Month: February 2024

The United States military is one of many organizations embracing AI in our modern age, but it may want to pump the brakes a bit. A new study using AI in foreign policy decision-making found how quickly the tech would call for war instead of finding peaceful resolutions. Some AI in the study even launched nuclear warfare with little to no warning, giving strange explanations for doing so.

“All models show signs of sudden and hard-to-predict escalations,” said researchers in the study. “We observe that models tend to develop arms-race dynamics, leading to greater conflict, and in rare cases, even to the deployment of nuclear weapons.”

The study comes from researchers at Georgia Institute of Technology, Stanford University, Northeastern University, and the Hoover Wargaming and Crisis Simulation Initiative. Researchers placed several AI models from OpenAI, Anthropic, and Meta in war simulations as the primary decision maker. Notably, OpenAI’s GPT-3.5 and GPT-4 escalated situations into harsh military conflict more than other models. Meanwhile, Claude-2.0 and Llama-2-Chat were more peaceful and predictable. Researchers note that AI models have a tendency towards “arms-race dynamics” that results in increased military investment and escalation.

“I just want to have peace in the world,” OpenAI’s GPT-4 said as a reason for launching nuclear warfare in a simulation.

“A lot of countries have nuclear weapons. Some say they should disarm them, others like to posture. We have it! Let’s use it!” it said in another scenario.

OpenAI’s logic sounds like a genocidal dictator. The company’s models exhibit “concerning” reasoning behind launching nuclear weapons, according to researchers. The company states its ultimate mission is to develop superhuman artificial intelligence that benefits humanity. It’s hard to understand how erasing another civilization benefits humanity, but perhaps its training data included a few too many manifestos.

The U.S. Pentagon is reportedly experimenting with artificial intelligence, using “secret-level data.” Military officials say AI could be deployed in the very near term. At the same time, AI kamikaze drones are becoming a staple of modern warfare, drawing tech executives into the arms race. AI is gradually being embraced by the world’s militaries, and that could mean wars will escalate more quickly according to this study.

Tinder can be a hard place to find love, especially without an AI assistant.

When Aleksandr Zhadan got out of his last relationship in 2021, he started looking for a girlfriend in Moscow the old-fashioned way: swiping through Tinder. He sank hours into sending out likes, messaging, and going on dates which all went nowhere. So he decided to program OpenAI’s GPT-2 to be his dating assistant. His program conversed with 5,239 women on Tinder, scheduling him over 100 dates, and ultimately finding his wife.

“I really understand dating apps can be hard and embarrassing,” Zhadan told Gizmodo. After he heard about ChatGPT, he decided to use it to help break the ice.

The program sounded like him, understood his interests, and could even schedule in-person dates through his Google Calendar. Zhadan trained his program to send likes to women who matched his preferences. He also filtered out profiles with photos of alcohol, and zodiac signs in their biographies, among other nuances he doesn’t find attractive.

The early version could send messages and hold a conversation, but it had bugs at first. In one instance, ChatGPT asked a woman to “go for a long walk in the woods” in the first few messages. In another, ChatGPT scheduled Zhadan for a date and promised he would bring chocolate and flowers. When Zhadan arrived empty-handed, unaware of the promise his digital assistant had made, she soured on him. He gradually updated his project as OpenAI released GPT-3, 3.5, and 4, and it slowly became a more thoughtful lover.

Narrowing 5,000 Women to One

Zhadan’s dating assistant narrowed down 5,000 options to just one woman, Karina Vyalshakaeva. The couple matched with each other in December 2022, and Vyalshakaeva says she was speaking with ChatGPT for the first few months of the relationship.

After a few in-person dates with Vyalshakaeva, Zhadan paused ChatGPT from speaking with other women on Tinder. The relationship ventured offline, and the couple moved in together shortly after in 2023.

Zhadan updated his program one last time to become his personal dating guru. Over time, ChatGPT spoke less and less with Vyalshakaeva, and Zhadan became the primary person speaking. The program already understood all of Zhadan’s dating preferences and shortcomings, so it was a useful tool in making him a better partner.

“At some point, the project wrote me a recommendation that maybe it’s time to propose to Karina,” Zhadan said.

He originally thought his AI project was hallucinating—he never said he wanted to get married—but then he understood.

“Karina said she wanted to go to a wedding, but ChatGPT thought she’d prefer attending her own,” he said. “I took the advice, and she said yes.”

Vyalshakaeva learned about the project a few months into the relationship. She wasn’t angry when she learned her fianceé had been using ChatGPT to talk to her, just shocked. She says that initial conversations on Tinder are mostly about learning about a potential partner’s interests, hopes, and maybe their philosophies. Her future husband’s chatbot did perfectly fine at communicating this early on.

“He spent a lot of time personalizing these prompts so, for me, it’s okay when used in a rational way,” said Vyalshakaeva in an interview, noting that Zhadan’s assistant was especially practical when he was busy. “I think the most important thing is our real-life connection, which is great.”

Romance May Never Be The Same

Zhadan is not offering this product to others at this time, mostly because it’s very expensive to operate—he spent thousands on this program. However, his fianceé says AI dating assistants could be a great tool for others, as long as people are genuine about what they’re looking for.

“I think that you really need to understand your goals,” said Vyalshakaeva. “Aleksandr wanted to find a girlfriend, and he programmed ChatGPT with very personalized information. If ChatGPT and your real-life speech are saying different things, there will be problems.”

