Artificial Super Intelligence – Can We Control It?
Artificial Super Intelligence – Can We Control It?
Vicar Sayeedi
January 17, 2021
What happens once Artificial Intelligence exceeds human intelligence? Will we be able to control such a Super Intelligent AI? It’s a concerning question. As we continue to progress in both our understanding of human intelligence and our implementation of Artificial Intelligence, scientists and philosophers are warning of the dangers of a super-intelligent AI that is beyond the scope of human understanding, and therefore human control. In a worrisome conclusion, a team of AI researchers from the Centre for Humans and Machines at the Max Planck Institute for Human Development in Berlin has recently confirmed that it would not be possible to control such a super-intelligent AI. Their results were published in early January 2021 in the Journal of Artificial Intelligence Research.
In theory, if we were to program such an AI system with intelligence superior to our own and then teach it to learn completely independently via an Internet connection, such an AI Agent would then have access to most if not all of the collective knowledge developed and published by humankind. Armed with such a comprehensive base of knowledge to meaningfully inform its thinking and adjudication, it is simply beyond human comprehension to understand the rationale for the decisions the Super AI has made or any additional decisions it might make, or how those decisions might impact humankind or our world.
For example, if a Super AI Agent concluded that the most efficient way to end human suffering were to simply eliminate humankind, what could we do to stop it? This is the question posed by researchers at Max Planck Society in their recently published research. Their answer suggests that given our current technological capabilities, we would not be able to stop the Agent and so we would be left to stand by and watch as the Agent initiated our destruction. Perhaps the most distressing reality in this discussion is that Super AI Agents capable of having a significant impact on humankind are not very far away in time and technical limitation.
Discussions about such super-intelligent AI Agents that might one day seize control from humans does seem like science fiction until we realize that there are already AI Agents that can perform very important activities without the ability of human programmers to reasonably comprehend how these Agents developed their response. This does make us worry and raises an important question: will we soon cross that point where things are beyond human control and how dangerous might this be for humankind?
Scientists have begun to explore different mechanisms for controlling such a super-intelligent AI. In one scenario, the capabilities of the Super AI could be physically limited by extricating any links to the Internet, other devices or technology. However, this would effectively make the super-intelligent AI significantly less capable or powerful. Alternatively, the AI’s design would incorporate the principles of human ethics so it would only pursue goals that are in the best interests of humankind. But researchers have shown that these and other contemporary and historical ideas for controlling super-intelligent AI have their limits, too.
Researchers have designed a theoretical containment algorithm that would ensure that any super-intelligent AI Agent cannot under any circumstances harm humans or their interests by first simulating the behavior of the AI Agent and stopping it if its direction or behavior appear harmful. However, the researchers believe that we are not able to build such an algorithm. Iyad Rahwan, the Director of the Centre for Humans and Machines at Max Planck, explains that breaking the problem down to basic rules from theoretical computer science demonstrates that an algorithm designed to prevent a Super AI Agent from destroying the world could possibly halt its own operation. Rahwan went on to say that, “if this happened, you would not know whether the containment algorithm is still analyzing the threat or whether it has stopped to contain the harmful AI. In effect, this makes the containment algorithm unusable.” Thus, such a containment approach is not feasible or possible. Additionally, researchers believe we are not really capable of realizing when we have crossed the Rubicon into the world of super-intelligent AI Agents.
The work of researchers at Max Planck and similar centers is of great importance given the power of Super AI and the pace at which science and technology are converging to realize this extraordinary attribute of the Fourth Industrial Revolution. Let's hope they get it right.
Vicar Sayeedi is a Computer Scientist and Engineer, a Lecturer and a Consultant. Vicar is also the author of several books. His most recent book is about Artificial Intelligence and is titled, The Génome Affair.
Vicar is most interested in the big questions facing humankind. He is particularly focused on studying at the confluence of the five great disciplines of Human History, Political Science and Thought, International Affairs, Science and Technology. This intersection offers a deep understanding and pedagogically important lessons of how advances in human endeavor have influenced and impacted civilization.
Vicar has been writing books, essays and poetry for many years. For the past 30 years, Vicar has worked in the Technology and Pharmaceutical industries. He is currently a Consultant in the Life Sciences Industry and lives in suburban Chicago with his wife and their three grown children [if they decide to visit during the holidays!]
You can find Vicar’s recent book on AI, The Génome Affair on Amazon at this link:
https://www.amazon.com/dp/B07YQ7M9Q7
The most recent episodes of Vicar’s ai & u! Podcast are on YouTube at the following links:
Episode 1 - https://youtu.be/yNYr28jtY_k
What is Artificial Intelligence? What is Human Intelligence?
