Assured Autonomy Workshop Series

Key Questions for Workshop 2

1. What are the key problems the field of the panel (referred to as “your field” hereafter) thinks about?

2. What are some problems that your field views as solved?

3. What are the current research trends in your field?

4. What are some problems that your field views as not solved but important?

This is the second in a series of three workshops on Assured Autonomy. See the landing page for this series here to learn more.

This workshop will focus on identifying existing capabilities, current research, and research trends that could address the challenges and problems identified in workshop 1.

The workshop will be structured in order to stimulate cross-disciplinary exchange. The tentative list of research areas to be covered includes the following:

• Safety, verification, test and certification.

• Security and privacy.

• Human-autonomy integration and trust.

• Policy, regulation and ethics.

Additional discussions—cutting across the research areas—will concentrate on several applications of  autonomy in mobility, health & medicine and the civilian infrastructure.

The expected outcome of the second workshop is an interim report which will provide inputs to the planning of and the discussions during the third workshop and to the final roadmap document.

February 20, 2020 (Thursday)

12:00 PM LUNCH | Sundance Room

01:00 PM Introductions | Cowboy Artist Room

02:00 PM Panel 1- AI | Cowboy Artist Room

• Tom Dietterich, Oregon State University

• Benjamin Kuipers, University of Michigan

• Alexander Ray, OpenAI

• Ashley Llorens, JHU APL (Moderator)

03:00 PM BREAK

03:30 PM Panel 2- Verification | Cowboy Artist Room

• Behcet Acikmese, University of Washington

• Julia Badger, NASA

• Swarat Chaudhuri, University of Texas at Austin

• Paulo Tabuada, UCLA

• Lenore Zuck, UIC (Moderator)

04:30 PM Breakouts | Cowboy Artist Room, Suite 324, Suite 3260

6:00 PM Discussion | Cowboy Artist Room

07:00 PM DINNER | Sundance Room

February 21, 2020 (Friday)

07:30 AM BREAKFAST | Sundance Room

08:30 AM Panel 3- Ethics/Policy/Societal Impacts | Cowboy Artist Room

• David Danks, CMU

• Michael Fisher, University of Liverpool

• Cara LaPointe, JHU APL

• Katina Michael, Arizona State University

• Nadya Bliss, Arizona State University (Moderator)

09:30 AM Panel 4- Security/Privacy | Cowboy Artist Room

• Anton Dahbura, Johns Hopkins University

• Jeremy Daily, Colorado State University

• Gregory Falco, MIT

• Ryan Gerdes, Virginia Tech

• Howard Shrobe, MIT (Moderator)

10:30 AM BREAK

11:00 AM Breakout | Cowboy Artist Room, Suite 324, Suite 326

12:00 PM LUNCH | Sundance Room

01:00 PM Discussion | Cowboy Artist Room

01:45 PM Panel 5- Human Interactions | Cowboy Artist Room

• Erin Chiou, Arizona State University

• Katherine Driggs-Campbell, UIUC

• David Hendry, University of Washington

• Ben Shneiderman, University of Maryland

• Nancy Cooke, Arizona State University (Moderator)

02:45 PM BREAK

03:00 PM Writing | Cowboy Artist Room

Autonomy is becoming mainstream. The anticipation is that cyber-physical-human systems and services enabled by autonomy will improve the future work conditions and the quality of life for humans and create new business models. To name a few examples, autonomous cars are test-driven on public streets by numerous companies, teams of robots that share the workspace with humans are showcased at airports and hospitals, new civilian and defense applications for drones surface by the day, and more and more human responsibilities in critical applications, including but not limited to infrastructure networks and medical diagnostics and hospital management, are shared with autonomous decision-makers.

On the other hand, a number of looming challenges—whether autonomous systems are safe and secure, whether we can assure their safety and security, whether humans will ever trust and work with them, whether we can integrate them at scale and whether we can do all these economically—overshadow the popular belief that a revolution driven by autonomy is imminent. Report after report warns of the technological and societal consequences if these challenges are not addressed.

Why does a revolution seem still so far despite the fact that the number of examples of autonomous systems is booming? We believe that the answer is not only in our lack of a scientific foundation, tools and engineering principles for sustainable development and deployment of autonomous systems with appropriate assurance on safety and security but also in the lack of awareness of the commonalities, misunderstandings and gaps in understanding across disciplines, sectors and communities. Furthermore, the barriers for the integration of autonomous systems at scale and with significant societal impact do not only stem from technical factors but also factors related to regulation, policy making, education, workforce development, capacity building and distribution of wealth.

