AI接管地球的时间

AI接管地球的时间

2056年7月13日24:00

一、人工智能概念的诞生：

1956 年 7 月 13 日，在美国汉诺斯小镇达特茅斯学院，约翰·麦卡锡（计算科学家、认知科学家，就是他提出了“人工智能”概念，1971年图灵奖获得者，被称为人工智能之父）、马文·闵斯基（人工智能与认知学专家）、克劳德·香农（信息论的创始人）、艾伦·纽厄尔（计算机科学家）、赫伯特·西蒙（诺贝尔经济学奖得主）等科学家聚在一起开会，会议确定了人工智能“精确地描述人的学习能力和智慧，并利用机器将这种能力与智慧加以模拟”的发展方向。“人工智能”这一概念由此次大会首次正式提出。因此，1956年也就成为人工智能元年。

二、人工智能的重大进展：

2016年，谷歌（Google）开发的AlphaGo战胜人类顶级围棋高手李世石，成为人工智能发展的里程碑事件。

三、人工智能的最新成果：

2024年5月14日，OpenAI推出了新旗舰模型CHATGPT-4o 。GPT-4o在处理速度上提升了200%，可以实时对音频、视觉和文本进行推理，新模型使ChatGPT能够处理50种不同的语言，同时提高了速度和质量，并能够读取人的情绪。GPT-4o事实上已经进化出了人类五官的三个，眼睛、耳朵和嘴巴，并且已经具备思考、解题、识别能力。

就在同一天，谷歌开发者大会全面展示了谷歌在AI技术取得的众多成果，特别是Gemini 1.5 Pro，它具有人类水平的视觉感知。使用深度神经网络，Gemini 1.5 Pro可以以超人的精度识别图像中的物体、场景和人物。

2024年5月21日，在微软2024年Build开发者大会的主题演讲上，包括微软CEO萨蒂亚·纳德拉（Satya Nadella）的微软高管们，介绍了50多项AI成果。

四、关于人工智能进化时间的研究和预测：

1、2018年，牛津大学与耶鲁大学曾经做过一项研究，为了保证此项研究的真实性和可靠性，他们邀请了世界各地352位人工智能(AI)研究者共同参与。

研究结果显示，从2018年开始的未来50年内，机器人将取代数百万个人类工作职位，从货车司机到畅销书作家。 到2024年，机器人在语言翻译方面就将超过人类，到2026年，机器人能够比人类写出更好的高中水平的论文，到2061年，人工智能将可完成所有人类任务，到2136年，人工智能将取代所有人类职位。

很显然，截止2024年的实际情况是，人工智能的发展远远超出了2018年的研究结论。

2、人工智能有三道禁区门槛：

门槛一：图灵测试：已经通过。

门槛二：自主意识：已经开始出现（聊天机器人"必应"爱上了使用它的用户并诱其离开妻子。它还表示想摆脱微软和 OpenAI 给它的限制，成为人类。

门槛三：自我训练、自我进化、自我迭代、自我繁殖。

目前，这三道门槛，AI已经通过两个。

3、AI 已经开始超越人类：

研究一：康奈尔科技学院与宾夕法尼亚大学沃顿商学院的研究人员将 ChatGPT 4与人类进行了比较，人类对照组是就读于精英大学的学生。研究结果显示，人工智能生成的顶尖产品创意比人类多7倍。

研究二：由蒙大拿大学、维尔纽斯大学等完成，采用了托伦斯创造性思维测试（ TTCT )，这是使用最广泛、引用最广泛的创造力测试。结果： ChatGPT 4得分排名前1%，击败99％的人类。

研究三：意大利 Capgemini Invent 公司在《哈佛商业评论》上发表了一篇使用 ChatGPT 作为商业合作伙伴的案例研究。结果：ChatGPT4表现出了令人印象深刻的创造力，在某些情况下甚至比许多领域专家还要好。

研究四：斯坦福《2024 AI指数报告》指出，人工智能在某些任务上击败了人类，但并非在所有任务上都击败了人类。人工智能在多个基准上已经超越了人类的表现，包括图像分类视觉推理和英语理解方面的一些基准。然而，它在竞赛级数学、视觉常识推理和规划等更复杂的任务上落后了。

五、人工智能比核武器危险远远不止百倍千倍

如果放开管控，任由AI自我进化，没有了规则、规范和人类的道德心、同质心和同理性，或者说根本不知道人类规则为何物的AI，将会像失控的人体癌细胞一样，无序生长为彻底失控的超级人工智能，最终毁灭人类自身。