Episode 2 - https://youtu.be/kAEgDNh1Nwc
How Did We Get Here? Machine Learning, Neural Networks & the AI Lexicon, Exciting AI Agents
Episode 3 - https://youtu.be/3PYCyv1pCgM
Deep Mind - How Powerful is Narrow AI? State Surveillance and Surveillance Capitalism
Episode 4 - https://youtu.be/WvobCMIM_H4
Impact on Societies That Possess Artificial Intelligence Versus Those that Do Not!
Episode 5 - https://youtu.be/4JVpvm4g79g
Surveillance Capitalism, Surveillance State, 4th Industrial Revolution, Bifurcation of Sapiens
Episode 6 - https://youtu.be/e1q2GgcWQok
AI, Blockchain & Quantum Computing - Implications for Global and National Governance
Episode 7 - https://youtu.be/4bDSUvrOdd4
Popular Perceptions of Intelligence, the Seat of Human Intelligence and Implications for AI
Episode 8 - https://youtu.be/3GjHqQZL7Pk
AI - Why Should I care? Plus Exciting Updates from Elon Musk and Neuralink!
Episode 9 - https://youtu.be/enpecqDecC8
AI - Excellence, Innovation & Genius
Episode 10 - https://youtu.be/UsriESlTjdA
Implications for AI as China and the West Decouple
Episode 11 - https://youtu.be/0CpaMb-yw3g
Artificial Intelligence: Ethics, Law & Society
Episode 12 - https://youtu.be/PLzodrLWMq0
AI & Lethal Autonomous Weapons Systems
Episode 13 - https://youtu.be/Lm_nrNyJRbA
AI & the Future of Work in Industry 4.0 – the 4th Industrial Revolution
Episode 14 - https://youtu.be/_3E1Mhr_Dt8
Artificial Intelligence and its Implications for Authoritarianism and Democracy
Episode 15 - https://youtu.be/IpMUqHOEd7w
Artificial Intelligence, Consciousness & the Future of Homo Sapiens
Episode 16 - https://youtu.be/mIf7_VMzGKo
AI and its Implications for America's Political Environment
Episode 17 - https://youtu.be/gsMRPnovZGI
Special Episode: Reading From The Génome Affair
Episode 18 - https://youtu.be/UxNxtWDCREM
Artificial Intelligence - Digital Tyranny, Datocracy and the Imitation Game
Episode 19 - https://youtu.be/Db8U0QQWWuY
Does Artificial Intelligence Really Replicate Human Intelligence?
Episode 20 - https://youtu.be/Np18AEq-6DM
AI & Rising Geopolitical Tensions Between China & The West
Episode 21 - https://youtu.be/fOMSvNiixA0
AI & In-vitro Fertilization [IVF]
Episode 22 - https://youtu.be/EQPpRhIV6Kw
AI & Human Intelligence: A Complex Relationship
Episode 23 - https://youtu.be/EAbQ7stY7JQ
AI Upends the World of Structural Biology
Episode 24 - https://youtu.be/Bsb1F8uoBpA
AI & COVID-19: An Extraordinary Contribution
Episode 25 - https://youtu.be/PqL5PCRc8tA
AI, Connectomics & Transhumanism: the Future of Humankind
Episode 26 - https://youtu.be/HI8Zjl8VM8U
Dramatic Upheaval Inside Google's Ethical AI Team
Episode 27 - https://youtu.be/TqWwJYc2GxQ
AI & Cyber Espionage | How Will the Sunburst Attack Affect Our Security?
Episode 28 - https://youtu.be/89BcThqgcfQ
AI & the Rapidly Escalating Threat from Deep Fake Videos
Episode 29 - https://youtu.be/zvOHRMmJNB4
AI, Facial Recognition Technology & The Digital Surveillance State
Episode 30 - https://youtu.be/br4ygfEWdk0
AI & Social Media | The Most Destructive AI We've Ever Seen?
Episode 31 - https://youtu.be/sOy12ii8zwM
The Democratization of AI & The 2021 Maiflower Expedition | Plymouth, England to Plymouth, MA
Episode 32 - https://youtu.be/bixdLL74reE
Sino-American Cold War | A Blessing in Disguise for America?
Episode 33 - https://youtu.be/tTRm312DgMc
AI & the Tyranny of Meritocracy
Episode 34 - https://youtu.be/FpPzC5VP6GM
#34 - In the Midst of Chaos and Pandemic | A Big AI Announcement From the Trump Administration
Episode 35 - https://youtu.be/bK3xCBwrG8w
Artificial Super Intelligence – Can We Control It?
