(2nd-January-2021)
• Today, we describe how numerical values assigned to outcomes fit into a mathematical framework for the way that AI systems interact with their environments. Here, we describes a way that AI systems can learn models of their environments. Together, numerical values assigned to outcomes and a model of the environment enable us to write down mathematical equations that define the behavior of future AI. We can use these equations to analyze the behaviors of future AI systems in order to design systems that help rather than harm humans.
• Also we discusses how mathematical equations for intelligence should be limited to finite amounts of information. Infinite sets are mathematically interesting but not necessary in our finite universe.
• And we describe how AI systems may choose actions harmful to humans, despite the fact that those actions are not intended in the AI system's design. These unintended, harmful actions make the design of future AI systems very difficult.
• The proposed AI system design as a set of mathematical equations. Advanced AI systems will need to learn their own environment models and our intentions for their behavior must be expressed in terms of their learned models. We discusses ways of assigning numerical values to outcomes based on learned environment models. This is also a way to avoid AI systems that intentionally delude themselves about their environments. We are talking about a way to assign numerical values to outcomes that expresses our human values while avoiding several pitfalls. Accurate expression of human values is the key to avoiding the unintended, harmful actions described inhere, AI
• Systems must exist in the real world, subject to the resource limits and vulnerabilities of any creature in the real world. Adapt the mathematical equations of intelligence to AI in the real world is discussed. We also discusses the problem of ensuring that AI system designs will remain ethical as they and humanity evolve together.
• Proposed a system design, based on our mathematical equations, for testing proposed AI designs while avoiding the risks they pose. Transparency and an ethical culture are vital components of AI development and testing.
• Seeing the politics of AI. If we know how to design ethical AI, how can we make sure that real AI systems conform to ethical designs? And how will the benefits of AI be allocated among humans?
• Finally, we are looking at the ultimate goal of building AI. If future AI systems help rather than harm humans, our physical needs will be easily met. At that point, the only challenge worthy of humanity and its super-intelligent AI systems will be to understand the nature of the universe and our place in it.
Comments