The challenge of establishing assurance in autonomy is attracting rapidly increasing interest of the industry, government and academia. A vast range of industrial sectors including defense, mobility, health care, manufacturing and civilian infrastructure are embracing the opportunities in autonomy yet face the barriers toward establishing the necessary level of assurance. Numerous government agencies are poised to tackle the challenges in assured autonomy. Controls, computer science, machine learning, artificial intelligence, human factors, communication, perceptual and cognitive sciences, business, law, ethics and public policy are only a few among the disciplines that address the underlying scientific problems in assured autonomy. Given the already immense interest and investment in autonomy, we argue that it is exactly the right time to organize an international workshop to facilitate a dialogue and increase awareness among the stakeholders in the industry, government and academia.

This series of three workshops aims to help create a unified understanding of the goals for assured autonomy and the research trends as well as near-term, mid-term and long-term research needs supporting these goals.

Workshop 1 (October 16-17, 2019 in Arlington, VA) — Identify current and anticipated challenges and problems in assuring autonomous systems within and across applications and sectors.

See the workshop website here.

Workshop 2 (February 20-21, 2020, Phoenix, AZ) — Identify existing capabilities, current research, and research trends that could address the challenges and problems identified in workshop 1.

Workshop 3 (tentatively April 2020, location TBD) — Create a roadmap for assured autonomy that will be usable by Government agencies for building and refining research and development programs and science and technology policy-makers.

Workshop 1 - Overview

This is the first in a series of workshops on Assured Autonomy.

This workshop was focused on identifying current and anticipated challenges and problems in assuring autonomous systems within and across applications and sectors.

The workshop was structured in order to stimulate cross-disciplinary exchange. It concentrated on several challenge areas and application domains. The challenge areas include the following:

• Security and privacy.

• Safety, trust and verification.

• Test, evaluation and certification.

• Human-autonomy integration.

• Policy, regulation and ethics.

The application domains that will be represented at the workshop include the following:

• Mobility.

• Defense.

• Health and medicine.

• Civilian infrastructure

• Space.

The expected outcome of the first workshop is an interim report which will provide inputs to the planning of and the discussions during the second workshop and to the final roadmap document.

Workshop 1 - Agenda

09:45 AM Panel: Mobility | Galaxy Room

Kevin Dopart, DOT
Natasha Neogi, NASA
Brian Sadler, ARL
Tichakorn Wongpiromsarn, nuTonomy

1:00 AM Panel: Privacy and Security | Galaxy Room

Miroslav Pajic, Duke
John Launchbury, Galois
Todd Humphreys, UT Austin
Howie Shrobe, MIT

2:45 PM Panel: Safety, Verification, Certification | Galaxy Room

Phil Koopman, CMU
Darren Cofer, Rockwell Collins
Laura Humphrey, ARFL
Julian Goldman, Mass General
Meeko Oishi, University of New Mexico

04:15 PM Panel: Space | Galaxy Room

Danette Allen, NASA
Joel Mozer, Air Force
Hiro Ono, NASA
Missy Cummings., Duke

08:45 AM Panel: Policy, Regulation, Ethics | Galaxy Room

Cara LaPointe, JHU ARL
Heather Roff, JHU ARL
Nadya Bliss, ASU
Lenore Zuck, UIC

09:30 AM Panel: Defense | Galaxy Room

Craig Lennon, ARL
Signe Redfield, NRL
Ashley Llorens, JHU APL

0:45 AM Panel: Human System Integration, Trust | Galaxy Room

Jessie Chen, ARL
Matthew Johnson, IHMC
Nisar Ahmed, University of Colorado
Missy Cummings, Duke
Nancy Cooke, ASU

Organizing Committee:

Nancy Cooke (Arizona State University)

Missy Cummings (Duke University)

Ashley Llorens (Johns Hopkins University, Applied Physics Laboratory)

Howard Shrobe (Massachusetts Institute of Technology)

Ufuk Topcu (University of Texas at Austin)

Lenore Zuck (University of Illinois at Chicago)

Before the Conference in discussions I had with my panel I noted the following (notes of my discussion taken by Helen Wright):

• We need to focus on the individual, human machine operator, who is enhanced technologically (black box recorder, feedback loops, wearables, implantable, etc.)

