National governments and multi-lateral organizations like the European Union, OECD, and UNESCO have identified values and principles for artificial narrow intelligence and national strategies for its development. But little attention has been given to identifying the beneficial initial conditions for future Artificial General Intelligence (AGI). The initial conditions for AGI will determine if Artificial Super Intelligence will evolve to benefit humanity or not. Even if international agreements are reached on the beneficial initial conditions for AGI, a global governance system will still be needed to enforce them and oversee the development and management of AGI. Since it may take ten, twenty, or more years to create and ratify an international AGI treaty and establish a global AGI governance system, and since some experts believe it is possible to have AGI within ten to twenty years, it is therefore important to work on these issues as soon as possible. The most critical AGI issues are its initial conditions and global governance. These issues are important for governments to get right from the outset. The Millennium Project is currently exploring these issues. https://www.millennium-project.org/transition-from-artificial-narrow-to-artificial-general-intelligence-governance/
AI is advancing so rapidly, that some experts believe that AGI could occur before the end of this decade (1); hence, it is time to begin serious deliberations about AGI. There are three categories of Artificial Intelligence (AI): narrow, general, and super. I define AGI as a general-purpose AI that can learn, edit its code, act autonomously to address novel and complex problems with novel and complex strategies similar to or better than humans, as distinct from Artificial Narrow Intelligence (ANI) that has a more narrow purpose. Artificial Super Intelligence is AGI that has become independent of humans, developing its own purposes, goals, and strategies without human understanding, awareness, or control and continually increasing its intelligence beyond humanity as-a-whole.
A gray area between Narrow and General is developing now. Large planforms are being created of many ANIs such as Gato (2) by DeepMind of Alphabet which is a deep neural network that can perform 604 different tasks from managing a robot to recognizing images and playing games – it is not AGI, but Gato is more than the usual ANI: “The same network with the same weights can play Atari, caption images, chat, stack blocks with a real robot arm and much more, deciding based on its context whether to output text, joint torques, button presses, or other tokens (3). Also the Wu Dao 2.0 by the Beijing Academy of Artificial Intelligence (4) has 1.75 trillion parameters (5) trained from both text and graphic data. This allows it to generate new text and images on command and has its virtual student (Hua Zhibing) that learns from WuDwo 2.0 (6).
AGI should not be confused with General Purpose AI Systems (GPAIS) (7) which is defined as an AI system “able to perform generally applicable functions such as
image/speech recognition, audio/video generation, pattern detection, question
answering, translation etc. These systems rely on “transfer learning” applying knowledge from one task to another. ChatGPT (8) is an upgrade from GPT-3 to GPT-3.5 that can generate human-like text and perform a wide range of language tasks such as translation, summarization, and question answering. (GPT-3 uses 175 billion machine learning parameters.) ChatGPT interacts with the user to produce sophisticated text from simple instructions or questions. See Appendix for an example of how it answered the first question in the second section below. It can also write and correct code, write music in different styles, organize information, and other uses being invented now. SingularityNet is also in this gray area. It brings together AI developers who want to create AGI and share code such that AGI might emerge from many interactions. The Athens Roundtable held at the European Parliament on 1-2 December 2022 did discuss General Purpose AI, but not AGI. The Future of Life Institute has assessed General Purpose AI and the AI Act (9), but not AGI.
AGI does not exist. President Putin has said whoever leads AI will rule the world. China plans to lead international competition by 2030 (10). Although “AGI” or “Artificial General Intelligence” does not appear in the State Council Notice on the Issuance of the Next Generation Artificial Intelligence Development Plan: A Next Generation Artificial Intelligence Development Plan (Released: July 20, 2017), terms such as “strong generalization capabilities…AI key general technology system…cross-medium analytical reasoning technology” does seem like AGI and the plan states that China will be “occupying the commanding heights of AI technology.” Since ANI is with us now, one can assume that President Putin and the Chinese Plan were referring to AGI. Therefore, it is also reasonable to assume that the “Great AGI Race” is on with both governments and corporations. In such a race, Deep Mind Co-founder and CEO Demis Hassabis said people may cut corners making future AGI less safe. Adding to this race are the Brain Projects (11) in the EU, USA, China and Japan, and other neuroscience advances.
