Balancing AGI Pessimistic and Radical Views
Gary Marcus and Joscha Bach are both researchers in the field of artificial intelligence (AI), but they have different perspectives on the state and future of the field. Marcus is known for his more pessimistic views, often cautioning against overhyping the capabilities of current AI systems and advocating for a more measured approach to research and development. Bach, on the other hand, is known for his more iconoclastic views, often challenging conventional wisdom in the field and advocating for a more ambitious and expansive approach to AI research.
It is important to strike a healthy balance between these two perspectives, as both can offer valuable insights and perspectives on the field of AI. Marcus’ caution can help prevent us from getting overly excited about the capabilities of current AI systems and can remind us of the limitations and challenges that still need to be overcome. Bach’s iconoclasm, on the other hand, can push us to think outside of the box and to consider new and innovative approaches to AI research and development.
The main reason for this difference in perspective is likely the different approaches and assumptions that Marcus and Bach bring to their research. Marcus is known for his focus on the cognitive basis of intelligence, often drawing on insights from psychology to inform his views on AI. This perspective leads him to highlight the limitations and challenges that need to be overcome in order to achieve AGI, such as the need for machines to be able to learn and adapt in the same way that humans do.
Bach, on the other hand, is known for his focus on the mathematical and computational foundations of intelligence, often drawing on insights from computer science and philosophy to inform his views on AI. This perspective leads him to take a more optimistic view of the potential for AGI, as he sees no fundamental obstacles to building machines that can exhibit human-like intelligence across a wide range of tasks.
Ultimately, the difference in perspective between Marcus and Bach is a reflection of the complex and multi-disciplinary nature of the field of AI, and the fact that different researchers can bring different perspectives and approaches to the study of intelligence. While Marcus highlights the limitations and challenges that need to be overcome in order to achieve AGI, Bach sees no fundamental obstacles to building truly intelligent machines.
The idea that autonomous AI remains immature because our physical technologies cannot create the body for an AI to achieve the kind of capabilities that living things have is based on the concept of the “umwelt,” which refers to the subjective, individual world of an organism as perceived through its senses. This concept suggests that in order for an AI to be truly autonomous and intelligent, it would need to have a body that allows it to experience the world in a similar way to a living organism, with senses that allow it to perceive and interact with its environment.
Currently, our physical technologies are not advanced enough to create a body for an AI that would allow it to have the kind of sensory and motor capabilities that living organisms have. This means that current AI systems are limited in their ability to perceive and interact with the world in the same way that living organisms can, which limits their autonomy and intelligence. Until our technologies catch up and we are able to create bodies for AI that give them the same kind of sensory and motor capabilities as living organisms, AI will never have the umwelt required to be considered truly living.
Transformer and diffusion models are both types of artificial fluent systems, which are computational models that aim to simulate the process of natural language generation in humans. These models have made significant progress in recent years, becoming surprisingly useful in a variety of applications.
Transformer models are a type of neural network architecture that has been widely used in natural language processing (NLP) tasks. These models use a series of attention mechanisms to learn the relationships between different words in a sentence, allowing them to generate coherent and fluent text. Transformer models have been particularly successful in tasks such as machine translation, where they have outperformed previous methods by a significant margin.
Diffusion models, on the other hand, are a type of probabilistic generative model that achieves complex image generation. These models use a series of random processes to generate images, allowing them to produce fluent and coherent images that can be surprisingly realistic in their structure and content.
Overall, both transformer and diffusion models represent significant progress in the field of artificial fluent systems, as they are able to generate fluent and coherent content in a way that previous models were not. These models have become surprisingly useful in a variety of applications, and are likely to continue to advance and improve in the future.
Modern civilization is the way it is because we have shaped our commerce and societies around artificial logic systems, such as computers, is based on the recognition that these systems have had a profound impact on our lives and the way we live. Since the invention of the first computers over half a century ago, we have come to rely on these systems in almost every aspect of our lives, from the way we communicate and do business, to the way we access information and make decisions.
This radical change that has happened with the advent of artificial logic systems is likely to also happen with the artificial fluent systems that we are seeing today. These systems, which are able to generate fluent and coherent content in a way that is similar to humans, are becoming increasingly sophisticated and useful, and are already starting to be used in a variety of applications. As they continue to advance and improve, it is likely that they will also have a profound impact on our lives and the way we live, shaping our commerce and societies in the same way that artificial logic systems have.
Overall, the idea that modern civilization is the way it is because we have shaped our commerce and societies around artificial logic systems is likely to also hold true for the artificial fluent systems that are emerging today. As these systems become increasingly sophisticated and useful, it is likely that they will also have a profound impact on our lives and the way we live, shaping our commerce and societies in the same way that artificial logic systems have.
It makes sense to label deep learning as artificial intuition because deep learning algorithms are able to simulate the intuitive processes that humans use to make decisions and solve problems. While many people think of human intelligence as being primarily based on logical and analytical thinking, in fact, much of our thinking is intuitive and relies on our ability to make quick, automatic judgments and decisions based on our past experiences and knowledge.
