Page 41 - Robotics and AI class 10
P. 41
• Attention Allocation: NARS allocates attention to different statements or tasks based on their relevance and
importance. The attention allocation process determines the level of processing resources dedicated to each
task. For example, when faced with a new piece of information, NARS determines how much attention to assign
to it based on its salience and potential impact on the reasoning process.
• Inference and Judgment: NARS performs both deductive and inductive reasoning to draw conclusions from the
available knowledge. Deductive reasoning involves applying logical rules to derive new statements from existing
ones. Inductive reasoning involves generalising from specific instances to form more abstract knowledge. The
system evaluates the plausibility of conclusions through a process called judgment, which assigns a degree of
truth and reliability to each statement.
• Learning and Anticipation: NARS has a learning mechanism called “anticipation and update” that allows it to
acquire new knowledge from experience. When faced with a novel situation, NARS generates anticipations or
predictions about what might happen. If the actual outcome matches the anticipation, the system reinforces its
beliefs. If the outcome contradicts the anticipation, the system updates its knowledge by revising or discarding
certain statements.
For example, let’s consider the following statements in NARS:
Statement 1: All mammals are animals.
Statement 2: Elephants are mammals.
Using deductive reasoning, NARS can infer a new statement:
Statement 3: Elephants are animals.
NARS assigns a degree of truth to Statement 3 based on the evidence provided by Statements 1 and 2, as well
as other knowledge in its system.
Important Note
Both NARS and DL (Deep Learning) are learning systems, though they correspond to very different understandings
of “learning”.
Deep Learning follows the machine learning tradition which gives “learning” an algorithmic definition in the
sense that:
• The resulting system realises a function, i.e., an input-output mapping, by following a (deterministic or
probabilistic) algorithm specially formed for this problem.
• The result is the output of a “learning algorithm” that takes the training data (representative cases of the desired
mapping) as input.
NARS, on the contrary, takes “learning” as adaptation or self-organisation that is non-algorithm, in the sense
that are as follows:
• At no time (either before any training or after significant training) does the system guarantees to work as a fixed
function. Instead, the system’s response to a problem is determined by history and context, so changes over time.
• The learning process happens in many forms in many places, and is lifelong and open-ended, never reaching a
final state or converging to such a state.
New Age Robotic Systems 39

