AI can be taught to do almost anything when trained properly with the right dataset. An example of that is what this magician did.
Andrew Mayne is a magician, illusion designer, author, and also a filmmaker. Having authored many book titles, DVDs and manuscripts, he created a project called AI Channels where the AI|Writer is developed.
This AI is able to simulate conversations with famous personalities, both real and fictitious.
It AI uses OpenAI’s text generator API that runs models with weights from the GPT-3 family to understand users' question by searching for patterns in the text.
It then uses the API‘s internal knowledge of the person the user wanted to ask, and guesses how that person would likely respond in written voice.
The digitized characters can answer questions about their work, explain scientific theories, and even offer their personal opinions.
For example, Marie Curie who was a physicist and chemist famously pioneered the research on radioactivity can give some of her answers about radiation, writer H.G. Wells can reveal his inspiration for the science fiction novella The Time Machine, while writer and poet Edgar Allen Poe can give advice on writing dark stories if asked nicely.
https://t.co/TPkHU9NIpa is an experiment using @OpenAI's API to create conversations with virtual historical figures.
Edgar Allen Poe can help you finish a poem. Isaac Newton will explain gravitation. Gottfried Leibniz will tell you how he feels about Newton...— Andrew Mayne (@AndrewMayne) July 4, 2020
According to its website:
"This is intended as a research project to examine the potential applications and considerations for interacting with highly-capable AI."
Users can ask the AI simple and complex questions.
And just like real people, the characters will respond based upon how compelling users' message is to them.
"If the real character was known to be brusque, the virtual version will be as well," the site wrote.
"Characters rarely reply to the same question in the same way. You can ask a question again and will likely get a different response."
What makes AI|Writer also unique is that, it can compare characters in their own eras, with the present day.
"Characters are capable of comparing and contrasting events from their own era and present day using relevant examples," the site said.
For example, when English novelist Jane Austen was asked how her characters would use social media in the 21st century, the AI would reply:
The AI can also impersonate fictitious characters, like what Mayne tweeted on Twitter, showcasing how the AI spoke as the Incredible Hulk:
Using my AI writing project based on @OpenAI's API, I asked this question and got a very sincere reply:
Dear Hulk,
Why Hulk smash?
Best,
Banner
---
Dear Bruce,
Hulk likes to smash. Why? Hulk not know why.
Please help.
Your friend,
Hulk— Andrew Mayne (@AndrewMayne) July 4, 2020
While the AI is indeed impressive due to the fact that it leverages the powerful AI from OpenAI, it does have some weaknesses.
For example, there’s an upper limit of about 300 words to what the AI will respond to.
Mayne also said that while the characters can do well with historical facts, it could be “quite erratic with matters of opinion if there's some debate about the topic” and “rarely reply to the same question in the same way.”
The magician author stressed that AI|Writer "should not be considered a reliable source nor an accurate reflection of their actual views and opinions."
He further stated that the AI is "meant for entertainment purposes only" because it is "merely an experiment to examine interactions with AI."