Embodiment and AI
Topic 1Embodiment. One criticism of cognitive science and its approach to intelligence and mind (PSS orANN),is that it is too abstract; it does not treat the cognitive system as an embodied system. There are different strengths to this claim: almost anyone would agree that a truly adequate account of cognition would include the mind’s connections to the wider world, while some might argue that there is a “core” to cognition that in fact involves information processing that is internal to subjects. Develop a critique of disembodied cognition from the embodied perspective. Three readings are provided under the heading Topic1, Embodiment, in the written assignment section on Courselink. You should address the following three questions in your answer: [1] In what ways are both PSSs and ANNs thought to be too abstract or disembodied?[2] How is embodiment to be defined? Provide an example(there are a number discussed in the readings)as an illustration of some embodiment thesis.[3] Are their aspects of cognition that cannot be dealt with by an embodied approach?(See Wilson article in particular).Note: embodiment means different things to different researchers. Be clear about the meaning of “embodiment” you are discussing Topic 3. Real AI. Almost everyone paying attention to the current “memosphere”has heard about “Big Data” and about “Machine Learning.” The Google Pixel camera is powered by “AI” (aka “machine learning”). In so far as this constitutes a “revolution” it is because we know have access to massivedata sets, and to unsupervised learningalgorithms. The so called “cat paper,” [every AI researcher in the world knows what you mean if you say this]was one of the first demonstrations of this powerful unification of unsupervised learning algorithms and big data sets. Levesque, who refers to big data-machine learning as “AML” thinks there is much of value in this approach, but he does not believe it can replace GOFAI and its focus on common sense. He argues that GOFAI is the approach to take to achieve the “Real AI” mentioned in his title. Read Lévesque, pp45-100. See also the tworeadings provided under the heading for Real AI, in the written assignment section on Courselink. Your answer should address the following questions[1] What is AML and how does it work, in outline?[2] What is the GOFAI approach Levesque favors? [3] Why is GOFAI likely to lead to “real AI” and AML not likely to lead to “real AI”
___________________________
Embodiment
1. In what ways are both PSSs and ANNs thought to be too abstract or disembodied?
Both PSSs and ANNs are thought to be too abstract or disembodied because they do not take into account the body and the environment in which the mind is situated. PSSs are typically implemented as software programs that run on a computer, and ANNs are typically implemented as hardware devices that are trained on large datasets. Neither of these approaches takes into account the fact that the mind is embodied in a physical body and that it interacts with the world through its senses and actions.
2. How is embodiment to be defined? Provide an example(there are a number discussed in the readings)as an illustration of some embodiment thesis.
Embodiment can be defined as the way in which the mind is shaped by the body and the environment. There are a number of different embodiment theses, but one common thesis is that the mind is a sensorimotor system that is constantly interacting with the world through its senses and actions. For example, the way in which we perceive the world is shaped by our bodies and our senses. We see the world from a particular perspective, and we can only see objects that are within our field of vision. Our perception of objects is also shaped by our motor abilities. For example, we can only reach objects that are within our reach.
3. Are their aspects of cognition that cannot be dealt with by an embodied approach?(See Wilson article in particular).
Some aspects of cognition may be difficult or impossible to deal with by an embodied approach. For example, it may be difficult to explain abstract concepts such as numbers and language using an embodied approach. However, it is important to note that embodiment is not the only factor that shapes the mind. The mind is also shaped by culture, education, and experience.
Real AI
1. What is AML and how does it work, in outline?
AML stands for Adaptive Machine Learning. It is a type of machine learning that uses artificial intelligence to improve its own performance. AML systems are able to learn from their mistakes and to adapt to new data. This makes them more powerful than traditional machine learning systems, which are typically trained on a fixed dataset and are not able to improve their performance over time.
AML systems work by using a feedback loop. The system starts by making a prediction about the output of a function. The prediction is then compared to the actual output of the function. If the prediction is wrong, the system learns from its mistake and makes a better prediction the next time. This process continues until the system is able to make accurate predictions.
2. What is the GOFAI approach Levesque favors?
GOFAI stands for Good Old-Fashioned Artificial Intelligence. It is a type of artificial intelligence that is based on symbolic reasoning. GOFAI systems are able to represent knowledge in the form of symbols and to use logic to reason about that knowledge. This makes them capable of solving a wide range of problems, including problems that are difficult or impossible to solve with traditional machine learning systems.
3. Why is GOFAI likely to lead to “real AI” and AML not likely to lead to “real AI”
Levesque argues that GOFAI is likely to lead to “real AI” because it is able to represent and reason about knowledge in a way that is similar to the way that humans do. AML systems, on the other hand, are not able to represent and reason about knowledge in the same way. This means that AML systems are not capable of solving the same kinds of problems that GOFAI systems are capable of solving.
In addition, AML systems are not able to learn from their mistakes in the same way that GOFAI systems can. This means that AML systems are not able to improve their performance over time. As a result, AML systems are not likely to lead to “real AI” in the same way that GOFAI systems are.
