Read Thousand Brain Intelligence Note 07 The Future of Artificial Intelligence Medium .

February**Dynamic Incentive Plan 1The future of machine intelligence.

1.1.There is no technical reason that prevents us from creating smart machines.

1.1.1.The obstacle is our lack of understanding of intelligence and the mechanisms needed to produce it.

1.2.History has shown that we can't afford the technological advancements that drive machine intelligence forward.

1.2.1.In 1950, no one could ** the innovations and advancements that accelerated the development of computers.

1.2.1.1.IC.

1.2.1.2.Solid-state memory.

1.2.1.3.Cellular wireless network communication.

1.2.1.4.Public-key cryptography.

1.2.1.5.Internet.

1.2.2.No one could have predicted how computers would change**, communications, and commerce.

1.2.3.Today it is equally unknown what a smart machine will look like and how we will use it in 70 years.

1.2.3.1.Machine intelligence will have more impact on the 21st century than computers will do in the 20th century.

1.3.By understanding the mechanisms by which the brain produces intelligence, we can know what is possible, what is not, and what kind of progress is possible to some extent.

2.1.The earlier parts of the human brain control the basic functions of life.

2.1.1.They create human emotions, desires to survive and reproduce, and human innate behavior.

2.2.When creating intelligent machines, we don't have to replicate all the functions of the human brain.

2.2.1.The neo-brain, or neocortex of the brain, is the organ that embodies human intelligence, and intelligent machines need to have something comparable to it.

2.2.2.Other parts of the brain, we can pick out some of the parts we want.

2.3.The map itself does not prescribe these uses, nor does it give any value to how it is used.

2.3.1.It's a map that is neither ** nor peace-loving.

2.3.2.Some maps may be better suited for war, while others are more suitable.

2.3.3.The desire to wage war or engage in ** comes from the people who use the map.

2.4.Intelligence is the ability to systematically learn models of the world.

2.4.1.Goals and values are provided by the system that uses the model.

2.5.The neocortex of the brain learns a model of the world that has no purpose or value in itself.

2.5.1.The neocortex of the brain does not create goals.

2.5.1.1.The neocortex of the brain, although much larger than the old brain, is made up of many relatively small elemental columns of the cortex.

2.5.2.The emotions that guide our actions are determined by the old brain.

2.5.3.If a person's old brain is aggressive, then it will use the model in the neocortex of the brain to better implement aggressive behavior.

2.5.4.If another person's old brain is benevolent, then it will use the model in the brain's neocortex to better achieve its benevolent goal.

2.6.Intelligent machines need a model of the world, and the behavioral flexibility that comes with it.

2.6.1.There is no need to have human-like instincts for survival and reproduction.

2.6.2.Designing a machine with human emotions is harder than designing a machine with intelligence.

2.6.2.1.In order to build a machine with human emotions, we have to rebuild the parts of the old brain.

2.7.Designing a smart machine can start with three parts.

2.7.1.Empodiment

2.7.1.1.To learn a new tool, we must hold it in our hands, constantly turn it, observe and pay attention to its different parts with our eyes.

2.7.1.2.Smart machines also need sensors and the ability to move those sensors.

2.7.1.2.1.It is called "embodied".

2.7.1.3.1.Actions and positions don't necessarily exist in physical space.

2.7.1.3.2.Smart machines can do the same thing without moving using software.

2.7.1.4.1.They do not have movable sensors and no frame of reference to determine the orientation of the sensors.

2.7.1.4.2.There is a limit to what can be learned without embodiment.

2.7.1.5.1.The main human senses are sight, touch, and hearing.

2.7.1.5.1.1.Human sight, touch, and hearing are all enabled by an array of sensors.

2.7.1.5.1.2.The senses of smell and taste are different in nature from the senses of sight and touch.

2.7.1.5.1.3.The sense of taste is also limited when it comes to perceiving what's in the mouth.

2.7.1.5.1.4.Smell and taste can help us judge whether food is safe or not, and smell can help us identify a general area, but we don't rely too much on them to understand the detailed structure of the world.

2.7.1.5.1.5.Hearing is somewhere in between.

> 2.7.1.5.2.Bats have sonar" 27.1.5.3.Some fish have the senses to emit an electric field" 27.1.5.4.A robot capable of rescuing humans in a collapsing building may have radar sensors so it can see things in the dark

2.7.1.6.1.Intelligent Protein Machine (intelligence

> 2.7.1.6.2.An intelligent machine that exists inside a single cell and understands proteins > 27.1.6.2.1.Proteins are long molecules that can naturally fold into complex shapes >27.1.6.2.2.Our brains aren't very good at understanding proteins > 27.1.6.2.3.Proteins also work much faster than the brain can process

2.7.1.7.1.The human brain's neocortex has about 150,000 cortical columns, each of which models a portion of the world it can perceive.

2.7.1.7.2.A smart machine may have millions of cortical columns and thousands of sensor arrays.

2.7.2.Old brain part.

2.7.2.1.To create an intelligent machine, the old brain part of the brain is needed.

2.7.2.1.1.Smart machines also need some old brain functions.

2.7.2.2.1.The neocortex of the brain does not directly control any muscles.

2.7.2.2.1.1.It sends signals to the old brain, which controls movement more directly.