• Human in the loop vs. human out of the loop 

• Looking at cross overs between surveillance of individual and how that affects their decision making, as opposed to their reliance on devices

• Evolutionary change in next 20-30 years

• What does a process of consent look like, what does it mean, what do we gain access to?

• Autonomy of individuals - their human rights - can we have autonomy when we have interactions with devices

• How do devices alter people’s decision-making: playing to a global theater (e.g. uberveillance)

After the conference, the 5 recommendations I provided at the end of the 2nd Assured Autonomy Workshop were as follows:

• Recommendation #1: Better understand the assured autonomy landscape in terms of governance- hard and soft laws; regulations and guidelines.

  • Milestone #1: The identification of existing laws and regulations that are relevant to assured autonomy.

  • Milestone #2: The identification of standards and protocols globally in use. These are specific to nation states.

  • Milestone #3: The identification of policies and procedures related to assured autonomy globally. These may be at the institutional or organisational levels of detail.

• Recommendation #2: Focus attention on human autonomy in the context of assured autonomy.

  • Milestone #1: Identify human values that may be overridden by inherent values carried by autonomous systems as applied to given use cases.

  • Milestone #2: Consider the psychological effects of “human out of the loop” technology on people and the potential for harm when a “human mule” is involved.

  • Milestone #3: Define how humans can always be in the loop in decision making use-cases of autonomous systems and address questions like “should autonomous systems kill”?

• Recommendation #3: The creation of meaningful stakeholder relationships for those who have a direct or indirect interest in the development of autonomous systems.

  • Milestone #1: Reach out to public interest technology advocates in the context of assured autonomy.

  • Milestone #2: Reach out to government and non-government organisations in the context of assured autonomy.

  • Milestone #3: Reach out to commercial entities already engaged in the development and deployment of autonomous systems, and ask them to share findings with citizenry.

• Recommendation #4: “Ordinary” citizens should be involved in the creation and development of autonomous systems that affect them, directly or indirectly.

  • Milestone #1: Use innovative “participatory” techniques to gather citizen sentiment about particular design decisions when it comes to new autonomous or semi-autonomous technologies.

  • Milestone #2: Use the opportunities when engaging with members of the public as an educational intervention to raise awareness of new technologies- civilian, commercial or military.

  • Milestone #3: Provide training opportunities for those who seek them with respect to “future of work” by offering free online courses to those who wish to engage further throughout the lifecycle of the development and deployment of a new technology. 

• Recommendation #5: Research technologies that are a clear breach of human autonomy and are on the market or under development today.

  • Milestone #1: Identify technologies that encroach on bodily integrity. E.g. non-medical implantables.

  • Milestone #2: Identify technologies in given use cases that are negatively impactful on the privacy of the individual. E.g. mobile apps that gather personal health information and use that information to construct addiction by design algorithms toward repeat use.

  • Milestone #3: Develop strategies toward blockchain technologies that provide auditability from the context of authorship, date, place, and source of data creation. Acknowledge that these oversight technologies in themselves are fallible and require human-led governance structures at multiple levels.

During the meeting I took these notes:

Tom Dietterich, Oregon State University

By construction

By runtime monitoring

Predict cyberattack: class probabilities, distribution, decision boundary, predict

Cockpit research mgmt.: be more reliable (Todd LaPorte, Gene Rochlin, Karlene Roberts): “High Reliability Organisations”

-          Situational awareness

-          Reluctance to simplify interpretations

-          Culture of safety: undiscovered failure modes of their system (novel undetected faults). Constantly be watching for anomalies and near misses; treating them how to figure out why that happened. Alternative hypotheses for what is going on—commitment to resilience

-          Sensitivity to operations

-          Deference to expertise

Design AI Systems to be HROs

-          Maintain situational awareness

-          Detect anomalies and near misses

-          Generate candidate explanations for anomalies and near misses

-          Improvise solutions

Design a Human + AI Teams as an HRO

-          Even powerful AI systems will be surrounded by a human team

-          Situational awareness

o   AI can track the situation

o   Humans must be aware of what version of the AI system they are using.

o   Detect anomalies and near misses

§  AI system must understand and predict behavior of a human team

§  AI and humans must work together: interactive anomaly detection

-          Generate human explanations

·       You don’t need all those people- only in emergencies—but that is the pressure but the reality is that we need the people.

