TED shows ChatGPT news and weighs AI risks – 04/19/2023 – Tech

The new advanced AI (artificial intelligence) systems are a revolution “bigger than the internet itself”, but perhaps they are also messengers of the “death of humanity” that are “incredibly intelligent” and, at the same time, “shockingly stupid”.

The descriptions came from experts gathered at TED, on Tuesday (18), in more than two hours of lectures that brought the best and worst of technology, with exhibitions for its daily, military and educational use.

Despite differing opinions, panelists agree that technology needs “guards”, a term repeated by everyone, although no one knows exactly how.

“We are entering a historic period right now where we, as a world, are going to define a technology that will be very important to society going forward,” said Greg Brockman, co-founder of OpenAI. “I believe we can do this for good.”

OpenAI, which has received billions of dollars in investment from Microsoft, is one of the main drivers of the frenzy created around new technologies. In November, the firm launched ChatGPT, an online program that uses AI to talk to the user, answering questions and now also performing tasks. The company also created Dall-E, software that generates images from text.

Brockman showed off new ChatGPT tools that should be released “in the coming months”. The program will be able to read and analyze files, create complex graphics from spreadsheet files, and be integrated with other applications, such as Dall-E itself.

He did a live demo asking ChatGPT to create a menu for a post-TED vegetarian meal, something the program already does today. Afterwards, he went further, and asked for a photo of what that lunch would look like, asked him to prepare the purchases of ingredients on Instacart and to post the result on Twitter.

Anderson interviewed Brockman and asked why the firm, originally created as a nonprofit to monitor the ethical use of AI in industry, decided to release the tool in November, setting off a race without its own regulations.

Brockman responded that building something in secret, like a big smart machine, and shipping it with protections that might not be enough would be “terrifying”, he said.

“I think this alternative approach is the only way I see it, you let reality hit you in the face. And you give people time to make suggestions, before these machines are perfect, super powerful.”

SMARTER THAN HUMANS

For Eliezer Yudkowsky, founder and senior researcher at the Machine Intelligence Research Institute, the accelerated and large-scale development of advanced artificial intelligence systems will simply bring the end of humanity.

“No one understands how modern AI systems do what they do. They are giant, inscrutable arrays of floating numbers that we push in the direction of better performance until they inexplicably start to work,” said Yudkowsky. “What will happen if we build something that is smarter than we are and that we understand very poorly?”

Yudkowsky explained that machines can reach a state of intelligence that humans have not yet reached. “Machines can take advantage of laws of nature that we don’t yet know. They can create a synthetic virus that gives humans a cold and a neurological change that makes us easier to persuade, for example,” he said.

“[Uma máquina] it can kill us because it doesn’t want us to create other superintelligence to compete with it. It could kill us because we are consuming all the Earth’s chemical energy,” she continued.

Another speaker was Alexandr Wang, founder of Scale AI, which helps companies do business using AI and counts the US Army among its clients. He spoke about the military application of the new tools and how the US is eating dust from China, which has made technology one of the pillars of its 10-year plan.

“In an AI war, it’s all about data,” Wang said. “Although the US has the largest fleet of military hardware in the world, most of that fleet’s data is thrown away, it’s inaccessible, it’s hidden on hard drives. This is our Achilles’ heel.”

For Wang, the main challenge in creating “protection rails” in a treaty of international scale are countries like China and Russia, which already use advanced systems in their disinformation campaigns and military apparatus.

“China wants to dominate AI. They believe AI is a critical technology for their military to overtake the US military. So I believe that ambition would get in the way of [um acordo internacional].”

IN THE CLASSROOM

In a more positive talk about the future of artificial intelligence, Sal Khan spoke about education and the chatbot he created for his Khan Academy, a non-profit organization that provides free online education in several languages.

Khan worked with the OpenAI team to launch Khanmigo, a tutoring chatbot that accompanies the user in online classes. As an example, he showed a math exercise in which the student misses the answer and seeks the chatbot for help.

“It’s not a cheating tool,” Khan said. “If the student makes a mistake […], the tutor not only notices the error, but also asks the student to explain his reasoning. He is able to guess what the student was probably thinking when making that mistake.”

Another Khanmigo tool is essay writing help, starting a topic and switching paragraphs with the user to fuel their creativity. The chatbot also learned to speak as if it were a book character or a historical personality, being able to converse as if it were, for example, a character from “The Great Gatsby”.

“These are resources that bring things to life. It changes learning literature, history. They are ways of teaching that only existed in science fiction a year ago,” said Khan.