2.7.2.2.2.Animals need to walk and run before they can evolve the neocortex of the brain.

2.7.2.4.Smart machines also need built-in security.

2.7.2.4.1.As with any product design, there are a number of security measures that need to be considered.

2.7.2.4.2.If the car detects that the driver is about to hit an obstacle, it ignores the driver's instructions and applies the brakes.

2.7.2.5.1.Human goals and motivations are complex, and some are driven by our genes.

2.7.2.7.Giving machines goals and motivations requires us to design specific mechanisms for goals and motivations, and then embed them into the embodiment of the machine.

2.7.2.8.Goals and motivations are not the result of intelligence and do not emerge on their own.

2.7.3.Neocortex of the brain.

2.7.3.1.The neocortex of the brain implements a near-universal algorithm, but this flexibility comes at a cost.

2.7.3.2.Rather than creating entirely new behaviors, it learns how to put existing behaviors together in new and useful ways.

2.7.3.3.Primitives of behaviors can be as simple as bending a finger or as complex as walking, but the neocortex of the brain requires that these behaviors exist in their own right.

2.7.3.4.The neocortex of the brain itself does not create goals, motivations, or emotions.

2.7.3.5.The cerebral neocortex is closely related to the way in which motivation and goals influence behavior, but the cerebral neocortex does not direct behavior.

2.7.3.6.Velocity.

2.7.3.6.1.It takes at least 5 milliseconds for neurons to make useful behaviors.

2.7.3.6.2.Silicon crystals run 1 million times faster than neurons.

2.7.3.6.3.The brain's neocortex, made of silicon, may be 1 million times faster than humans can think and learn.

2.7.3.6.4.The fact that a part of an intelligent machine is 1 million times faster than a biological brain does not mean that the speed of the entire intelligent machine can reach this level, nor does it mean that it will acquire knowledge so quickly.

2.7.3.6.5.With the use of intelligent machines, the whole process of creating a habitable environment on Mars by humans may be several times faster, but not a million times faster.

2.7.3.6.6.Computers can complete tasks that were previously done by hand at a rate of 1 million times faster than humans.

2.7.3.6.6.1.Intelligent machines will have a similar impact on human society and the speed at which humans make scientific discoveries.

2.7.3.6.7.Replacing humans with intelligent machines to study neuroscience can speed up scientific discovery, but not 1 million times faster.

2.7.3.6.8.Time spent reading**, discussing all possible theories, and writing software.

2.7.3.6.8.1.In principle, some of this work can be done much faster by intelligent machines.

2.7.3.6.8.2.But our software simulations still take a few days to run.

2.7.3.6.8.3.Our theories are not developed in a vacuum, but rely on experimentation.

2.7.3.6.8.3.2.We cannot dramatically speed up the development of telescopes and particle detectors, nor can we reduce the time it takes for them to collect data.

> 2.7.3.6.9.Intelligent machines can also dramatically speed up research work in some areas >27.3.6.9.1.The main job of a mathematician is to think, write, and share ideas > 27.3.6.9.2.Theoretically, intelligent machines can process certain mathematical problems 1 million times faster than human mathematicians > 27.3.6.9.3.The speed at which an intelligent web crawler learns is limited by the speed at which it "moves" through a tracker link and opens a file. The process can be very fast

2.7.3.7.1.Our brain's neocortex has gotten bigger, and we've gotten smarter.

2.7.3.7.1.1.Machine intelligence can also follow the same mechanism.

2.7.3.7.2.There is no obvious limit to the size of an artificial brain that can be created.

2.7.3.7.2.1.Enlarging certain areas of the brain's neocortex may have some effects, but it won't give you some kind of superpower.

2.7.3.7.2.2.Most people would think that humans are smarter than monkeys, and that our brains build a deeper and more comprehensive model of the world.

2.7.3.7.2.3.Intelligent machines can surpass humans in depth of understanding.

2.7.3.7.2.3.2.It's like even though I couldn't have made a discovery like Einstein's, I could understand his discovery.

> 2.7.3.7.3.When we are born, there are too many brain circuits in the neocortex of the brain. Over the next few years, the number of brain circuits will decline significantly" 27.3.7.4.Intelligent machines have no constraints associated with brain circuits

2.7.4.1.We can copy a smart machine at any time, clone it.

2.7.4.2.As long as the bot's capabilities reach a satisfactory level, we can migrate the connections it learns to a dozen other identical bots to create replicas.

3.1.Every time a new technology is created, we imagine using it to replace or improve something we are familiar with.

3.2.New uses that no one could have anticipated emerged, and it was these unexpected uses that often became the most important and would change society as a whole.

3.3.When the Internet Protocol first appeared, few people could have foreseen these societal changes.

3.4.The goal of artificial intelligence is to imitate humans, a concept that is embodied in the famous "Turing Test".

3.5.Hazardous work and work that is harmful to health, such as deep-sea repairs or cleaning up toxic spills, may be too risky for humans.

3.6.Use smart machines to complete tasks that lack sufficient manpower, such as caring for the elderly.

3.7.Learning about science is an exciting application.

3.7.1.Human beings want to learn, so they are attracted to explore, to seek knowledge, to understand the unknown.

3.7.2.When intelligent machines can think faster and deeper than we do, can perceive things we can't, and can travel where we can't, who knows what we'll learn.