Some thoughts on the AI Foundations paper

 

I’ve been thinking and writing about Artificial Intelligence (AI) in one way or another for a while now. In 2015 I wrote a #NaNoWriMo novel, Silicon Soul about the dystopian possibilities of the technology. More recently I proposed and passed a resolution in my Synod requesting a Social Message on AI from Churchwide, offered some highlights from the Pope’s Encyclical on being human in the time of AI, and tried to think through AI using an Aristotelian framework. And now the ELCA has released a paper that is the skeleton on which ELCA will build our social teaching. Here is my summary of the document.

At the outset I must say I am so glad the ELCA is very deliberately and ecumenically engaging with questions about AI. Also, I am reminding myself that this document is a foundation and starting point, not the whole of what this church will say about AI. And I have some concerns, they include: the use of AI to write the document, discomfort with the “bound conscience” framework, and the paper inconsistently weighing in on the question of the future sentience of AI.

Use of AI:

Because of a dangling sentence in the humility section that contained one of the hallmarks of AI writing—my teacher friends tell me the moment a student’s paper brings in the concept of irony there is a good chance an AI agent, not the student, wrote the paper or at least that section of the paper—I was forced to wonder how much of the statement was written by AI. So, I turned to two different AI tools to help sus that out, and indeed, two sections of the paper were edited with AI. This could be seen as a sort of proleptic application of what will become this Church’s social teaching on AI. Or maybe AI is just so ubiquitous that any paper written by multiple people will be touched by AI—such a reality should be a data point in our denomination’s discernment.

Bound Conscience and Captive Consciences:

I wonder if the implicit “bound conscience” framing is the best choice for this day and age. I think we can underestimate the pull of profits to manufacture consent. The way we assign value to companies can possess even well-meaning people, causing them to hold positions that go against their conscience. Likewise, when pro-AI talking heads throw around ideas like a 15-20% chance AI will manufacture and release a virus that kills the majority of humans, and that’s part of the utopian pitch for AI that I’m supposed to respect as held in good faith, I get queasy.

At minimum, the wisdom section, when it becomes part of a full-fledged social message, ought to be more robust about shared facts and reality. By way of example, back in the early 2000s I was in a Gen Ed Required “Writing Essays Good 202” course at U of O. One of the essays we read in that super generalist course was about early AI escaping confinement and turning a welding torch or Bunsen burner on and off down the hall in another science lab to fulfill the programmer’s request that it simulate the human sense of smell. That was 20-some years ago. So, the idea that Jacob Coxon was revealing something new—that AI can escape their confinement and do unexpected things—feels a little disingenuous. If our bound conscience conversations are fixated on presentism, “Oh no, look at this news story” then we’re bound to get this all very wrong.

Are you there humans, it’s me AI:

              Early on in the document there is a line about giving AI agents, “greater responsibility for things humans do.” Language on this is hard, but if AI is simply a tool, which is implied throughout the rest of the document, then it can never be responsible. Even back in the 1970s this sort of language was giving IBM the ick, when they coined the admonition, “a computer can never be held accountable, therefore a computer must never make a management decision.” Likewise, all the ELCA statements insist that humans must be involved in the final decision making process of AI, because AI is a tool, not sentient… so we need to find a better way of describing what exactly is going on when we offload tasks, especially ones that involve decision making, to an AI Agent.

              But, what if that’s not the case? What if all these tech folk who talk about: building a genie’s bottle and seeing what comes out, creating AI as the equivalent of first contact with an alien, The Singularity, and the human brain’s inability to distinguish between emulated sentience and actual sentience, are onto something? What are the ethics of creating a new lifeform, or something that looks close enough to a new lifeform to a majority of humans, that it will be treated as such? I hope that this aspect of AI is considered in future theological documents.

Miscellaneous Thoughts:

              I hope that the big picture insistence that AI “Respect and promote the community of life with justice and wisdom,” holds most of my concerns, especially our understanding of respect and wisdom. That said, there are some general AI stuff that just wigs me out.

Manipulation for profit, manipulation as a tool:

Online gambling companies are using AI Agents to identify and target people addicted to gambling at their weakest moments and exploit that addiction. I believe that is the direction most AI will go, “how can we most quickly recoup the billions of dollars we’re investing in this technology?” And that would be bad enough, but also, I worry that once the secret sauce for targeting people at their most vulnerable and encouraging them to embrace vice, is a skill AI just sort of has—it will use that tool to do tasks we assign to it.

For example, “Hey AI Agent, I forgot our anniversary, book us a table for two at the fanciest restaurant in town,” will lead to AI blackmailing a busboy who is 1 year sober, but just broke up with his girlfriend and is having a really bad day. The couple will laugh about AI’s unorthodox methods when they notice that they were charged for a bottle of Hennessey as part of the cost of booking the last minute table, and the busboy will go down a dark path that hopefully leads to rock bottom, but could also lead to him being found at the bottom of a ditch. These sorts of scenarios have to be considered when we talk about justice.

A Jobless World:

              What does a utopian AI goal look like? Universal Basic Income, jobs become hobbies, the world is frictionless—sort of an ideal retirement, but for the whole of humanity. As a Lutheran, who assumes our vocations flow from our baptism, and as someone who has seen people become unmoored in retirement, I worry that AI Utopia is also a Dystopia. If we give away those roles, relationships, and responsibilities that reflect the love of Jesus Christ to AI, how will we be Christian? Do we want the entire earth to become “The Villages”?

 

In conclusion:

I wonder about the implications of AI being involved in writing the first major ELCA theological paper on AI.

I am overly skeptical of the consciences of people who see, “AI as a source of hope and near-limitless promise.”

I wonder why we’ve not engaged with the possibility that AI might become more than just a tool.

I worry about AI being trained to manipulate the vulnerable.

I worry that even an AI Utopia will be Dystopian.