Greetings,
Two ideas today, one about AI and our degraded public discourse with some detours into the philosophy of mind and yet another (short I promise) thought on LDS finances.
Are You an AI Chatbot?
This will be my attempt to write a think piece on AI. I apologize for adding to the deluge of writing in this genre over the last year or two. The good news is that I promise to make no dire predictions for how AI threatens the future of humanity with replacement. Rather, my thesis is that AI suggests that we are thinking a lot less than we often assume
.
How do we tell if a machine has thoughts? This has been one of the set piece debates in the philosophy of mind for the last half-century or more. It begins with the so-called Turing Test. According to Turing, if in a conversation with a machine I cannot tell if the answers are being authored by a human being or a computer, then we can say that the computer has intelligence or consciousness. The most influential response to the Turing Test was the Chinese Room Argument proposed by John Searle. Imagine a man in a room filled with Chinese characters. He does not speak or read Chinese, but has a complex rulebook that instructs him on what characters to send out of the room when he receives particular Chinese characters from outside. A Chinese speaker on the outside of the room would be able to send notes in Chinese into the room and receive perfectly intelligible replies from the man inside. The Chinese Room passes the Turing Test, but the man inside the room understands nothing of Chinese. Accordingly, Searle argued, to say that a system is intelligent we need something more than the Turing Test. We need some additional process of thought and understanding. This debate has created an entire academic industry of refinements, refutations, and responses.
ChatGPT works by predicting language. It doesn't "think." It just creates language that is a statistically likely set of words associated with the prompt. For simple linguistic tasks this proves entirely sufficient to pass the Turing Test with flying colors. As the task becomes more conceptually complex, ChatGPT's prediction model does worse. Some of my colleagues, for example, fed ChatGPT their exam questions and asked for an answer. They were given well phrased legal gibberish that managed to accidentally make a few valid legal points, basically a D- exam answer. On any issue of any legal complexity, ChatGPT's answers are obviously machine generated, reading like something written by a very confident person with a good legal style manual and zero legal knowledge.
ChatGPT does a pretty good job of producing political arguments and commentary. Give it a story from current events and then invite it to write a progressive or conservative argument about the story modeled on a particular voice, and it will produce an entirely plausible bit of political boilerplate, a well-written version of the kind of stuff that you would read in the comments on a long political Facebook thread or Twitter scrum. It turns out that what passes for political debate is really predictable, with a lot of pre-scripted rhetorical moves. It’s the kind of language that ChatGPT excels at producing. Our political debates pass the Turing Test, and AI can slip into a lot of these conversations without too much difficulty.
The Turing Test and the Chinese Room Argument go to the question of whether or not a machine is thinking. ChatGPT, I think, forces us to ask the question of ourselves. If our language cannot be distinguished from the language of ChatGPT on a topic, are we really thinking? For so much of our political discourse, those doing the talking feel to me like the man inside the Chinese Room. We are not so much thinking, as following the well-established rules of some symbolic game, a minuet of narrative and counter-narrative in which each move can be predicted in advance. The remarkable thing to me is the speed with which these narratives can now be established. Among the very online the progressive or conservative narrative about events is established almost instantaneously and then propagated at scale with astonishing rapidity. The very speed of this process suggests to me that what is going on is less a matter of thought than of simply putting out the words that our side expects. We are ChatGPT.
The most moral-panicky of the AI commentary suggests that in the future, we will all be replaced by AI chatbots. My concern is that for so much of our public discourses, we functionally ARE already AI chatbots. This is depressing, because I side with Searle against Turing. The chatbots aren’t thinking.
Yet Another Thought on LDS Finances
I got a number of thoughtful responses to my last missive on the finances of the Church of Jesus Christ of Latter-day Saints and the choices that the Church hierarchy faces. Exchanges with my friends Dan Nielsen and Whitney Clayton, got me thinking a lot about the complex dilemmas that increased humanitarian spending will create for the Church. In a nutshell, Dan, a political scientist at the University of Texas at Austin who specializes in development, argues that the most effective strategy for the Church to pursue would be targeted donations to other humanitarian groups based on rigorous, evidence-based evaluation of the programs’ actual results. This seems entirely correct to me. If you want to just maximize the good consequences, this is the way to go. One objection to this model of checkbook charity is that Church members could simply write checks on their own, rather than writing checks to the Church that then writes checks to charities. There are two advantages, however, to Latter-day Saints centralizing a lot of giving through the Church. First, the Church’s financial management of its assets creates a potential multiplier effect to donations. By investing donations and then giving away some portion of the appreciation, each donation can become a little perpetual income stream that can be devoted to good works indefinitely. Over time, the overall size of the donation will be larger and given compounding investment and long-time horizons dramatically larger. Second, the Church can acquire evidenced-based expertise more easily than individual donors, meaning that under Dan’s approach Church-based donations would be more efficient. Both advantages accrue from the fact that the fixed costs of financial management and humanitarian evidence assessment would be quite high relative to an individual’s donations, but quite low relative to donations by the Church as a whole. All well and good.
Whitney Clayton, without being aware of Dan’s thoughts, however, provided a very nice counterpoint to them. Checkbook charity is in all likelihood the most efficient way of doing good in the world. However, it would have a very small cultural footprint in the lives of Latter-day Saints. In all likelihood it would have an increasingly large impact at Church headquarters, where managing a Dan-style, evidenced based donations, would be a major facet of the hierarchy’s work. But for ordinary Latter-day Saints, life would look the same. You would go to Church, make tithing donations, and perhaps hear with greater frequency of big donations made by the Church. In a sense, this is what current Church culture already looks like. What is missing would be big and exciting humanitarian projects that the Saints could be directly involved with. In short, the experience of service. In contrast to Dan’s model, one could imagine a missionary program that involved much greater levels of humanitarian service through Church-run and Church-funded institutions. Such a shift would have a big cultural impact within the Church, and would provide an answer to the perennial question of what does life as a Latter-day Saint offer. The problem with Whitney’s vision of Latter-day Saint devotional life pervaded by participation in LDS humanitarian institutions is that such institutions would by definition be built around low-expertise, unspecialized labor. To be sure, the Church could sponsor high-skilled operations like medical clinics. Those would create many service opportunities for LDS nurses and doctors but fewer for 19-year-old missionaries. What does a massive, capital intensive, high-spending, low-skilled labor humanitarian institution look like? It’s really hard to say. I strongly suspect that the reality is that broadly participatory humanitarian institutions run by the Church would likely be expensive and, relative to checkbook charity, highly ineffective.
I don’t know which is the best strategy to pursue in some imagined future of massively increased LDS humanitarian spending. One could imagine a balancing of approaches. Given certain financial futures (a high growth/low inflation scenario that I frankly think is unlikely), it’s even possible that the Church could have the resources to pursue both strategies at scale simultaneously. I don’t know. These are genuinely hard questions.
Until next time,
Nate