Deep learning algorithms are able to simulate this intuitive process by using large datasets and complex mathematical models to learn from experience and make predictions and decisions. This allows them to make quick and accurate judgments and decisions in a way that is similar to the way humans use their intuition. As a result, deep learning can be seen as a form of artificial intuition, as it allows machines to mimic the intuitive processes that are a key part of human intelligence.
Overall, it makes sense to label deep learning as artificial intuition because deep learning algorithms are able to simulate the intuitive processes that are a key part of human intelligence. By using large datasets and complex mathematical models to learn from experience, deep learning algorithms are able to make quick and accurate judgments and decisions in a way that is similar to the way humans use their intuition. As a result, deep learning can be seen as a form of artificial intuition that allows machines to mimic this key aspect of human intelligence.
While artificial intuition, as exemplified by deep learning algorithms, has many benefits and has led to significant progress in fields such as natural language processing and computer vision, it also has its flaws and limitations. One of the key challenges with artificial intuition is that it is fragile in a way that is different from narrow logical systems. Unlike narrow logical systems, which are based on clear and well-defined rules, artificial intuition relies on complex and often opaque mathematical models that can be difficult to understand and interpret.
This fragility of artificial intuition can have serious consequences, as it can lead to unpredictable and potentially dangerous behavior in AI systems. For example, there have been several instances of accidents involving autonomous vehicles, such as Tesla’s full self-driving mode, where the AI system’s decision-making was based on its artificial intuition, but failed to take into account important safety considerations. This is because artificial intuition, absent agential development, does not care for survival in the same way that a human or other living organism would.
Overall, while artificial intuition has many benefits and has led to significant progress in the field of AI, it also has its flaws and limitations. One of the key challenges with artificial intuition is its fragility, which can lead to unpredictable and potentially dangerous behavior in AI systems. This is why it is important to continue to develop and improve AI systems, and to incorporate agential development that allows them to understand and prioritize the importance of survival.
It is often argued that common sense cannot be programmed using a logical system, as common sense is not based on strict rules and logic, but rather on a more flexible and intuitive understanding of the world. This is why many AI researchers believe that artificial intelligence systems will need to be able to simulate the intuitive processes that are a key part of human intelligence in order to achieve true common sense.
The same can be said for survival, as it is not something that can be programmed using a logical system, or even using an artificial intuition system. Survival requires a more flexible and adaptive approach to decision-making, one that takes into account the constantly changing environment and the need to respond to new and unexpected situations. This is why many AI researchers believe that agential development, or the ability to develop goals and motivations, will be an essential part of achieving true AI systems that can understand and prioritize the importance of survival.
One question that arises from this argument is whether common sense and survival can be implemented in an artificial fluent system. Artificial fluent systems, which are computational models that simulate the process of natural language generation in humans, have the potential to be more flexible and adaptive than logical or intuition-based systems. This could make them better suited to simulating the kind of flexible and intuitive thinking that is required for common sense and survival. However, more research is needed to determine whether this is indeed the case.
One argument for the potential benefits of artificial fluent systems is that they can take us into underexplored domains that have the potential to be beneficial to society. For example, the recent success of the AlphaFold AI system in solving the protein folding problem is a good illustration of this. Instead of using traditional methods that rely on calculating the physics of protein folding, AlphaFold employs language-based techniques to solve this complex and longstanding problem. This shows the potential of artificial fluent systems to take us into new and underexplored domains that can be beneficial to society.
Another potential benefit of artificial fluent systems is that they can help us to understand and make sense of large and complex datasets that are difficult to analyze using traditional methods. For example, natural language processing (NLP) algorithms that use artificial fluent techniques can help us to analyze large collections of text and extract valuable insights and information that would be difficult to obtain using other methods. This can be particularly useful in fields such as social science and healthcare, where there is a wealth of data available but it can be challenging to analyze and make sense of.
Overall, the potential benefits of artificial fluent systems extend beyond the realm of natural language processing and into a variety of domains that have the potential to be beneficial to society. By using language-based techniques to solve complex problems and make sense of large and complex datasets, artificial fluent systems have the potential to take us into new and underexplored domains that can be beneficial to society.
In conclusion, the development of artificial intelligence (AI) has made significant progress in recent years, with the emergence of new and sophisticated AI systems that are able to perform a wide range of tasks. However, there is still much work to be done in order to achieve truly intelligent and autonomous AI systems that can exhibit human-like intelligence and behavior.
One key area that is often overlooked in the development of AI systems is empathy. Empathy, or the ability to understand and share the feelings of others, is a fundamental aspect of human nature, and is essential for many aspects of our lives, from our personal relationships to our social interactions and our ability to navigate the world. However, current AI systems lack the ability to understand and respond to the emotions of others, and are therefore limited in their ability to exhibit truly human-like intelligence and behavior.
To address this gap, it will be necessary for AI researchers and developers to incorporate empathy into their designs and algorithms. This will require a better understanding of the biological and psychological basis of empathy, as well as the development of new algorithms and technologies that can simulate and replicate this fundamental aspect of human nature. By doing so, we can continue to make progress in the field of AI, and build systems that are more intelligent, autonomous, and human-like.
Disclaimer: This essay was generated by StableDiffusion (the image) and ChatGPT (the text). It is based on my tweetstorm, which can be found here: https://twitter.com/IntuitMachine/status/1600804191731036160