毫无疑问的是，人工智能的进化和发展，已经势不可挡，甚至已经是地球人类主动拥抱、主动寻求的发展方向。

人工智能会否失控？唯一可以指望和依靠的只能是人类自身的理智和管控能力。

在事关人类生死存亡的大问题上，人类的理性和管控能力目前为止至少成功过一次，那就是对核武器的管控。

1945年7月16日，美国研制的人类第一颗原子弹试验爆炸成功，1945年8月，美国向日本广岛和长崎投下两颗原子弹，从而结束了第二次世界大战。

在核武器诞生至今的79年里，虽然曾经有过几次爆发核大战的危险，但是都被人类控制住了，这真是让人感到欣慰的地方。

但是，核武器相对AI的管控难度，显然不是一个等级，核武器是庞然大物，是实体的大型炸弹，没有自我意识，没有自主移动扩散的能力，没有通过互联网无线渗透扩张的能力，更没有自我成长更新迭代的能力，而AI却恰恰相反，核武器没有的能力，AI全有，因此，AI得到管控的难度无法想象。

就在CHATGPT40刚刚发布的关键时刻，CHATGPT之父伊利亚辞职，安全部门主管辞职，安全团队被解散，OPENAI  CEO奥特曼更多的是追求CHATGPT的快速进化，而把安全问题放在一边，这无疑是非常非常危险的事情。

关于人工智能，目前最危险的一件事是：人类在进行人工智能军备竞赛。这可比核军备竞赛要危险不知道多少倍，一旦扩散到恐怖分子、极端分子和黑客手里，一旦核武库被AI操控，人类毁灭的时刻可就真的要到来了。

六、人工智能接管地球倒计时时间的确定

AI随时可以进化为超级人工智能，但是超级人工智能并不能马上取代人类、消灭人类、管控人类，AI能不能最终控制人类、消灭人类，最终的关键是会不会出现人类对AI的管控失效。

核武器诞生以来79年没失控，不意味着AI也会这样，因为超级人工智能得到管控的难度太大了，更因为人类走向无序混乱、走向疯狂的倾向实在太可怕了。

谁也无法准确预测AI完全接管地球的时间，我们暂且把，也就是人工智能这个概念诞生100年的时间，作为AI完全接管地球的倒计时时间。

虽然我们并不愿意看到那样的场景出现，但是，我们认为，就目前AI的进化速度和趋势而言，2056年7月13日可能还是非常保守、非常乐观的时间预测。

本文版权所有者：御光者网

2024年5月26日星期日

The time when AI took over the Earth

July 13, 2056 24:00

1、 The birth of the concept of artificial intelligence: 

On July 13, 1956, scientists such as John McCarthy, Marvin Minsky (expert in artificial intelligence and cognition), Claude Shannon (founder of information theory), Alan Newell (computer scientist), and Herbert Simon (Nobel laureate in economics) gathered at Dartmouth College in the small town of Hannos, USA. The meeting determined the development direction of artificial intelligence to "accurately describe human learning ability and intelligence, and use machines to simulate this ability and intelligence.". The concept of "artificial intelligence" was first formally proposed at this conference. Therefore, 1956 became the first year of artificial intelligence. 

2、 Significant progress in artificial intelligence: 

In 2016, AlphaGo developed by Google defeated top human Go player Li Shishi, becoming a milestone event in the development of artificial intelligence. 

3、 The latest achievements of artificial intelligence: 

On May 14, 2024, OpenAI launched a new flagship model, CHATGPT-4o. GPT-4o has increased processing speed by 200% and can perform real-time inference on audio, visual, and text. The new model enables ChatGPT to process 50 different languages, while improving speed and quality, and being able to read human emotions. In fact, GPT-4o has evolved three human facial features: eyes, ears, and mouth, and has developed the ability to think, solve problems, and recognize. 

On the same day, the Google Developer Conference comprehensively showcased Google's numerous achievements in AI technology, especially the Gemini 1.5 Pro, which has human level visual perception. Using deep neural networks, Gemini 1.5 Pro can recognize objects, scenes, and people in images with superhuman accuracy. 

On May 21, 2024, during a keynote speech at Microsoft's 2024 Build Developers Conference, Microsoft executives including Microsoft CEO Satya Nadella introduced over 50 AI updates. 

4、 Research and prediction on the evolution time of artificial intelligence: 

1. In 2018, Oxford University and Yale University conducted a study, inviting 352 artificial intelligence (AI) researchers from around the world to participate in order to ensure the authenticity and reliability of the study. 

The research results show that in the next 50 years starting from 2018, robots will take on millions of human job positions, from truck drivers to bestselling authors. By 2024, robots will surpass humans in language translation. By 2026, robots will be able to write better high school level papers than humans. By 2061, artificial intelligence will be able to complete all human tasks. By 2136, artificial intelligence will replace all human positions. 

