Due to the rapid development of artificial intelligence, warnings about the potential risks of artificial intelligence have increased sharply in recent months, according to the latest CAIS statement.
The statement was signed by a number of AI scientists, including Turing Prize winners Jeffrey Hinton and Yoshua Bengio, as well as heads of leading AI laboratories, including Sam Altman, founders of OpenAI - developer of the ChatGPT bot, and Demis Hassabis. ) Demis Hassabis of DeepMind, a subsidiary of Google, as well as a number of senior executives in the company.
The warning is one of many that have been issued in recent months about the potential dangers of artificial intelligence. Some of the early developers of the technology say that we are heading first towards the destruction of humanity, while others warn that regulation is urgently needed.
“There are many scenarios related to the existential risks of AI, but it is difficult to identify the most dangerous scenarios, but I don't do it,” said David Krueger, an expert in artificial intelligence and assistant professor at the University of Cambridge. I think we have enough time to prepare for the upcoming preparation of potential threats.
With that in mind, here are some potential AI issues that are of concern to experts:
1. Artificial intelligence is out of control.
artificial intelligence outside the developers' control; By far one of the most frequently mentioned risks, which experts are urgently warning about.
Artificial general intelligence (AGI) refers to a powerful form of artificial intelligence capable of solving a wide variety of problems and possessing human-like abilities that some believe will, for a short time, surpass humans.
But current AI systems are not sentient, but humanoid. Dr. Janice Wong of the Alan Turing Institute, for example, said that the ChatGPT bot is designed to make users feel like they are talking to another human being.
Until now, experts have disagreed on how to define artificial intelligence. However, it is widely recognized as a threat to humanity and requires research and regulation.
"The clearest example of these risks is military competition between countries," Kruger said. "Military competition in the development of smart weapons (that is, weapons that rely on sensors, computer vision, and software) has been rapid and is now intensifying." The impact of these systems would be significant should they emerge. "out of control."
2. Artificial intelligence takes jobs away from people.
There is a growing awareness that artificial intelligence is a threat to some jobs. “The prospect of unemployment from AI is the most real and imminent existential threat, so we must consider the absurdity that people can feel in the face of mass unemployment,” said Abhishek Gupta, founder of the Montreal Institute of Ethics. Intelligence.
CEOs of major companies are open about their plans to use artificial intelligence: IBM CEO Arvind Krishna announced last month that the company plans to hire new employees in some of its other companies.
On this allegation, Gupta said, "Four or five years ago no one would have made such a statement and taken it seriously."
3. AI bias
The problem of bias in AI systems is the oldest problem the field has faced since its inception, with many studies and studies warning of its potential dangers. For example, a recent study by the Center for Existential Risk Research at the University of Cambridge showed that despite its huge potential, AI also poses risks to privacy, bias, inequality, and security.
There are already many examples of bias in generative AI systems, including an early version of the ChatGPT bot that prompted OpenAI to add more firewalls to help the bot avoid responding to user requests for objectionable content.
In this regard, Gupta said, "If there is an undetected bias in an AI system used for real-world decisions, such as approving benefits, it can have severe consequences."
As Dr. Janice Wong believes that the training data used in the model is often based on English data, and the resources needed to train AI models in different languages are very limited, leaving many people stuck. , or the educational impact of these models. In some languages it is less than in others, which will certainly affect the output.