The Power of Obligation in the Age of AI

As soon as I saw your title I figured there was a connection to other aspects of consciousness.  Glad it was interesting to you.  You bring up a very good question about AIs and regret. Considering alternative plans for the future is part of what the AI people call optimization. I have never seen them discuss regret.  Logically, regret would start with alternative *plans about the past* (“could have done differently”).  It would then *motivate* possible future plans and actions that, some of which, taken together, constitute obligations.  

\
The analogy to this in AI training is reinforcement learning with negative feedback.  The analogy is limited, as all analogies are.  In this case, machine reinforcement learning doesn’t have the social dimension that human obligations do.  I’m not aware of any attempts to add social obligations to an AI, maybe for good reason.  Like anything with an AI, the challenge is how to limit how far it will go in pursuing whatever goals it has.  Like, you don’t want an AI that is loyal/obligated to particular humans to harm other humans when fulfilling its obligations.  Putting boundaries on the behavior of a powerful AI is an unsolved, possibly un-solvable problem: <https://app.hackernoon.com/stats/ais-will-be-dangerous-because-unbounded-optimizing-power-leads-to-existential-risk >

\
Also, regret is an emotion.  Emotional AI seems dangerous to me.

The closest thing (from bonafide consciousness researchers) to this idea that I’ve seen is a theory that regret and responsibility are functions of having consciousness.  And consciousness thus evolved because it made advanced sociality work better by enforcing committments/obligations.  My phrasing of this is poor, but the article from Chris Frith and Thomas Metzinger is not: <https://www.researchgate.net/publication/304657860_What's_the_use_of_consciousness>

Artificial Persuasion Takes Over the World

Monica - There’s a very active and super smart AI Alignment community, but their stuff is not well known because … well, it’s hard to understand.  I think one way to help their ideas reach the public is to write stories like mine where abstract possibilities are turned into imagined realities.  Even so, failure of alignment (aka the end of our world as we know it) will not be a simple process.  My story reflects this complexity, but I hope that some of your readers will enjoy the challenge.

\
\


More at Sentient Artifact!

Nominated for 2022 - HackerNoon Contributor of the Year - Artificial Intelligence

How AI and the Internet Can Create An Immortal Persona

AIs Will Be Dangerous Because Unbounded Optimizing Power Leads to Existential Risk

Why Would Anything Ever Go Right?

What We Should Learn from the Tension Between Mind and Machine

Intelligence Explosion by AI and Its Impact on Superintelligence 

The AI Who Was Not a God

Time and the Artificial Ego

Making Sense of an Interview with AI [Transcript]

A Farm-Born AI Tells the Story of its Stages of Conscious Development

Coinbase #1 for All the Wrong Reasons, Apple Becomes 'Beta'

The Challenging Quest for Authentic Experiences 

A Catalog of Eternal Moments 

Microsoft’s Minotaur Moment

Ted Wade