Vicar Sayeedi
January 17, 2021
What happens once Artificial Intelligence exceeds human intelligence? Will we be able to control such a Super Intelligent AI? It’s a concerning question. As we continue to progress in both our understanding of human intelligence and our implementation of Artificial Intelligence, scientists and philosophers are warning of the dangers of a super-intelligent AI that is beyond the scope of human understanding, and therefore human control. In a worrisome conclusion, a team of AI researchers from the Centre for Humans and Machines at the Max Planck Institute for Human Development in Berlin has recently confirmed that it would not be possible to control such a super-intelligent AI. Their results were published in early January 2021 in the Journal of Artificial Intelligence Research.
In theory, if we were to program such an AI system with intelligence superior to our own and then teach it to learn completely independently via an Internet connection, such an AI Agent would then have access to most if not all of the collective knowledge developed and published by humankind. Armed with such a comprehensive base of knowledge to meaningfully inform its thinking and adjudication, it is simply beyond human comprehension to understand the rationale for the decisions the Super AI has made or any additional decisions it might make, or how those decisions might impact humankind or our world.
For example, if a Super AI Agent concluded that the most efficient way to end human suffering were to simply eliminate humankind, what could we do to stop it? This is the question posed by researchers at Max Planck Society in their recently published research. Their answer suggests that given our current technological capabilities, we would not be able to stop the Agent and so we would be left to stand by and watch as the Agent initiated our destruction. Perhaps the most distressing reality in this discussion is that Super AI Agents capable of having a significant impact on humankind are not very far away in time and technical limitation.
Discussions about such super-intelligent AI Agents that might one day seize control from humans does seem like science fiction until we realize that there are already AI Agents that can perform very important activities without the ability of human programmers to reasonably comprehend how these Agents developed their response. This does make us worry and raises an important question: will we soon cross that point where things are beyond human control and how dangerous might this be for humankind?
Scientists have begun to explore different mechanisms for controlling such a super-intelligent AI. In one scenario, the capabilities of the Super AI could be physically limited by extricating any links to the Internet, other devices or technology. However, this would effectively make the super-intelligent AI significantly less capable or powerful. Alternatively, the AI’s design would incorporate the principles of human ethics so it would only pursue goals that are in the best interests of humankind. But researchers have shown that these and other contemporary and historical ideas for controlling super-intelligent AI have their limits, too.
Researchers have designed a theoretical containment algorithm that would ensure that any super-intelligent AI Agent cannot under any circumstances harm humans or their interests by first simulating the behavior of the AI Agent and stopping it if its direction or behavior appear harmful. However, the researchers believe that we are not able to build such an algorithm. Iyad Rahwan, the Director of the Centre for Humans and Machines at Max Planck, explains that breaking the problem down to basic rules from theoretical computer science demonstrates that an algorithm designed to prevent a Super AI Agent from destroying the world could possibly halt its own operation. Rahwan went on to say that, “if this happened, you would not know whether the containment algorithm is still analyzing the threat or whether it has stopped to contain the harmful AI. In effect, this makes the containment algorithm unusable.” Thus, such a containment approach is not feasible or possible. Additionally, researchers believe we are not really capable of realizing when we have crossed the Rubicon into the world of super-intelligent AI Agents.
The work of researchers at Max Planck and similar centers is of great importance given the power of Super AI and the pace at which science and technology are converging to realize this extraordinary attribute of the Fourth Industrial Revolution. Let's hope they get it right.
Vicar Sayeedi is a Computer Scientist and Engineer, a Lecturer and a Consultant. Vicar is also the author of several books. His most recent book is about Artificial Intelligence and is titled, The Génome Affair.
Vicar is most interested in the big questions facing humankind. He is particularly focused on studying at the confluence of the five great disciplines of Human History, Political Science and Thought, International Affairs, Science and Technology. This intersection offers a deep understanding and pedagogically important lessons of how advances in human endeavor have influenced and impacted civilization.
Vicar has been writing books, essays and poetry for many years. For the past 30 years, Vicar has worked in the Technology and Pharmaceutical industries. He is currently a Consultant in the Life Sciences Industry and lives in suburban Chicago with his wife and their three grown children [if they decide to visit during the holidays!]
You can find Vicar’s recent book on AI, The Génome Affair on Amazon at this link:
https://www.amazon.com/dp/B07YQ7M9Q7
The most recent episodes of Vicar’s ai & u! Podcast are on YouTube at the following links:
Episode 1 - https://youtu.be/yNYr28jtY_k
What is Artificial Intelligence? What is Human Intelligence?
