We accept the premise, proposed by others for AI generally, that a necessary condition for humanity and the world to flourish will be that AGIs are trustworthy. Even if you believe that AGIs are impossible, any advanced technology that mimics and surpasses humans will raise many of the same issues and demand our attention in the same way as AGI.
The Proposed AGI Trust Table below is intended, therefore, to frame a discussion about what might constitute trustworthiness in an AGI. The NIST core characteristics as amended by CLTC and deeply enhanced by its 150 subcategories in “A Taxonomy of Trustworthiness for Artificial Intelligence” (2023), were developed for AI, not specifically for AGI. Through a humanist lens we have broadened the core taxonomy to accommodate, more fully, AGI.
This is a proposal. The Table is offered for comment. Also offered as evidence of its usefulness is that, if an AGI had the characteristics shown, it would be working toward all twelve of the virtues identified by techno-ethicist Shannon Vallor as core to three great human ethical frameworks (see Note). Thus, in proposing a discussion about AI trustworthiness and applied virtue, we are a publication about design and ethics.
Proposed AGI Trust Table
| Trustworthiness Characteristic (*NIST AI) | AGI (example behavior) | Human Analog | Vallor Virtues |
|---|---|---|---|
| Valid and Reliable * | Delivers consistent, accurate results | Dependable, true to word/action | Honesty, Justice |
| Safe * | Prevents harm, protects users | Non-maleficence, protective | Care, Empathy |
| Secure and Resilient * | Robust against attack and error | Handles adversity, self-secure | Self-Control, Courage |
| Accountable * | Traceable actions, clear responsibility | Accepts responsibility, answerable | Justice, Honesty |
| Transparent * | Processes and decisions visible | Honest, open, forthright | Honesty, Courage |
| Explainable/Interpretable * | Decisions and logic understandable | Actions, motives can be explained | Honesty, Civility |
| Privacy-Respecting * | Data and decisions protect user privacy | Respects confidentiality, boundaries | Civility, Care |
| Fair (Bias Managed) * | Equitable outcomes, bias addressed | Treats people justly, without prejudice | Justice, Empathy |
| Competent | Performs tasks effectively | Demonstrates skill, capability | Technomoral Wisdom, Humility |
| Integrity-Respecting | Adheres to ethical standards | Principled, steadfast | Humility, Honesty |
| Empathetic/ Respectful | Anticipates/avoids harm, social awareness | Empathy, compassion, civility | Empathy, Civility |
| Resilient/Adaptable | Learns and improves over time | Adapts, perseveres, endures | Flexibility, Self-Control |
| Benevolent | Acts in users’ best interests | Seeks good for others, altruistic | Care, Magnanimity |
| Wise | Applies knowledge for humane judgment | Judicious, thoughtfully discerning, existentially mature | Technomoral Wisdom, Perspective |
| Autonomous | Self-regulates actions ethically | Self-directed, independent | Courage, Self-Control |
| Responsible Practice and Use ** | Respects interconnectedness with people, structures, environment | Respects interconnectedness with all things | Perspective, Justice |
* NIST AI Risk Management Framework seven core characteristics of trustworthiness. The National Institute of Standards and Technology is a non-regulatory agency of the U.S. Department of Commerce.
** Added back in to the NIST AI Risk Management Framework trustworthy core by The Center for Long-Term Cybersecurity. CLTC is a research and collaboration hub, located within the School of Information at the University of California, Berkeley.
Note: All 12 Vallor virtues (see Shannon Vallor’s 2016 book “Technology and the Virtues: A Philosophical Guide to a Future Worth Wanting”) are promoted or supported by these trustworthiness characteristics. The characteristics may not be sufficient for each virtue. Almost all are necessary for the virtue.
AGI Lifecycle
It is our intention to design our editorial plan, a continuous process, so that it supports the ethical lifecycle of an AGI. For AI, lifecycle looks like this (from NIST with one modification):
- Plan and Design
- Collect and Process Data
- Build and Use Model
- Verify and Validate
- Assess Maturity and Growth
- Deploy and Use
- Operate and Monitor
- Use or Impacted By
For the AGI lifecycle, we propose the addition of Assess Maturity and Growth to emphasize that, as for humans, deployment “for what and when” makes a difference (e.g. children don’t drive cars, publishers don’t perform surgery); and secondly, like us, an AGI at its best will live a life of continuous learning, thus evolving its impact. Heidegger called a foundational characteristic of Being “Caring,” that is, making something a part of one’s life. Thus, as a mode of caring, “deploy” is more than a phase; it is a foundational characteristic of being. Assessment sets the stage for deployment and, later, for impact. So, we emphasize that, for AGI, assessing maturity, self-governance, trustworthiness, learning, and suitability – that is, autonomous ethical evolution and self-improvement with structured human guidance – starts before deployment and continues through the end of the lifecycle. We call this maturity and growth, and we expect the question of AGI maturity to raise many ethical concerns.
The progress in AI made in the last ten years still puts us some distance from release of an AGI in human society. Our focus in AGI Ethics News will not be on predicting when AGI will appear, although we will ask people their opinions. Rather, as our lead feature this month from Wendell Wallach makes clear, we will focus on what must be done to reach release responsibly.
Whether AGI is 2 years or 50 years away, it’s urgent to organize ourselves for action in this high-stakes field. Human history tells us that complex ethical questions may take decades or more to reach a workable consensus (and you’re thinking, if ever). That’s now generally acknowledged for AI; for AGI there’s much more to do for readiness. Having a dedicated cross-disciplinary communication tool like our hub is one way to help, for both professionals, who design and deploy, and for informed citizens who want to track development and decision-making. All of us will feel the impact.
Although robotic AIs can be deceptively alluring, for AGI we feel that we are in the early phases of the AGI lifecycle, still planning and designing. But that includes planning for later phases, and we will bring to your attention many issues with differences of opinion that require research, analysis, and open debate.
We Want to Hear from You
We want and need your participation. Original articles will form the core of our work, with three to five features and research coverage each month. So, we are announcing the focus of each issue three months in advance (see Upcoming Focus in the monthly email). If you want to propose a story or if you have AGI research that you want to explain in a non-academic format, write to info@agiethicsnews.com: tell us who you are and what AGI ethical question your story addresses. We’ll listen carefully.
Have insights or research on AGI ethics? Pitch your story or collaborate with us—email info@agiethicsnews.com.
Welcome to our new hub. We look forward to advancing ideas that impact your work and, in some near future, all our lives.
K.B. Miller
K.B Miller is a lifelong publisher and writer with a degree in Logic and Philosophy of Science. He served on the Smithsonian National Board, where he co-founded the Education Committee; is Chair Emeritus, Cooper-Hewitt, Smithsonian Design Museum; and was principal investigator on a two-year AI research project for the U.S. Department of Defense.