External interactions are relevant outputs in order to compare animal, human and artificial intelligences.
Efficiency of messages and actions of a given intelligent being can be checked.
We consider these outputs now.
The main elements of a closed loop system are:
- Reference input
- Comparator
- Control elements and actuators leading to output
- Feedback elements
An intelligent being is constantly comparing sensor signals to reference inputs. Reference inputs for example give limits on cold, warm, noise, hunger, generally: pain thresholds. A set of sensor signals can be considered as a point in a multidimensional space. In this space there are areas to be avoided. Other areas are to be reached.
Reference inputs are characterized in the same multidimensional space by these "good" and "bad" areas. Most of sensor signal space belongs to neutral areas (neither good nor bad ones). Most of sensors are neutral themselves.
The comparator output gives control elements a gradient to be followed in order to reach certain areas or in order to avoid other ones.
Basically, there is always a wired response to a strong signal located in a "painful" area. Basically, all intelligent beings can learn more or less a customized response from their own experience.
Less evolved intelligent beings give a "wired" response to each gradient.
More evolved intelligent beings can learn more and give learned responses.
The control elements command actuators. They interact with the universe of the given intelligent being. This intelligent being perceive, at least from time to time, a part of this interaction thanks to its sensors.
The feedback loop is closed.
That's all for the main feedback loop in a self-learning AI.
Still to come:
What about stability, positive or negative feedback?
What about self-learning?
No comments:
Post a Comment