Can AI talk like a human? Experts warn of threat to humanity

 Executives at both Meta and OpenAI have stated that they hope to make generative AI that can truly talk, reason and plan like humans, rather than the kind of AI that is only good at one-time tasks or has limited capabilities. The reason is that when AI has the ability to reason and plan or plan, it can be called "Artificial general intelligence" (AGI). The so-called AGI refers to AI that has the same intelligence as humans or surpasses humans, and can express all intelligent behaviors that normal humans have.

Both OpenAI and Meta stated that they will launch GPT-5 and Llama 3 modules soon, hoping to reach AGI level. Tesla CEO Musk said on the X platform on April 8, “I predict that AI will be smarter than any human being by the end of 2025, and their capabilities may completely surpass all humans in the next five years (before 2029). "

In addition, an account named Jackson on the X platform published a report on the controversial AI Q* (pronounced Q-Star) expected to be released by OpenAI in 2027. The report stated that there is a lot of indirect evidence that Q* may be released in 2027, when it will have reached the level of AGI and have an IQ equivalent to a human of 145.

Q-Star was exposed in 2023. At that time, a report letter from former employees of OpenAI to the board of directors was circulated on the Internet. The letter mentioned that the company's senior management concealed from the entire society and the board of directors that they were conducting AI Q* projects that would threaten human safety.

In September last year, 75-year-old former Google executive Geoffrey Hinton, known as the "Godfather of AI," warned that in the next 10 years, AI technology will be smarter than humans in some aspects, which will bring devastating disaster to mankind.

He predicts that AGI will appear within five years, leading to the emergence of a large number of "killer robots", fake news and unemployment.

Hinton invented technology that became the bedrock of systems like ChatGPT and Google Bard.

On April 5, Adecco Group, one of the world's largest human resources services companies, announced an interview survey of executives from 2,000 large companies around the world. The results show that about 41% of executives believe that the number of employees will decrease in the next five years. The main reason is that AI will replace real people. Some experts even believe that AI may take away 80% of human jobs in the next few years.

In addition, some people engaged in cultural and artistic creation have begun to worry that AI is replacing their wisdom. Recently, the advent of two new music-generating AIs, Suno and Udio, and text-generating image AI, Sora, has made this worry even more real. Rolling Stone writer Brian Hiatt said last month that he was surprised and disturbed by the songs Suno created because they were "completely alien."

In addition, more than 200 famous artists and musicians, including Billie Eilish and Nicki Minaj, recently signed an open letter calling for a ban on "predatory practices" in the music industry. AI" takes away their jobs and motivation for artistic creation. They also called for a ban on the unbridled use of AI or the theft of artists’ voices and likenesses, as this violates the rights of creators and damages the entire music ecosystem.

Zack Kass, a senior director of marketing at OpenAI, boldly predicted in January this year that if it continues to develop, "(AI) may be mankind's last technological invention, and it will replace people's Business, culture, medicine, education and other occupations and professions.”

Japanese computer engineer Kiyohara Jin told The Epoch Times, "The development of AI will make the already competitive creation and work market more intense, which may lead to many humans being eliminated or replaced. This is very important for the development of mankind. It’s a terrible thing. Many people will be unemployed and many problems will be difficult to solve.”

The rapid development of AI is also testing the carrying capacity of power grid systems in many countries, because the development of AI is inseparable from chips, energy and data storage space, and modern technology is particularly dependent on energy and networks. However, power grid systems in various countries are prone to large-scale power outages due to natural disasters or hacker attacks.

Although Musk believed during the interview that the power and chip problems may be solved within one or two years, major chip manufacturers such as TSMC have indeed successively expanded factories in the United States, Japan, Europe and other places, and at the same time invested in more The funds and manpower are needed to solve the chip shortage, but producing more chips means more energy is needed.

When AI performs machine learning, calculations and data storage, it requires huge power support, and currently developing electric vehicles also require huge charging station support. These projects are likely to overload the already fragile and stressed power system, and may even threaten the power people need for normal life.

Alex de Vries, a doctoral student at Vrije Universiteit Amsterdam, used a computer to calculate that the AI ​​field will consume at least 85 to 134TWh (one megawatt hour) of electricity per year by 2027. This figure is equivalent to the annual energy demand of the Netherlands and accounts for 0.5% of global electricity, and the International Energy Agency provides a similar estimate.

Previous reports have estimated that just training a language model such as ChatGPT's GPT-3 is estimated to cost 1,300MWh (millions of watt hours) of electricity, which is approximately equivalent to the electricity consumed by 130 American households per year, or 162.5 people watching Netflix. The amount of electricity required for 10,000 hours. Additionally, some estimate that training GPT-4 to learn might cost 51,772MWh of electricity.

However, these statistics do not take into account the electricity, water and other costs that technology companies spend to maintain servers, GPUs, data storage and AI to answer questions. Some scientists have found that training GPT-3 requires 700,000 liters of clean water, and answering 20 to 50 simple questions requires 500 ml of clean water. OpenAI CEO Sam Altman said that 100 million people ask ChatGPT to answer questions every day. If hundreds of millions of ChatGPT users ask GPT-4 to answer questions, the water consumption will be even more difficult to estimate.

In addition, OpenAI, Meta and Google have also successively stated that they will launch more powerful AI models, which also means that these AIs need to consume more energy, chips and clean water, which may impact the electricity and water sources that humans rely on for survival.

Although two companies, Microsoft and OpenAI, have proposed promoting small nuclear power plants to provide power for AI computing and data storage centers, and at the same time provide part of the power for ordinary households, they have little regard for the location of the nuclear power plant and the nuclear waste generated. Storage, no clear solution was given.