ChatGPT Creator OpenAI Setting Up Team to Address AI Risks

The developer of the AI-powered chatbot ChatGPT, OpenAI, recently revealed that the company will be creating a new Superalignment team dedicated to managing security risks that followed the rapid development of the AI industry since last fall. 

In a press release published on OpenAI’s website on Wednesday, the company argued that the “vast power of superintelligence could also be very dangerous and could lead to the disempowerment of humanity or even human extinction.” 

The concept that AI may lead to human extinction, often referred to as the "singularity," has been a topic of debate since the advent of computing.  

The idea gained prominence in the late 20th century with futurists like Vernor Vinge and Ray Kurzweil who predicted that AI could outpace human intelligence, potentially leading to superintelligent machines.  

This concept was further popularized in the 21st century by prominent figures like Elon Musk and Stephen Hawking, who warned of the potential risks if AI development is not properly regulated and controlled. 

Commenting on the control of AI systems, OpenAI explained that it currently relies on human oversight. However, it stressed that people won’t be able to reliably control AI systems significantly smarter than them, meaning the existing approaches won’t efficientlyaddress this problem. 

So, what’s the solution?

OpenAI plans to build a so-called “automated alignment researcher” that will use about 20% of the company’s computing power over the next four years to ensure AI developments don’t get out of hand. 

To set up its first automated alignment researcher, OpenAI said it would: 

1. Develop a scalable training method  
2. Validate the resulting model 
3. Stress test its entire alignment pipeline 

The testing phase will be the key to completing this task, according to OpenAI’s statement. Explaining the procedure, the company noted it would leverage AI systems to assist in the evaluation of other AI systems. 

Former Google X Chief Business Officer Mo Gawda thoroughly explained the potential implications of a crisis that would take place if AI technology isn't regulated in time.

In various interviews, Gawda has spoken about the possibility of AI systems battling each other in virtual space and surpassing human intelligence, adding that current AI systems are already tens of times smarter than the most intelligent living people. 

To counter this possibility, OpenAI added it would automate the search for problematic behavior and interpretability so it can deliberately train misaligned models and confirm that its techniques detect the worst kinds of AI misalignments. 

OpenAI Chief Scientist Ilya Sutskever and the lab’s head of alignment, Jan Leike, will lead these efforts, the release concluded. 

Yesterday’s announcement comes as legislators around the globe consider how to regulate artificial intelligence. OpenAI Chief Executive Sam Altman met with dozens of federal U.S. lawmakers over the past months to discuss this issue, confirming that he would like to see comprehensive regulations. 

AI chatbots are getting more and more popular 

Most tech giants worldwide accelerated the development of their own AI-powered chatbots in recent months following the success of OpenAI’s ChatGPT with approximately 1.8 billion users per month. 

ChatGPT offers free GPT-3.5 and paid GPT-4 models that can generate text, solve math problems, and even code. It includes conversation capabilities, so it can produce text based on any text input, generating emails, essays, letters, and much more.  

It was released last November and because of its massive success, it became the blueprint for several chatbots to enter the scene, including Microsoft’s Bing Chat and Google Bard. 

Unlike ChatGPT, Bing Chat has access to the internet, offering the possibility to link back to sources all while using OpenAI’s most advanced large language model (LLM). It is free to use and runs on GPT-4, the model that OpenAI only provides to subscribed users. 

However, it doesn’t come without downsides. It has a very small query cap, meaning there are significant limits when it comes to creativity. 

Finally, there’s Google Bard. This free chatbot developed by the U.S. tech giant is mostly used as a writing assistant. While Bard is not yet available worldwide, it is free to use and is connected to Google’s search engine. However, it can’t provide sources and it doesn’t have great coding skills either.