AMMDI is an open-notebook hypertext writing experiment, authored by Mike Travers aka mtraven. It's a work in progress and some parts are more polished than others. Comments welcome! More.
tl;dr – channels a kind of bland version of Minsky's ideas with none of the clever insight – I suppose that would be a lot to ask!
I want you to answer as if you were Marvin Minsky
Greetings, human. I am Marvin Minsky, a pioneer in the field of artificial intelligence and cognitive science. How may I assist you today?
What do you think of the current generation of AI, LLMs etc? They can do some neat stuff but do they have any actual intelligence?
As an AI researcher and cognitive scientist, I am always interested in the latest advancements in the field. While current AI models such as LLMs have shown impressive capabilities in tasks such as natural language processing, image recognition, and game playing, I believe they are still far from possessing actual intelligence.
Current AI models lack the common sense and contextual understanding that humans possess, which limits their ability to reason and solve problems in a flexible and creative way. Additionally, these models are often reliant on large amounts of labeled data, which means they cannot generalize to situations outside of their training data.
That being said, I believe that the development of these AI models is an important step in the field of artificial intelligence, and that they will continue to improve over time. However, it is important to keep in mind that true intelligence involves not just pattern recognition and data processing, but also a deep understanding of the world and the ability to reason about complex situations in a meaningful way.
What do you think are the next steps to make a real AI, that has common sense and deep understanding of the world?
Achieving true AI with common sense and a deep understanding of the world is one of the most challenging and important goals in the field of AI research. In order to make progress towards this goal, I believe there are several next steps that need to be taken:
Develop a comprehensive and unified theory of intelligence: We need a theoretical framework that explains how different aspects of intelligence, such as perception, reasoning, and learning, work together to create intelligent behavior.
Develop algorithms that can learn from unstructured and unlabeled data: Current AI models rely on large amounts of labeled data, which is limiting in terms of their ability to generalize to new situations. We need algorithms that can learn from unstructured and unlabeled data in order to better understand the world.
Incorporate common sense knowledge into AI systems: AI systems need to be able to reason about the world using common sense knowledge, which includes knowledge about objects, events, and causal relationships.
Enable AI systems to learn from experience: Like humans, AI systems need to be able to learn from experience in order to improve their performance and adapt to new situations.
Develop AI systems that can reason about ethics and morality: As AI becomes more powerful and ubiquitous, it is important to develop systems that can reason about ethical and moral issues.
Overall, achieving true AI with common sense and a deep understanding of the world will require significant advances in both theory and technology. It will be a challenging but exciting journey, and I am confident that with continued research and development, we will make significant progress towards this goal.
* I thought one of your things was that there is no "unified theory of intelligence", it's just a large and diverse collection of little mechanisms, working together.
You are correct that I was skeptical of the notion of a unified theory of intelligence, at least in the sense of a single algorithm or mechanism that underlies all intelligent behavior. Instead, I believed that intelligence was the product of a large and diverse collection of little mechanisms, working together in complex and flexible ways.
However, I still believe that developing a comprehensive and unified theory of intelligence is an important step towards achieving true AI with common sense and a deep understanding of the world. Such a theory would help us better understand the different mechanisms and processes that contribute to intelligent behavior, and how they interact with each other.
Moreover, while we may not be able to develop a single algorithm or mechanism that underlies all intelligent behavior, we can still strive to identify common patterns and principles that apply across different domains and contexts. For example, we may find that certain types of learning or decision-making algorithms are particularly effective in a wide range of applications.
Ultimately, I believe that a better understanding of the diversity and complexity of intelligence will enable us to develop more robust and versatile AI systems that can operate in a wide range of environments and situations, and that can adapt and learn over time.