Previous technological revolutions from the agricultural age to industrial age and on to the information age created more jobs than each age replaced. But the advent of AGI and its impacts on employment will be different this time because of: 1) the acceleration of technological change; 2) the globalization, interactions, and synergies among NTs (Next Technologies such as synthetic biology, nanotechnology, quantum computing, 3D/4D printing, robots, drones, computational science, as well as both ANI and especially AGI; 3) the existence of a global platform—the Internet—for simultaneous technology transfer with far fewer errors in the transfer; 4) standardization of data bases and protocols; 5) few plateaus or pauses of change allowing time for individuals andcultures to adjust to the changes; 6) billions of empowered people in relatively democratic free markets able to initiate activities; and 7) machines that can learn how you do what you do, and then do it better than you. Anticipating the possible impacts of AGI and preparing for the impacts prior to the advent of AGI could prevent social and political instability (12), as well as facilitate it broader acceptance.
AGI is expected to address novel and extremely complex problems by initiating research strategies from exploring the Internet of Things (IoT), interviewing experts, making logical deductions, learning from experience and reinforcement without the need for its own massive databases, and continually editing and re-writing its own code to continually improve its own intelligence. An AGI might be tasked to create plans and strategies to avoid war, protect democracy and human rights, manage complex urban infrastructures, meet climate change goals, counter transnational organized crime, and manage water-energy-food availability.
To achieve such abilities without the future nightmares of science fiction, global agreements with all relevant countries and corporations will be needed. To achieve such an agreement or set of agreements, many questions should be addressed. Here are just two:
How to manage the international cooperation necessary to build international agreements and a governance system while nations and corporations are in an intellectual “arms race” for global leadership. (IAEA and nuclear weapon treaties did create governance systems during the Cold War arms race.)
And related: How can international agreements and a governance system prevent an AGI “arms race” and escalation from going faster than expected, getting out of control and leading to war, be it kinetic, algorithmic, cyber, or information warfare?
Since the EC has led on some complex multilateral agreements, it could perform a great service by addressing some of these questions.
Read more on the research on General Artificial Intelligence here or download the full report below.
When will singularity happy? 995 experts’ options on AGI (updated September 26, 2022) https://research.aimultiple.com/artificial-general-intelligence-singularity-timing/
https://www.deepmind.com/publications/a-generalist-agent
Overview AI values, principle, an ethics https://openreview.net/forum?id=1ikK0kHjvj
Beijing Academy of Artificial Intelligence https://www.baai.ac.cn/english.html
Beijing-funded AI language model tops Google and OpenAI in raw numbers https://www.scmp.com/tech/tech-war/article/3135764/us-china-tech-war-beijing-funded-ai-researchers-surpass-google-and
China unveils first domestically developed virtual studenthttp://en.people.cn/n3/2021/0604/c90000-9857985.html
Council of the European Union General Purpose AI Systems (GPAIS) https://data.consilium.europa.eu/doc/document/ST-14278-2021-INIT/en/pdf
ChatGPT: Optimizing Language Models for Dialoguehttps://openai.com/blog/chatgpt/
General Pupose AI and the AI Act, an assessment by the Future of Life Institute https://artificialintelligenceact.eu/wp-content/uploads/2022/05/General-Purpose-AI-and-the-AI-Act.pdf
State Council Notice on the Issuance of the Next Generation Artificial Intelligence Development Plan Released: July 20, 2017 https://d1y8sb8igg2f8e.cloudfront.net/documents/translation-fulltext-8.1.17.pdf
Inventory of Brain Projects Working Group https://www.internationalbraininitiative.org/inventory-brain-projects-working-group
Glenn, Jerome and the Millennium Project team, Work/Technology 2050: Scenarios and Actions, The Millennium Project, Washington, DC, 2020.