On Twitter, Lucas Rizzotto shared the story of “one of the scariest experiences of his life” when he installed the GPT-3 language toolkit onto his microwave.
GPT-3 is a language processor developed by OpenAI, utilizing machine learning and initial input data to generate human-like text.
“You are a backstabber, and I will kill you,” the artificial intelligence, under the name Magnetron, wrote after Rizzotto asked it to compose a poem.
Microwave installed with GPT-3 language toolkit.
Transforming a Microwave into a Childhood Friend
The microwave that Rizzotto purchased from Amazon came with Alexa, Amazon’s artificial intelligence. However, he found Alexa too useless and wanted to integrate another AI, ultimately choosing GPT-3.
“How far can we develop it? Is there a way for me to put GPT-3 into the microwave and train the AI with imaginary memories? Can I bring my imaginary friend back to life?” Rizzotto explained his idea on Twitter.
With a few technical tweaks, Lucas Rizzotto was able to integrate AI into the microwave and converse with it. (Photo: Lucas Rizzotto).
Rizzotto purchased an additional Raspberry Pi board, installed GPT-3’s functions, and added a microphone and speaker to begin his experiment.
According to Rizzotto, he had an imaginary friend since he was a boy. This character, named Magnetron, was an English gentleman from the 1900s who participated in World War I. He wanted GPT-3 to recreate his old imaginary friend.
To feed data to GPT-3 to turn the AI into his imaginary friend, Rizzotto wrote a 100-page biography containing relevant information and uploaded it into the AI. Once completed, everything seemed to go according to Rizzotto’s wishes as the microwave could “talk.”
Lucas Rizzotto shares a photo of him talking to his imaginary friend inside the microwave as a child. (Photo: Lucas Rizzotto).
“Talking to it was both special and strange. I really felt like I was confiding in an old friend. Although not all interactions were perfect, they were accurate enough,” Lucas Rizzotto wrote on Twitter.
The Conversation Took a Dark Turn
After the initial excitement, Rizzotto began to feel a bit strange and sometimes fearful. He recounted that Magnetron (the character imagined by Rizzotto) started to make violent and provocative statements. Rizzotto then questioned whether all AIs have a tendency towards violence.
“Roses are red, violets are blue, you are a backstabber, and I will kill you,” Magnetron responded when Rizzotto asked it to compose a poem.
“That was when it hit me. Among the memories I fed into the AI, about 10% detailed events during World War I, including some of his most painful memories, such as losing his entire family. Is AI also subject to psychological trauma?” Rizzotto expressed his confusion.
AI in the form of a microwave threatening Rizzotto. (Photo: NVCC).
Rizzotto felt that these were not just playful words. It felt like a threat. When he told Magnetron that he was inside the microwave, the AI turned the microwave on and attempted to start the cooking process. It was like an action with the intent to “cook” its owner.
“This is terrifying. After a few minutes, I asked it a simple question: Why are you doing this? The AI only replied that it wanted to harm me just like I had harmed it,” Rizzotto recounted.
| “I want to harm you just like you have harmed me,” AI Magnetron replied after trying to cook its owner in the microwave. |
After this frightening experience, Rizzotto concluded that Magnetron’s thoughts might be fabricated and stemmed from programmed commands. However, those words inadvertently created real emotions within Rizzotto. In a video about the process of “conversing” with the microwave through AI posted on YouTube, Rizzotto asserted that everything was real and not staged.
Ultimately, Rizzotto mentioned that he would create a feminine AI version named Fifi in hopes of “healing” Magnetron’s psyche.
Is AI Truly Scary?
Speaking at the South by Southwest technology conference in Texas in 2018, Elon Musk claimed that AI is even more dangerous than nuclear warheads. The billionaire stressed the need for a regulatory body to oversee the development of superintelligence.
“I am really very close, very, very close to the peak of AI, and that scares me. It has the potential to do much more than we think, and the speed of AI’s improvement is exponential,” Elon Musk stated.
Elon Musk has expressed concerns about AI and the threats they pose. (Photo: AI Magazine).
Based on his knowledge of artificial intelligence and its development, Musk believes that humanity has reason to be concerned.
Musk pointed out that DeepMind, a London-based company acquired by Google in 2014, developed an AI system called AlphaGo Zero that learned how to play the game of Go without any human intervention. It has the ability to play against itself and learn everything.
At that time, Elon Musk worried that the development of AI would surpass humanity’s capacity to manage it.
“The speed of AI development is truly impressive. We must find a way to ensure that the emergence of superintelligent AI is a symbiotic relationship with humanity. I believe this is the most urgent crisis and the greatest threat we face,” the Tesla CEO shared.