Benjamin Kuipers, University of Michigan

Ethics is a tool for a society to encourage its individual members to behave in cooperative ways that benefit the society.

Trust enables cooperation. Distrust discourages cooperation and damages society.

Social norms and ethical principles.

-          We will drive on the right side of the road

Trust is the willingness to accept vulnerability, with confidence that it will not be exploited.

              Cooperation requires vulnerability

A prospective cooperative partner must be trustworthy

Ethical Principles leads to

Trustworthiness leads to

Trust

Cooperation                                   Social Norms

Positive Sum outcomes                Save resources for defence and recovery

More resources for society

Knowledge and humility

The world is infinitely complex.

Knowledge is finite.

-          “The Blind Men and the Elephant”

Moral Philosophy

Deontology (What is my duty, to do or not to do)

-          Pattern matched rules and constraints

Virtue Ethics (What would a virtuous person do?)

-          Case based an analogical reasoning

Utiliarianism (What action maximises utility for all)

-          Game theory/ decision theory

- These are human constructs

We are designing intelligent agents that participate in our society

-          Other intelligent agents (humans, institutions) also participate in our society.

What does the purpose of ethics imply?

-          For society to thrive, its members (humans, AIs, institutions) should behave ethically

-          To create trust, to encourage cooperations

-          Otherwise, society suffers.

Which social norms for AIs?

-          If a social norm is not respected by members of the society, it is weakened. People stop being able to trust it.

What do I need to trust?

-          My vulnerabilities will bot be unfairly exploited

Questions in context:

-          What vulnerabilities do I have?

-          What are the potential exploitations?

-          What social norms would discourage the exploitations?

-          What punishments should violators receive?

A proposed methodology for designing ethics—contexts.

Cooperation is essential.

-          There are existential threats to humanity.

o   Super-intelligence.

o   Climate change. NEED TO THINK ABUT THI

-          To meet these existential threats

o   Cooperation will be essential

o   Cooperation depends on trust

o   Trust is being eroded

o   We must do what we can

OpenAI – Alexander Ray

-          Even though data is being generated by humans we don’t know if it is valid- or valuable.

-          Metrics and quantitative.

https://en.wikipedia.org/wiki/ImageNet (ImageNet)

Negligible – can be neglected.

Widen the distribution

Balance- robustness and making progress…

But if you have a wide distribution you cannot zone in on what the other policy of the other cars.

How do you go back to balancing safety with more assertiveness.

Google: How do the cars obey the road rules? But that is wrong way to approach the problem. “I’m going to next”.

Shoddy AI: implementations where people do not understand what they are doing. Package slick and they have somethingà but different methodologies won’t work if people are not paying attention. Change of shifts, personnel change, throw system off. Is there enough awareness to adjust the AI decision system as well.

Team A and B: A thinks they did great; but second learning system is the credit for it and not 1.

A/B testing.

Change the world and then change our inputs.

Model the feedback loop to work.

https://www.openai.com/

·       Big data and internal questions…

What are the inputs and outputs? Do we have storage to do everything? Capture? Privacy issue—voice recorder… the flight recorder… blackboxes for machine learning algorithms…

Autonomy

Can we be sure make right decisions, make right actions, if you have devolved decisions what will they do? Trustworthiness?

Is this reliable?

Does it have good intentions?

If you have a friend who makes a mistake, they have good intentions

Devolving decisions to a system?

Screen on chest of what it is thinking

Regulator

Strong verification is good; formal verification is too complex and takes a long time; big stack of neural netsà

We wrote the algorithms, we can explain what they are thinking and what they are doing? Expose those things in a symbolic component.

Convince people: “why did you do that”? Symbolic AI; machine learning. Record decisions. Regulator have assurance? Always right reasons; it will be safe; learning. If you build systems well; do strong verification; trustworthy autonomy.

33M pounds: Trustworthy Autonomous Systems (Trust, Resilience, Ethicists, Psychologists)

Explainable AI. Information to users is ‘great’ if you are a ML researcher. You never use ML – decision making process is analyzable, transparent—not an opaque box. The regulators would just get scared.

Viewing as a property of AI; use of AI from functions.

