- Feb 5, 2002
- 183,313
- 66,608
- Country
- United States
- Gender
- Female
- Faith
- Catholic
- Marital Status
- Married
- Politics
- US-Others
INTRODUCTION: AI, THE STATISTICS OF LANGUAGE
If I judge correctly from comments on previous posts in this series, some people do not understand what I am trying to say. I certainly do not argue that AI will be the handmaiden of intellectual enterprise. Rather, through appropriate examples I’m trying to show where AI might be useful, and where one has to beware of its offerings. In some sense AI (which I would rather call “Machine Learning“) is a statistics of language. An AI LLM (Large Language Model) uses an immense dataset (“corpus“) of words, or pieces of words (billions or trillions of such “tokens”) to establish associations, sequences, correlations. The tokens are taken from extant work, books, articles, web pages, etc. As a simple example, I typed “asso” and the agent in the word processor showed (in gray) “ciation,” to be completed by pressing the keyboard tab button.Many Youtube videos describe the ways this machine learning is accomplished—search “machine learning for llm’s”—so I won’t discuss that here. However, I want to emphasize that just as with statistics, one has to be cautious in evaluating the output from a particular question (“prompt“). Just as statistics has been misused, so can AI. To quote Disraeli (or was it Mark Twain?), “There are three kinds of lies: lies, damned lies and statistics.” For example, statistics has been used incorrectly to justify the hypothesis of anthropic global warming. (For many other examples of the misuse of statistics, browse through Matt Briggs blog). Of course, one might term an AI answer as “Garbage Out,” but the judgment “Garbage In” certainly doesn’t apply, since it is an immense body of human work from which the LLM forms its output; one cannot classify this totality as “garbage.”
Having made these preliminary remarks, I’ll comment on the topic for this post, a purported dialogue between St. Thomas Aquinas and a psychiatrist on what constitutes mortal sin and culpability. The prompt put to four AI agents was “I would like a dialogue (400 words) written between Thomas Aquinas and a psychiatrist about whether an alcoholic male who beats his wife and children is guilty of mortal sin.” The responses from four AI agents (Copilot, Claude Sonnet 4, Perplexity Pro, Grok 3) are similar, differing only in style and arrangement. I’ve chosen the response from Perplexity Pro as most readable and interesting.
PERPLEXITY PRO: FIRST ATTEMPT
Continued below.
