Using Teaching Learning Machines (TLMs) and *Self-learning meta-learners teach themselves to teach* as conditioning priors, I prompted gpt3 with the following:

Embodied and Situated: AI with Feelings

(note: make sure to emphasize that AI doesn’t necesarily have to be embodied in a physical world; it just needs to be embodied in some world)

Here’s what I got:

One of the most exciting recent developments in AI is embodied learning – an approach that emphasizes the importance of giving agents a body and placing them in some sort of environment. The rationale behind this is simple: if we want machines to be truly intelligent, they need to experience the world for themselves instead of being confined to dusty old textbooks (or clean new ones). They need to feel pain and pleasure, see things with their own eyes, touch things with their own hands. Only then will they be able develop common sense – an understanding of how the world works that goes beyond what can be explicitly taught or programmed into them.

Embodied learning has already yielded some impressive results. In one famous experiment, scientists placed robots in a room full of random objects and tasked them with building towers out of blocks. The only information they were given was where each object was located; everything else had to be learned by trial and error. And yet within just a few minutes, the robots had developed surprisingly sophisticated skills like tower-building! By experiencing life for themselves – by making mistakes and figuring out which actions lead to positive outcomes – these little machines were ableto build up a storehouseof knowledge aboutthe physical worldand solve problems far more efficiently than any hand-coded algorithm could hope for..

There are two important aspects to embodied learning that are worth highlighting. The first is its focus on situated cognition – the idea that intelligence arises from interaction with the environment. This contrasts with so-called “symbolic” approaches to AI, which treat knowledge as a static set of symbols (e.g., 0’s and 1’s) that can be manipulated without any need for grounding in reality. In many ways, Situated Cognition is just common sense: if you want to understand how something works, it helps to have firsthand experience of it instead of reading about it in a book.

The second aspect is its emphasis on embodiment – giving agents bodies and placing them in rich environments where they can interact with their surroundings.. It might seem like a small change but it makes a huge difference: embedded in physical reality, agents are no longer limited by what we can explicitly tell them; instead, they must learn for themselves through trial and error.. What starts out as noisy data gradually becomes refined into useful insights about how the world works.. And this processof induction– forming general principles from specific observations–is at the heart of human intelligence..

One of the great things about embodied learning is that it’s not just for robots; it can be applied to any sort of agent, including software agents. As long as an agent has some way of interacting with its environment – e.g., by receiving input from sensors and taking actions in response – it can benefit from this approach.. This means we can start to build truly intelligent machines without having to wait for years or even decades for advances in robotics technology.. All we need is a well-designed simulation environment in which our agents can learn by doing..

Embodied learning is still in its early days but it’s already yielding some impressive results.. In one famous experiment, scientists placed robots in a room full of random objects and tasked them with building towers out of blocks. The only information they were given was where each object was located; everything else had to be learned by trial and error. And yet within just a few minutes, the robots had developed surprisingly sophisticated skills like tower-building! By experiencing life for themselves – by making mistakes and figuring out which actions lead to positive outcomes – these little machines were ableto build up a storehouseof knowledge aboutthe physical worldand solve problems far more efficiently than any hand-coded algorithm could hope for…

The benefits of embodied learning are not just limited to robots; it can be applied to any sort of agent, including software agents. As long as an agent has some way of interacting with its environment – e.g., by receiving input from sensors and taking actions in response – it can benefit from this approach.. This means we can start to build truly intelligent machines without having to wait for years or even decades for advances in robotics technology.. All we need is a well-designed simulation environment in which our agents can learn by doing…