Different explanations of members of public.

Intentions: drone and intention?

Which stakeholders are you talking about?

Accountability; responsibility

Civilian deaths—from the operator of the drone?

Autonomy of the drones—flying the drones; kinetic targeting.

Liability; regulators; insurance; where does that liability fall?

Terrified how to insure autonomous vehicles—liability discussion (ethical and public perception)

Explainability and accountability

License—use everywhere “no more restrictions”

Pharmaceuticals: gradually over time allow it with like a certification—over time not too restricted

A/S safe operating envelope: what is safe and what is not? Tech can inform reg?

How do we minimize liability?

Quantification of ethics: a/s under water vehicle proj. Score to maximise. Could drown passengers and get to place faster.

Run out of rules – ethical principles and norms—bunch of priorities—avoid hurting people.

Trolley problem occurs—no plan about that; no regulation.

Ethics as a practice. What are the ethically acceptable and permissible things?

Are these ok?

Space of permissible things.

>> brain… independent oversight

Internet of Brains…

Oversight—domain specific. Developing new concepts; new distinctions.

A/S vehicle: Level 1-5 (5 is unreachable; level 3 is infeasible… level 4 is key concept: operational design domain: Phx during good weather A/S can be released… operational design domain can increase)… concept that can be addressed technically—can incorporate and when not ready.

Assured Autonomy Session

Security is necessary condition for Assured Autonomy

An autonomous systems attack

New modes of attacking—libraries, TensorFlow vulnerability list (image processing libraries)

Cyber vulnerabilities—specifically autonomous level and then underlying levels of software. Conventional cyber attacks.

Adversarial machine learning for visual object covering- how does the machine know?

Risk-sensitive adversarial learning for a/s

Competition between agent and environment that trying to reconfigure itself. Feedback loop.

Flagship- transportation- public safety & security, smart cities, robotics, unmanned drones, and healthcare space.

Cybertruckchallenge.org – Jeremy Daily

>> Cybersecurity—trucks smashing what if they are. We like breaking things.

>> Can cyber attacks lead to kinetic results and how can we tell?

Transport of high-risk/high-value cargo

How to generate talent to address specific problems.

Develop next generation workforce- industry, government, professionals, establish community of interest for heavy vehicle cyber that transcends individual companies

Security challenges

-          Human

-          Corporate

-          Machine

-          Network

-          Ethics and Responsible disclosure?

Cybertruckchallenge June 21-26, MI

API Security

Scale + from automating to autonomy

Ryan Gerdes – Virginia Tech (Break things that tend to move)

Legacy systems

Detecting and Localising Compromise—what is happening?

Update a system when it is compromised

Lack of adversarial mindset

Adaptiveness via machine learning

Convergence of systems

Systems have UNKNOWN VULNERABILITIES IN THEM

Manipulate pixels before classifier; or modify objects of interest

·       Inject noise injected into the camera….

Safety and security is not the same

Safety thinking cannot be applied to security problems.

-          Can jam messages between planes; spoof aircraft and make phantom aircraft (messages are authenticated); messages are periodic and there is delay—attacker is closer or father away.

- Few thousands dollars--- ADSB messages can be spoofed. Secure localization. Sensors for systems cannot be trusted. Radar systems can be hacked. We need ways to make secure measurements.  Authenticated communications- making it work in legacy systems.

Define the criticality, complexity, assurance, harm…

Nancy Cooke Panel

Humans are more important than machines

Human considerations need to happen early and often

Erin Chiou

Human-Agent Coordination

Soft constraints

-          Achieve operational excellent

-          Obtain competitive advantage

Popular approaches:

-          Proceduralise

-          Shift control

-          Add technology

Procedures address the knowns, reduce ambiguity

Front-end control enables discretion when unknowns occur and requires expertise

-          Expertise is expensive

Technology increases control… tech should not be evaluated at internal structure alone, but on their role as coordination mechanisms.

Human systems; human cognition.

System resilience is graceful extensibility

Requires people are able to make right decisions under defined circumstances.

The proposed approaches:

-          Trust

-          Cooperation

-          Accountability

·       Situation structure that the human and autonomy… whether trust is needed; is cooperation a solution.

Trust is useful in the absence of complete control.

Trust guides but does not completely determine

Cooperation to forego individual goals for shared goals

Accountability are about social pressures that impact decisions.

