Discussion fin de modèle

diff --git a/tex/4-loop.tex b/tex/4-loop.tex
index 650323c631798e6c46c6e9e24bd428b4e9a90f0c..41d461967e3e6617e55cd66187656fb2904bc3fd 100644 (file)
--- a/tex/4-loop.tex
+++ b/tex/4-loop.tex
@@ -1030,9 +1030,9 @@ This is basically the \emph{human-in-the-loop} approach I mentioned above, favor
   % Et donc, comment faire ? Certain le font déjà (robotique comportementale basée sur la cognition, comme par exemple l’équipe FLOWERS à Bordeaux), mais comment inclure ça dans les systèmes interactifs ? Est-ce que ton modèle “seven stages of reaction” permettrait de mieux prendre ça en compte, ou le prendre en compte dans le système, etc. ?
   % Et quelles perspectives tu ouvres ?
 
-The idea here is that interaction is not separate loops on the user and system side.
+The idea here is that interaction is not separate loops on the user and system sides.
 Once connected, these loops form a Human-System loop that cycles between the user and the system.
-Therefore, a smooth interaction between a user and an interactive system requires efficient connections between them.
+Therefore, smooth interaction between a user and an interactive system requires efficient connections between them.
 Here we focus on computer-as-a-tool systems in which the users have the initiative.
 %The user usually has the initiative, it is the case for computer-as-a-tool systems.
 Hence, the system must not be a limiting factor to the user's sensorimotor loop, it must react in \defword{real-time}.
@@ -1041,12 +1041,26 @@ The visual system requires a \qty{100}{\hertz} loop.
 The touch system requires a \qty{1000}{\hertz} loop.
 The audio system requires a \qty{40000}{\hertz} loop.
 We know that breaking this loop affects perception.
-For example a disruption of the human visual input stream prevents people from seeing changes in their visual field~\cite{rensink97}.
-At the opposite, as we discussed in \refchap{chap:output}, force-feedback devices use force models computed over \qty{1000}{\hertz}, but the force model to be applied can be updated at a lower frequency without affecting the perception of a shape.
-Hence, the fine characterization of the bottlenecks in this Human-System loop is necessary to both avoid undesirable effect.
+For example, a disruption of the human visual input stream prevents people from seeing changes in their visual field~\cite{rensink97}.
+On the opposite, as we discussed in \refchap{chap:output}, force-feedback devices use force models computed over \qty{1000}{\hertz}, but the force model to be applied can be updated at a lower frequency without affecting the perception of a shape.
+Hence, the fine characterization of the bottlenecks in this Human-System loop is necessary to avoid undesirable effects.
 But we can also leverage the limits of our sensorimotor loop to create illusions to extend interaction.
 More generally, combining the user and the system at the same level with complementary roles facilitates the holistic approach of my research.
 
+The similarity of the user and the system in this model can give the wrong impression that humans and systems have the same capacities.
+Indeed, this is not the case.
+Humans are good at adapting to many situations, they have intuition, and they have high perceptual and motor skills.
+On the opposite, systems are good with repetitive tasks, precise calculations, and automation in general.
+In the Artificial Intelligence community, this is known as Moravec's paradox which states that contrary to general assumption, sensorimotor behavior is much more complex to implement than reasoning and computation \cite{moravec88}.
+This statement should not be surprising after reading this manuscript.
+My analysis of this paradox is that first, the assumption was a clear underestimation of the complexity of perception.
+Second, the inductive nature of systems and the co-inductive nature of humans are the core of this difference.
+Systems are made to execute algorithms with high efficiency.
+On the opposite, the human sensorimotor loop enables humans to explore, try, analyze, and therefore understand their environment.
+Research started to apply this principle of \emph{curiosity} as a building block of a robot behavior\cite{laversannefinot18}.
+Thanks to a kind of sensorimotor loop, a robotic arm can manipulate objects without explicit instructions on how to actually do that.
+Maybe if we manage to build systems that can adapt their behavior to the users, we will stop considering that humans must use their adaptation capacity to compensate for system failures.
+
 %User model and system model. Both models include how they behave and how they perceive its environment, which includes the other (resp. interactive system and user). 
 %We study the system with rational methods, but we study the user and the interaction of both with empirical methods.