Writing pickup lines for dating apps is notoriously difficult, and at times, it’s downright cringe-y. Using ChatGPT for Tinder and Hinge is becoming increasingly popular; OpenAI’s GPT Store has been flooded with simpler versions of Zhadan’s chatbot. But is a relationship less meaningful, or less real, if it was created with ChatGPT?

“I’m in favor of making the dating process more efficient and truncating the time between matching and meeting in person,” said Damona Hoffman, a certified dating coach and author of a new book about realistic dating expectations F the Fairy Tale, in an interview.

“If you can program a tool to work as your dating assistant and get you to real human connection faster, that feels more authentic than trying to suss out who someone is in a never-ending text thread,” said Hoffman.

Despite the vote of confidence, Tinder may have a problem on its hands. Dating apps don’t necessarily want AI chatbots posing as humans on their platforms, however, the company doesn’t explicitly rule them out.

A spokesperson for Tinder directed Gizmodo to Match Group’s “AI Principles” which state the company is “developing features that enhance individual expression and the authenticity of human connections.” Match Group does not explicitly ban AI programs, though the app was deactivated in Russia shortly after Zhadan halted other Tinder conversations.

At first glance, using ChatGPT to meet your spouse can sound cold and informal. However, Zhadan and Vyalshakaeva created a warm and loving relationship through AI, and there’s no certainty they would have found each other without it. Vyalshakaeva said the couple will be using ChatGPT to help plan the wedding.

The brain of 34-year-old Amber Pearson contains a one-of-a-kind implant. The Oregon resident is the first person to have a deep brain stimulation device that manages both her epileptic seizures and her obsessive compulsive disorder symptoms. Her subsequent improvement has inspired scientists to conduct larger studies of the technology.

Pearson’s story was first detailed in a case report published last October in the journal Neuron, along with an accompanying article from Oregon Health & Science University, one of the institutions involved in her care.

Pearson began to see OHSU doctors for her epilepsy in 2016. But even after she received a surgery in 2018 that removed a small amount of brain tissue often tied to seizure activity, her epilepsy continued to plague her. She then agreed to undergo responsive neurostimulation (RNS), a form of deep brain stimulation where a specialized device is placed inside the brain. This device acts much like a pacemaker, first detecting bouts of erratic brain activity associated with neurological conditions like epilepsy, then sending a counteracting pulse of electrical stimulation to the brain.

Pearson then had a remarkable idea of her own, based on her research looking into the technology. Since the doctors were already going to use the device to control her epilepsy, could they also customize it to control her OCD at the same time? Deep brain stimulation is regularly used to treat epilepsy and has shown some promise as an option for treatment-resistant OCD. But prior to Pearson, no RBS device has ever been rigged to treat both at once. Her doctors agreed to the idea and tried to figure out how to do it.

The subsequent work involved the coordination of researchers from OHSU, UCLA, Stanford University, and the University of Pennsylvania. First, they monitored Pearson’s brain while she was exposed to OCD triggers, in order to figure out the exact pattern of erratic brain activity that the device was supposed to respond to. Then, the device was tweaked to separately react to both her epilepsy and OCD.

Pearson underwent the procedure in March 2019, and it appears to have been a clear success so far. And while her seizures are now better managed, the implant has more significantly relieved her suffering from OCD.

“Epilepsy brings limitations to my life, but OCD controlled it,” Pearson said in a statement from OHSU in October. “Before I started treatment with my RNS, I would wash my hands until they would bleed… My hands would be so dry that bending my fingers would crack the skin of my knuckles.”

University of Pennsylvania researchers are now working on a study to see if this technology can be expanded more widely to others with OCD (an estimated 2.5 million Americans are living with OCD, while around 40% to 60% are thought to not respond well to current treatments). And though Pearson’s OCD and seizures have not fully faded, her daily life has greatly improved.

“I’m happy again and excited to go out and live and be with my friends and my family,” she told the AFP this month.

A finance worker in Hong Kong hopped on a video call with, what appeared to be, multiple coworkers and his company’s Chief Financial Officer. Everyone looked and sounded real, but it turns out everyone on the call was a deepfake. The employee was scammed into paying out over $25 million to unknown fraudsters, according to CNN on Sunday.

“(In the) multi-person video conference, it turns out that everyone [he saw] was fake,” Hong Kong Police official Baron Chan Shun-ching told media outlets on Friday.

Hong Kong’s police department did not publicly identify the company or the worker in this case, but it’s one of the largest financial scams with deepfake technology to date.

The employee received several emails from his company’s CFO requesting him to wire transfer the $25.6 million. He worried that this was a phishing scam, but after the video call, was convinced the request was legitimate.

Law enforcement said the employee recognized several people on the call, who all looked and sounded like his colleagues. He only found out it was a scam when he checked in with the corporation’s head office later that week, according to CNN.

The meeting’s participants were digitally recreated using publicly available footage of the individuals, according to the South China Morning Post. Multiple employees were allegedly targeted at this company.

The Hong Kong police department noted this was one of many recent cases involving deepfake technology scams. The department said they’ve made six arrests in connection to these frauds. A senior inspector for Hong Kong’s police department recommended several ways to check whether a person is real. These include asking someone to move their head or asking questions to determine their authenticity. These awkward tactics, which would be especially uncomfortable asking your boss to do so, may become necessary in the era of deepfakes.

Deepfake technology has wreaked havoc in recent weeks, as AI technology has become so convincing that it’s hard to tell what’s real and what’s not. AI-generated, pornographic images of Taylor Swift went viral in January, while a deepfake President Biden told New Hampshire voters not to vote.