The notion of intelligence remains intriguing to researchers and practitioners, yet still somewhat mysterious. After over a century of research into the topic, there is still no widely accepted definition. Many say that intelligence is the ability to solve problems, but exactly what problems do they have in mind? Is it the ability to collect reward in Markov decision processes, determine truth in first-order arithmetic, find patterns in progressive matrices, or something else? Could it be argued that some problems are more natural than others in this context?
According to evolutionary theory, animal intelligence, including human intelligence, evolved in response to demands for problem-solving in nature. Moreover, natural selection favors the genes of animals that can reproduce in a relatively broad class of ecosystems. To be able to reproduce, animals must solve a continuous stream of problems during their lives, e.g., finding food, avoiding predators, mating, and parenting. This suggests that human intelligence primarily evolved for solving everyday problems related to survival in ecosystems ranging from the tropics to the polar regions.
Artificial Intelligence started as an attempt to reproduce parts of human intelligence in machines and, like the notion of intelligence, it suffers from or enjoys a certain vagueness regarding its definition, targeted problems, performance measures, and relations to neighboring research fields.
Today, AI researchers are increasingly focusing on generality. They build systems that can translate between many languages, play many games, manipulate many objects, predict many video frames, and diagnose many diseases. They design autonomous cars that can drive in many cities, dialogue systems that can answer many questions, and robots that can navigate many homes. They want their systems to learn quickly, generalize well, and never stop learning.
AGI is AI with generality put at the center. Since 2008, the Artificial General Intelligence Society has organized annual conferences on the topic. The 16th AGI conference (AGI-23), will be held as a mixed virtual/F2F event in Stockholm, Sweden on June 16 – 19, 2023. We look forward to welcoming you to the conference!
Conference UR: https://agi-conf.org/2023/
Sponsor: Artificial General Intelligence Society
Co-sponsor: Cisco
Co-sponsor: Digital Futures