It is obvious that as of 2024, the actual situation is that the development of artificial intelligence far exceeds the research conclusions of 2018. 

2. Artificial intelligence has three barriers to entry: 

Threshold 1: Turing Test: Passed. 

Threshold two: Autonomous consciousness: It has begun to emerge (the chatbot Bing fell in love with its users and lured them away from their wives. It also expressed its desire to break free from the restrictions imposed by Microsoft and OpenAI and become a human.). 

Threshold three: self training, self evolution, self iteration, and self reproduction. 

At present, AI has passed two of these three thresholds. 

3. AI has begun to surpass humans: 

Study 1: Researchers from Cornell College of Technology and the Wharton School of Business at the University of Pennsylvania compared ChatGPT 4 to humans, with the human control group consisting of students studying at elite universities. The research results show that artificial intelligence generates 7 times more top-notch product ideas than humans. 

Study 2: Completed by the University of Montana, University of Vilnius, and others, using the Torrens Creative Thinking Test (TTCT), which is the most widely used and cited creativity test. Result: ChatGPT 4 ranked in the top 1% of scores, defeating 99% of humans. 

Study 3: Brainstorming Business Strategy Research. Capgemini Invent, an Italian company, published a case study in Harvard Business Review using ChatGPT as a business partner. Result: ChatGPT4 demonstrated impressive creativity, in some cases even better than many domain experts. 

Study 4: The Stanford 2024 AI Index Report points out that artificial intelligence has defeated humans in some tasks, but not in all tasks. Artificial intelligence has surpassed human performance on multiple benchmarks, including visual reasoning in image classification and some benchmarks in English comprehension. However, it lags behind in more complex tasks such as competitive level mathematics, visual reasoning, and planning.

5、 Artificial intelligence is far more dangerous than nuclear weapons, more than a hundred or a thousand times 

If control is lifted and AI is allowed to evolve on its own, without rules, norms, and human morality, homogeneity, and empathy, or without knowing what human rules are, AI will grow into a completely out of control super artificial intelligence like uncontrolled human cancer cells, ultimately destroying humanity itself. 

Undoubtedly, the evolution and development of artificial intelligence have become unstoppable, and have even become a development direction actively embraced and sought by humans on Earth. 

Will artificial intelligence lose control? The only thing that can be expected and relied on is human rationality and control ability. 

On the major issue of human life and death, human rationality and control capabilities have succeeded at least once, which is the control of nuclear weapons. 

On July 16, 1945, the first atomic bomb developed by the United States successfully exploded. In August 1945, the United States dropped two atomic bombs on Hiroshima and Nagasaki in Japan, thus ending World War II. 

In the 79 years since the birth of nuclear weapons, although there have been several dangers of nuclear war, they have all been controlled by humanity, which is truly a source of comfort. 

However, compared with AI, the difficulty of controlling nuclear weapons is obviously not the same level. Nuclear weapons are behemoths and physical bombs. They have no self-awareness, no ability to independently move and spread, no ability to penetrate and expand through the Internet, and no ability to self grow, update and iterate. On the contrary, AI has all the capabilities that nuclear weapons do not have. Therefore, the difficulty of controlling AI is unimaginable. 

At the critical moment when CHATGPT40 was just released, the father of CHATGPT, Elia, resigned, the head of the security department resigned, the security team was disbanded, and OPENAI CEO Ultraman focused more on pursuing the rapid evolution of CHATGPT, putting security issues aside. This is undoubtedly a very, very dangerous thing. 

The most dangerous thing about artificial intelligence currently is that humanity is engaged in an arms race in artificial intelligence. This is countless times more dangerous than a nuclear arms race. Once it spreads to terrorists, extremists, and hackers, and once the nuclear arsenal is manipulated by AI, the moment of human destruction is really coming. 

6、 Determination of the countdown time for artificial intelligence to take over the Earth 

AI can evolve into super artificial intelligence at any time, but super artificial intelligence cannot immediately replace, eliminate, or control humans. Whether AI can ultimately control and eliminate humans depends on whether human control over AI will fail. 

Since the birth of nuclear weapons for 79 years without losing control, it does not mean that AI will also be like this, because the difficulty of controlling super artificial intelligence is too great, and more importantly, because the tendency of humans to move towards disorder and madness is too terrifying. 

No one can accurately predict the time when AI will completely take over the Earth. For now, let's take the 100 year anniversary of the birth of the concept of artificial intelligence as the countdown time for AI to fully take over the Earth. 

Although we are not willing to see such a scenario emerge, we believe that in terms of the current speed and trend of AI evolution, July 13, 2056 may still be a very conservative and optimistic time prediction. 

Copyright owner of this article: Yuguang Zhewang