Episode 2 - https://youtu.be/kAEgDNh1Nwc
How Did We Get Here? Machine Learning, Neural Networks & the AI Lexicon, Exciting AI Agents
Episode 3 - https://youtu.be/3PYCyv1pCgM
Deep Mind - How Powerful is Narrow AI? State Surveillance and Surveillance Capitalism
Episode 4 - https://youtu.be/WvobCMIM_H4
Impact on Societies That Possess Artificial Intelligence Versus Those that Do Not!
Episode 5 - https://youtu.be/4JVpvm4g79g
Surveillance Capitalism, Surveillance State, 4th Industrial Revolution, Bifurcation of Sapiens
Episode 6 - https://youtu.be/e1q2GgcWQok
AI, Blockchain & Quantum Computing - Implications for Global and National Governance
Episode 7 - https://youtu.be/4bDSUvrOdd4
Popular Perceptions of Intelligence, the Seat of Human Intelligence and Implications for AI
Episode 8 - https://youtu.be/3GjHqQZL7Pk
AI - Why Should I care? Plus Exciting Updates from Elon Musk and Neuralink!
Episode 9 - https://youtu.be/enpecqDecC8
AI - Excellence, Innovation & Genius
Episode 10 - https://youtu.be/UsriESlTjdA
Implications for AI as China and the West Decouple
Episode 11 - https://youtu.be/0CpaMb-yw3g
Artificial Intelligence: Ethics, Law & Society
Episode 12 - https://youtu.be/PLzodrLWMq0
AI & Lethal Autonomous Weapons Systems
Episode 13 - https://youtu.be/Lm_nrNyJRbA
AI & the Future of Work in Industry 4.0 – the 4th Industrial Revolution
Episode 14 - https://youtu.be/_3E1Mhr_Dt8
Artificial Intelligence and its Implications for Authoritarianism and Democracy
Episode 15 - https://youtu.be/IpMUqHOEd7w
Artificial Intelligence, Consciousness & the Future of Homo Sapiens
Episode 16 - https://youtu.be/mIf7_VMzGKo
AI and its Implications for America's Political Environment
Episode 17 - https://youtu.be/gsMRPnovZGI
Special Episode: Reading From The Génome Affair
Episode 18 - https://youtu.be/UxNxtWDCREM
Artificial Intelligence - Digital Tyranny, Datocracy and the Imitation Game
Episode 19 - https://youtu.be/Db8U0QQWWuY
Does Artificial Intelligence Really Replicate Human Intelligence?
Episode 20 - https://youtu.be/Np18AEq-6DM
AI & Rising Geopolitical Tensions Between China & The West
Episode 21 - https://youtu.be/fOMSvNiixA0
AI & In-vitro Fertilization [IVF]
Episode 22 - https://youtu.be/EQPpRhIV6Kw
AI & Human Intelligence: A Complex Relationship
Episode 23 - https://youtu.be/EAbQ7stY7JQ
AI Upends the World of Structural Biology
Episode 24 - https://youtu.be/Bsb1F8uoBpA
AI & COVID-19: An Extraordinary Contribution
Episode 25 - https://youtu.be/PqL5PCRc8tA
AI, Connectomics & Transhumanism: the Future of Humankind
Episode 26 - https://youtu.be/HI8Zjl8VM8U
Dramatic Upheaval Inside Google's Ethical AI Team
Episode 27 - https://youtu.be/TqWwJYc2GxQ
AI & Cyber Espionage | How Will the Sunburst Attack Affect Our Security?
Episode 28 - https://youtu.be/89BcThqgcfQ
AI & the Rapidly Escalating Threat from Deep Fake Videos
Episode 29 - https://youtu.be/zvOHRMmJNB4
AI, Facial Recognition Technology & The Digital Surveillance State
Episode 30 - https://youtu.be/br4ygfEWdk0
AI & Social Media | The Most Destructive AI We've Ever Seen?
Episode 31 - https://youtu.be/sOy12ii8zwM
The Democratization of AI & The 2021 Maiflower Expedition | Plymouth, England to Plymouth, MA
Episode 32 - https://youtu.be/bixdLL74reE
Sino-American Cold War | A Blessing in Disguise for America?
Episode 33 - https://youtu.be/tTRm312DgMc
AI & the Tyranny of Meritocracy
Episode 34 - https://youtu.be/FpPzC5VP6GM
#34 - In the Midst of Chaos and Pandemic | A Big AI Announcement From the Trump Administration
Episode 35 - https://youtu.be/bK3xCBwrG8w
Artificial Super Intelligence – Can We Control It?
No comments have been added yet.
