TAKASAKI, Japan – Like nuclear weapons and biotechnology before them, artificial intelligence has brought the world to an existential crisis in which, according to some experts, humanity’s future could be at risk if proper checks are not put in place on a global scale.
Against this backdrop, AI models like ChatGPT were high on the agenda during a two-day Group of Seven Digital and Tech Ministers’ Meeting in Japan that concluded Sunday, with policymakers agreeing on the urgent need for continued discussion on how to govern the rapidly advancing technology.
In its joint declaration, the G-7 agreed to promote “responsible” use of AI while calling for broader stakeholder participation in developing international standards for governance.
ChatGPT, launched in November 2022 as a prototype, stands for Chat Generative Pre-trained Transformer and is trained on massive amounts of data, enabling it to process and simulate human-like conversations with users.
Running on the GPT-3.5 model upon its initial release, the chatbot took the world by storm for the enormous number — said to be 355 billion — of adjustable variables it could use to generate text, a huge leap from previous language models that typically only had a few million parameters.
On March 14, U.S.-based developer OpenAI released the next iteration of the model, known as GPT-4, which is more powerful than its predecessors and has multimodal capabilities — meaning it can take both text and images as prompts.
While the seemingly limitless potential for generative AI has caused concerns that technological development may be getting out of hand, expert opinions are varied on whether it may signal doom for humankind.
In March, the Future of Life Institute, a think tank focusing on the responsible development and use of technology, published an open letter calling for a minimum six-month pause in the training of AI systems more powerful than GPT-4.
How AI could upend the world even more than electricity or the internet
Citing dangers such as perpetuating biases, misinformation, destabilizing labor markets and the concentration of power in a small number of corporations in an attached paper on policy recommendations, the letter collected over 27,000 signatures as of April 30, including from Tesla’s Elon Musk, who co-founded OpenAI, and Apple Inc. co-founder Steve Wozniak.
Advanced AI systems “could themselves pursue goals, either human- or self-assigned, in ways that place negligible value on human rights, human safety, or, in the most harrowing scenarios, human existence,” the think tank wrote.
Musk, who is one of the institute’s backers, also sounded the alarm on hyper-intelligent AI in an interview with Fox News earlier this month, saying it “has the potential of civilization destruction.”
“(If) we only put a regulation after something terrible has happened, it may be too late to actually put the regulations in place. The AI may be in control at that point,” he said.
Katja Grace, co-founder and lead researcher of AI Impacts, a project that focuses on the long-term consequences of sophisticated AI, said she estimates there is a 19 percent chance humanity’s failure to control AI will result in humankind’s extinction.
“I think the biggest risk is that current progress leads soon to AI systems that are as good at making decisions about everything as current AI systems are at making decisions about Go or chess, and are strategizing to bring about objectives that are contrary to human welfare, leading to the destruction of humanity,” she said.
The systems are still far from perfect, however. Inaccurate information presented as fact, dubbed hallucinations, remains a challenge in large language model technology, making them impossible to rely on for critical applications.
Satoshi Kurihara, chairman of the Ethics Committee of the Japanese Society for Artificial Intelligence, said AI currently only exists as a tool for humans, and “it is humans who will destroy humankind.”
“I believe we can avoid extinction if we can learn to coexist with the highly autonomous and versatile AI that will likely become a reality in the future,” he said during a recent written interview.
Kurihara stressed the need to adhere to guidelines like peace, cultural diversity and integrity, as well as controlling the scope and transparency of AI use, during the development of such advanced systems.
In the joint declaration issued this weekend, the G-7 said it recognized “the need to take stock in the near term of the opportunities and challenges” of generative AI and “continue promoting safety and trust” given its global prominence and fast-paced development.
Less apocalyptic but more immediate concerns surrounding generative AI models revolve around their unauthorized collection of user data, ability to manipulate public opinion and potential use for nefarious purposes like deepfakes and revenge pornography.
A repository managed by AI, Algorithmic, and Automation Incidents and Controversies, an initiative that tracks unethical misuse of AI, also logged several incidents of fake news and disinformation during Russia’s invasion of Ukraine.
But Charlie Pownall, founder of AIAAIC, said that while it is important to curb such abuses, doing so “may prove difficult to regulate proportionately without unduly abusing user privacy, confidentiality, and other rights.”
International regulation of AI is further complicated by differing attitudes toward technology around the globe.
Japan’s emphasis on generative AI’s potential utility, for example, means the government has so far taken a more cautious stance toward regulation than the European Union, which has proposed what it describes as the first-ever legal framework on AI.
“It seems unlikely that China, the U.S., EU, the United Kingdom and other major markets will be on the same page on many important aspects of AI legislation given political, economic and legal differences and the wide divergence in public perception and expectations of AI across different countries and cultures,” Pownall said. (Kyodo)
