AI Archives - Tales of Times Forgotten

My Position Regarding Use of Generative AI

Because generative AI is in everything now, a shockingly high number of other blogs have adopted it, and, on Reddit, other people have occasionally accused my comments on that platform of being AI-generated, I’ve decided to make this short post clarifying my own position regarding its use.

I began writing this blog in November 2016, long before the emergence of generative AI. I write every post I make on this blog entirely myself, and it takes me many hours to research, write, and edit each post. I’m sure that, if I did use generative AI, then I could produce many more posts in far less time, but I don’t want to do that. I want every post on this blog to be as accurate as possible and to reflect my own actual thoughts, and I believe that using generative AI would compromise both of these goals. Moreover, I still firmly believe that presenting AI-generated content as one’s own work without disclosure is intellectually dishonest, despite how normalized it has become in the past three years.

ChatGPT Is Impressive for a Bot, But Not for a Human

The academic humanities are facing a new challenge—and, this time, it’s robots. On 30 November 2022, the artificial intelligence research lab OpenAI released as a prototype a chatbot called ChatGPT, which is extremely good at processing language input and producing coherent language output that seems like it could have been written by a human. As a result of this, for nearly the past three months, secondary school teachers and professors in the humanities have been panicking about the possibility that students may use this chatbot to cheat on assignments, noting that it consistently produces more coherent writing than many high school students and undergraduates.

I actually tried out ChatGPT to see what all the hype was about. I’ll admit that, knowing that it is an AI, not a human, I was surprised to find that its English is quite fluent (although it is still not exactly eloquent). Nonetheless, ChatGPT is still far from living up to human standards, at least when it comes to my particular field of classics and ancient history. It bungles translating ancient languages, it frequently makes serious factual errors, and it is incapable of any kind of original thought. When I gave it a prompt to write a historical essay, it completely failed to engage with any primary or secondary sources whatsoever, failed to display even the most basic level of historical analysis, and also made several outright factual errors that I was able to catch.

How Accurate Are “Photorealistic” Portraits of Roman Emperors?

In 2020, the digital artist Daniel Voshart used a neural net called Artbreeder to create a series of “photorealistic” portraits of fifty-four Roman emperors spanning from Augustus (ruled 27 BCE – 14 CE) to Carinus (who died in 285 CE). As the term “photorealistic” suggests, his portraits look almost like photographs. Unfortunately, a lot of people do not realize that these portraits are modern artistic impressions, not scientific recreations of what the Roman emperors really historically looked like.

I’ve seen many people over the past year cite Voshart’s portraits and others like them as though they were authoritative, scientific recreations of what the Roman emperors really looked like. In particular, I’ve noticed a worrying number of white supremacists trying to cite these kinds of portraits as “evidence” that the ancient Romans were all white. In this article I would like to discuss why Voshart’s portraits—and others like them—should be taken with several grains of salt.