Raj Reddy: AGI, ASI and AI Bill of Rights, Talk at ICAIS in Beijing Oct 19, 2026
Publicado no IA Dots ·
AGIAI SignalASIdot-agidot-artificial_intelligencedot-asidot-google_deepmindidioma-enInglêsyt-mFfhfChhU6A
Vídeo em Inglês, publicado pelo canal CMU Robotics Institute.
About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.
Video summary
The speaker discusses the role of artificial intelligence in scientific discovery, tracing its history from early foundational models by Herb Simon to modern achievements like AlphaFold. The presentation also explores the definitions and implications of Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI), proposing the concept of an AI "Guardian Angel" to protect humanity from disasters.
Key points
- Herb Simon and his students studied how AI can assist in scientific discovery, showing that creativity involves systematic symbolic search.
- AlphaFold, developed by Google DeepMind, successfully mapped protein structures and earned its creators a Nobel Prize in chemistry.
- AGI is described as the ability to answer questions on any topic correctly at least 90% of the time, demonstrated by passing Advanced Placement tests.
- Artificial Superintelligence (ASI) is defined as an intelligence with god-like qualities of omniscience, omnipresence, and omnipotence.
- The speaker proposes a "Guardian Angel" concept enabled by ASI to protect humanity from unexpected disasters by providing the right information at the right time.
Topics
Artificial IntelligenceScientific DiscoveryAlphaFoldArtificial General IntelligenceArtificial SuperintelligenceGuardian AngelAI Bill of Rights
Video transcript
Transcript from available captions; it may contain errors.
Dr. Liu, distinguished guests and friends, it is a pleasure to join you today virtually for the international conference on AI scientists in Beijing. Since the beginning, it was clear that AI can play a significant role in scientific discovery. Herb Simon, one of the founders of AI, and his students were studying how scientific discovery can be assisted using AI. Their explorations have been summarized in a book by MIT Press called Scientific Discovery, Computational Explorations of Creative Processes. I recommend you this book. Simon suggested that scientific discovery is a discovery of invariance and correlations in the raw data, much like it is happening today within AI foundation models using deep learning and LLMs. They suggested that discovery and invention is not some mysterious creative genius, but rather the use of tools such as induction, abstraction, memory retrieval that people use in day-to-day problem-solving and solving puzzles. Several PhD students of Herb Simon have produced systematic demonstration of this hypothesis by rediscovering Kepler's laws, Boyle's law, Ohm's law entirely from the scratch using the same data that was available at the time of discovery. Their work demonstrated that creativity is a result of systematic incremental symbolic search that can be simulated by a computer. More recently, we've had a dramatic demonstration of the use of AI in scientific discovery. Working at Google DeepMind, Hassabis and Jumper developed AlphaFold, an AI model trained on massive databases of known protein sequences and physical structures resulting in mapping of the structure of virtually all known diseases. Hassabis and Jumper and Baker received a Nobel Prize in chemistry in 2024 for their work. This proves that Nobel Prize quality research can be done using AI assistance in the future. Next, I would like to discuss the implications of artificial general intelligence, AGI, and artificial superintelligence, ASI, and their potential use for scientific discovery in the future. We do not now have a commonly accepted definition of these two terms, AGI and ASI. Expert systems of the '80s and '90s were built using narrow domain-specific knowledge. The term general intelligence was coined to contrast the expectation from the conventional expert systems to include human-level common-sense knowledge and to distinguish it from special intelligence embodied in expert systems. The term AGI usually means that you can ask a question on any topic and expect to get a correct answer at least 90% of the time. In the past, such people who could answer questions on any topic were called polymaths. Since the advent of ChatGPT, we have been able to demonstrate systems that demonstrate AGI on almost all topics. And one way to demonstrate and quantitatively measure the performance is the use of AP Advanced Placement Tests of senior high school students. On a specific benchmark, GPT-4 was able to answer correctly all the science and math questions over 90% of the time and were were able to take all the other AP tests and get a passing grade. Any human being that can exhibit such capability to take all AP tests would be considered a genius or more correctly a polymath. I believe we have now demonstrated AGI that satisfies such a definition. Now we come to artificial superintelligence. What is ASI? An ASI is a an intelligence that's almost godlike, omniscient omnipresent and omnipotent. Knows everything, sees everything, and controls everything. Superintelligence. And can be anywhere, and can control events. Let me illustrate such a capability by considering a specific example. You may remember the Indian Ocean tsunami of 2004. In that disaster, over 300,000 people lost their lives. My colleague, Professor Carbonell and I were contemplating whether there was any way we could have saved their lives using AI. We came up with the term Guardian Angel for such intelligent agents that can help humanity from unexpected disasters. A Guardian Angel has to be able to provide the right information to the right people at the right time in the right language at the right level of detail. We propose that such a capability be called the AI Bill of Rights. A little reflection on the AI Bill of Rights will immediately show that we need to endow a guardian angel with the capabilities of know everything, see everything, and control everything. An ASI superintelligence with god-like qualities. We know we have achieved superintelligence or ASI when we can ensure that every person on Earth is provided with a personal guardian angel which provides the right information at the right time in the right language. The tools and techniques that we develop to realize AGI and ASI will ultimately prove to be the most important inventions of the human race and will lead to advances in science and technology and scientific discovery that will eclipse everything that human race has accomplished to date. Human wisdom will be essential to harness AI effectively and govern responsibly. Realizing this vision will require collaboration across disciplines, institutions, cultures, and generations. Three decades ago, I gave it the Turing Award lecture titled To Dream a Possible Dream. I still believe that such dreams are important today. Science begins with a dream. But then, we must do the work to ensure that the dream becomes possible. So, to the young scientists gathered here in Beijing, dream the possible dream. Dream beyond what is reasonable today. Ask the questions that today's technology cannot yet answer. Collaborate and learn from each other, and do the patient, rigorous work required to turn these dreams into reality. Can we create in the future a AI guardian angel to provide every person on Earth the AI Bill of Rights? To provide the right information to the right people at the right time in the right language to anyone anywhere on the planet? Then, in the next century of AI that you build, will realize this dream. Thank you. I wish you all a wonderful and interesting 21st century.
Assistir ao vídeo original no YouTube
Assuntos: AGI, ASI. Publicado no YouTube em 2026-10-19.
Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.