In 1950 Alan Turing published the paper Computing Machinery and Intelligence, in which he wrote:
Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s? If this were then subjected to an appropriate course of education one would obtain the adult brain.
Which is the fundamental idea on how Machine Learning works today: a model with no knowledge about the problem is built, and then data is fed to it in order for it to get more knowledge.
From that, we can reason on how does a baby learn:
- Be multi-modal: use multiple sensory inputs to gather information (vision, sounds, smell);
- Explore by navigating through the environment;
- Try to learn how things work via trial and error;
- Interact with the physical world
- Be social, interact with other agents
- Use natural language.
Embodied AI can be defined as the study of intelligent agents that are able to:
- See, or in a broader sense perceive their surroundings through vision, audition and other senses)
- Talk with natural language
- Listen, comprehend and respond to audio inputs
- Navigate their environment and interact with it
- Reason on the long term consequences of their actions.
Moravec’s Paradox
Easier tasks like grasping an object can be very hard for a robot to do, since it’s complex for a robot to estimate the weight of an object just by its appearance and the prior knowledge.
On the other hand, robots are good at tasks that are difficult to perform by humans, like be very good at chess.
Embodied AI misconceptions
- Embodied AI is NOT just robotics, since it can also include virtual agents, such as chatbots, that interact with the physical world through sensors and other devices;
- Embodied AI is NOT only concerned with movement, but it also involves perception, cognition and decision-making;
- Embodied AI is NOT only for physical tasks, since it can be used to create virtual assistants that help people with digital tasks such as scheduling and email management;
Traditional AI vs Embodied AI
The main difference from Embodied AI and traditional AI is that traditional AI analyzes data that is static, downloaded from a static dataset, and most of the time does not utilizes sensory data to adapt to the environment.
Embodied AI, on the other hand, analyzes data that is taken from the environment, so it’s very dynamic.