary Uses
Perceiving People
Perceiving Vehicles
Primary Uses
Image Capture
Aggregation &
Storage
Specific Incident Analysis
Continuous Analysis
42
Recognition Identification Tracking
Recognition Identification Tracking
Comfort with Capability Scenarios
43
Comfort with Capability Scenarios
How comfortable are you with the scenario?
44
Comfort with Capability Scenarios
45
Comfort with Capability Scenarios
DifferenTaTon is less clear than for
likelihood quesTons
46
Comfort with Capability Scenarios
DifferenTaTon is less clear than for
likelihood quesTons
47
ExcepTon: recogniTon of
vehicles
Comfort with Capability Scenarios
48
Very Uncomfortable
Q25. I would feel _______ if my car was tracked each time it encountered a self-driving car.
Uncomfortable Neutral Comfortable
Why are people (un)comfortable?
Very Uncomfortable Uncomfortable C VC Neutral
49
Why are people comfortable?
Reasonable benefit
50
Why are people comfortable?
Reasonable benefit Ubiquity
P52: “Phones are tracking in the same sense”
51
Why are people uncomfortable?
P107: “Massive privacy invasion.
Not necessary for Uber to navigate their cars.”
Necessary for AVs
Ubiquity Reasonable benefit
52
Why are people uncomfortable?
Necessity for AVs Consent
Ubiquity Reasonable benefit
53
Question: How much effort would people expend to opt out?
Q36. How many minutes would you spend in the system to successfully opt out?
Effort to Opt Out
0 1-5 5-10 11-15 16-20 21-25 26-30 >30 Minutes
0%
10%
20%
30%
40%
54
Only Priming had a significant effect on opt-out time
(M-W U = 9847.5, p=.02)
Q36. How many minutes would you spend in the system to successfully opt out?
Effort to Opt Out
0 1-5 5-10 11-15 16-20 21-25 26-30 >30 Minutes
0%
10%
20%
30%
40%
55
56
In the Paper: - More correlation
tests - Comparison of
privacy and safety comfort
- Uber-related exposure and bias
57
Takeaways
58
Takeaways 1. People differentiate
between primary and secondary data uses
59
Takeaways 1. People differentiate
between primary and secondary uses
2. Justifications focused on necessity, consent, and ubiquity
60
Takeaways 1-2: Policy Application
61
Takeaways 1-2: Policy Application Primary uses are reasonable, while secondary uses are not
62
Takeaways 1-2: Policy Application Primary uses are reasonable, while secondary uses are not Possible Exception: Recognition of vehicles
63
Takeaways 3. Misconceptions about
new information
64
Takeaways 3. Misconceptions about
new information 4. Priming had the only
significant effect on effort to opt-out
65
Takeaways 3-4: Policy Application People will likely react strongly to conversations about autonomous vehicle privacy
66
Takeaways 3-4: Policy Application People will likely react strongly to conversations about autonomous vehicle privacy
And, it may be difficult to relay accurate information
67
The Big Picture - Companies should
self-regulate
68
The Big Picture - Companies should
self-regulate .. to get ahead of the narrative .. to fulfill reasonable expectations
69
The Big Picture - Companies should
self-regulate
- Policy should restrict secondary uses of AV-collected information
70
The Big Picture - Companies should
self-regulate
- Policy should restrict secondary uses of AV-collected information
Self-driving Cars & Data Collection
Privacy Perceptions of
Networked Autonomous Vehicles
___________________________
Cara Bloom, Joshua Tan, Javed Ramjohn, Lujo Bauer
71
72
73
Exposure to Uber & AV technology -‐ 78% Pgh and 42% non-‐Pgh were exposed to media
-‐ 64% Pgh and 3% non-‐Pgh had seen one as a pedestrian
74
Bias against Uber -‐ 17% would have answered differently if Uber hadn’t been the example
-‐ 18% would trust a different AV company over Uber to have their best interests in mind
75
76
Radar
Front facing cameras
Side and rear cameras
GPS and wireless data
Data processing and storage
LiDAR
77
Progression of General Scenarios
General Proximity
78
Walking Near
Driving Near
Cycling Near
Being Near in Snow Riding In
Changes in Job Market
Image Capture
Aggregation & Analysis
Accident Liability
Becoming More Common
General Privacy
24% 61%
85% 77% 30%
25%