-          Especially when social sanctions are present

Agent behaviours affect human behaviours!

Interaction structures affect teams

Increasing accountability pressures – pause and think than reactionary?

Consider process-based performance and the ability to resolve conflicting goals.

Build capabilities that allow for graceful extensibility

Human systems integration; interoperability

Cannot do just studies in the lab

Understanding naturalistic settings

Theory development and implementation… how to keep the human in the loop; designing for worker empowerment, human-machine interdependence and system resilience

Problems in Human-Centered Autonomy

Safety over time… Emergency of Autonomy in Planes

All Aircraft

1st Generation

2nd

3rd

Fourth

1960-2010…. Spike initially… Dominique Chatterenet, Air Transport Safety Tech and Training, ETP 2010. Airbus—how to train pilots…

Fatalities per million departures… even if we have automation that is validated and tested we have to think what happens when we put it to practice…

-          Transportation (autonomous driving)

-          Agriculture

-          Manufacturing

Measure of structure vs Measure of Autonomy

-          Full automated assembly line

-          Collaborative task completion

-          Assurances—how much structure specific to problem. What is domain specific.

David Hendry

-          Value Sensitive Design Lab (Seattle, UW)

-          Why this sociotechnical future?

o   Who will benefit from this future? What other ones have we put aside?

o   Who will absorb the risks? Engineering process.

-          Moral imagination; technical imagination

-          EU AI Feb 19, 2020

o   Human agency and oversight

o   Etc…

Human dignity:

-          Vague values that need to be defined (lift patience up)

-          See van Wynsberghe, A. (2013). Patient looking at robot

-         Exoskeleton: Augmentation of person.

-          Care ethics… explored the space.

-          More appropriate approach…

Prof Batya Friedman in 1990s

-          Value Sensitive Design: social impact statements of technology

-          An interactional theory and method that accounts for human values in a principled and structured manner throughout the process.

Level of human experience

-          Individual… small group… organisations… social policy… global

-          Van de Hoven, J. 2017:

o   Two solitudes; bridge the gap

o   Applied philosophers; legal theorists (values, norms, laws, policies)

o   Engineers work in black and white (how do we bring them together)

Human values: human flourishing, environment, sustainability, human fignity, quality of life, well-being, justice, equity

Engineering Values

Value hierarchy:

-          Values: quality of life – for sake of links

-          Norms – objectives, goals, constraints

-          Design requirements – infrastructure, device, policy

Traceable way of doing analysis… we need a working definition of goals, constraints…

Responsibility

Theory and practice

Appropriation

Co-evolve tech and policy

Standardisation

Education

Ethics washing

Stakeholders

Value tensions

Accounting for power

Worst case planning

Saying no

Planet finite, yet regenerative

Ben Schneiderman

Human-Centered AI

-          Amplify, augment, enhance, empower people

-          It is a design sciences--- how to people become 1000x more effective

-          Reliable, safe and trustworthy

Reliable, Safe, Trustworthy

Audit trail, product logs, civil aviation, flight recorder/voice recorder- petabyte of data crossing Atlantic. Then there is analytics of it.

Large systems… flight data recorder in it.

Testing, fairness, explainable

Independent Oversight: regulation, auditing, insurance, industry (trustworthy)

Organisation: culture of safety, openness (save)

Systems: Flight Data recorder, tested, fair, explainable (reliable)

Do people cover over their faults? Where are the near misses?

National algorithm safety board (NASB)

The familiar model is the zoning codes and the zoning boards… how to build a building!

Assured Human Autonomy

Assured Human Control

Trustworthy Technology

Reliable Automated Systems

Enhancing Human Control of Advanced Systems

Reliable, Safe & Trustworthy Technology

·       Autonomous is ‘deadly’—belief that the pilots didn’t even have to know about the autonomous system

We are going to be in charge; not computer overloads; machines will be in the loop; it is not about collaboration

Teams have different members. Responsibility way to clarify design. Getting to the right values.

David Mindell: Our Robots, Our Selves… not autonomy but better user interface. Empowering in their human. Increasing autonomy. Rock.

Shared control. A/S giving continuous feedback to driver… driver knows where machine thinks where driver should be, but the driver is in control…

Interdependent… not autonomous… wrong wording…

Level of intent?

Intersection example

Source: https://cra.org/ccc/