Challenges - MindRACES

Mind RACES:
some Emerging Challenges
Cristiano Castelfranchi
Institute of Cognitive Sciences and Technologies
National Research Council of Italy
MindRACES, First Review
Meeting, Lund, 11/01/2006
1
Some EMERGING CHALLENGES
Let’s put aside
• ATTENTION
• LEVELS of ACTION CONTROL
2
MindRACES, First Review Meeting, Lund, 11/01/2006
Some EMERGING CHALLENGES
for example: 1
A lot of Emotional responses are related
a) to expected events or outcomes or to
b) the match or mismatch between the anticipated
event or value and what is achieved or happens
>> Does the Agent get more or less than expected? (IST)
>> Is the Agent expecting a harm or danger ?
>> Is the Agent realizing a lack of knowledge
and high uncertainty ? (ISTC)
Specific Emotional states or responses
are related to these Anticipatory situations
3
MindRACES, First Review Meeting, Lund, 11/01/2006
Some EMERGING CHALLENGES
for example: 1
Emotional responses
>> will we be able to model and implement some of these
responses and their adaptive functionality ?
which is the use of fear, joy, anxiety,
felt frustration, relief, etc. ?
>> will we be able to integrate some of these functions
with other anticipatory mechanisms like decision, planning,
attention, etc. ?
4
MindRACES, First Review Meeting, Lund, 11/01/2006
Some EMERGING CHALLENGES
for example: 2
Bases & Mechanisms for PREDICTION
>> can we model and implement other kinds of predictions
beyond the fundamental statistical learning,
probability, or analogical reasoning ?
- Predictions based on Laws and Norms?
- Predictions based on the Ascription
of goals and beliefs to the other Agent?
>> Aren’t these functions and their integration
fundamental for interaction & coordination with other Agents
and humans ?
5
MindRACES, First Review Meeting, Lund, 11/01/2006
Some EMERGING CHALLENGES
for example: 2
Relationship between
sensory-motor expectations controlling the action and
Symbolic/propositional predictions and goals
No Goal can be matched against the world if not “translated”
into sensory expectations or signs
6
MindRACES, First Review Meeting, Lund, 11/01/2006
Some EMERGING CHALLENGES
for example: 3
SURPRISE
>> WHAT IS ‘SURPRISE’?
>> WHAT IS SURPRISE FOR?
>> ARE THERE DIFFERENT KINDS of SURPRISE?
ONLY ORGANISMS ABLE TO PRODUCE
ANTICIPATORY REPRESENTATIONS
CAN BE ‘SURPRISED’
7
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
'Surprise' is the automatic reaction to a
mismatch. It is:
- a (felt) reaction/response;
- of alert and arousal;
- due to an inconsistency (mismatch, non-assimilation, lack of integration)
between incoming information and our previous knowledge, in particular an
actual prediction or a potential prediction (error in prediction);
- invoking and mobilizing resources at disposal of an activity for a better
epistemic processing of this 'strange' information (attention, search, belief
revision, etc.), but also for coping with the potential 'threat';
- aimed at solving the inconsistency;
- and at preventing possible dangers (the reason for the alarm) due to a lack
of predictability and to a wrong anticipation.
8
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
Not just a ‘degree’ on a continuum.
Not simply matter of ‘intensity’
No
surprise
0
1
Maximum
surprise
Different Kinds and Levels of Surprise
with different ‘qualities’ and ‘transitions’
9
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
“First-hand surprise” the most peripheral one,
just due to perceptual mismatch based on sensory-motor expectations
Deeper and slower forms of surprise
due to symbolic representations of expected events, and to the process of
information integration with previous long-term knowledge and explanation
of the perceived data (Meyer et al. 1997; Reisenzein 2000).
This is surprise due to implausibility, un-believability
of the new information
10
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
i) Firsthand Surprise,
or Surprise in Perception/Sensing or
Active-Prediction-based Surprise,
or Mismatch-based Surprise
I'm actively checking whether P or not P but with the forecast that P, that
is, I have an endogenous anticipatory representation of the incoming
input and I attempt to match the stimulus against it. If there is a
mismatch between the two (sensory) representations there is
surprise:
11
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
>>The more sure I was about the more surprised I am
Intensity of Surprise = Function (Bel%)
But also
>>The larger the mismatch the more surprised I am (??)
Surprise based on the subjective probability of the event (in a given context)
-”information theory”
- other cognitive approaches
12
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
Moreover, if I was also concerned (as usually is for leaving systems)
that is if there was
not just a forecast but an "expectation"
then I will be more surprised,
meaning that my response/reaction will be more intense.
My concern affects the surprise
which is not only an "epistemic" reaction
>>The more important the Goal
the more Surprise I will be
13
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
>>The more important the Goal the more Surprised I will be
better:
the more intense the response
>>Not fully independent dimensions
The desire/wish/need and its value
can impact on the belief (prediction)
While
The certainty of the prediction (belief)
can impact on the value of the goal
When there is an expectation not simply a forecast, when the Ag is ‘concerned’
there will be also specific and richer affective reactions
14
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
ii) Passive Prediction-based Surprise
Passive expectation:
I have formulated an explicit prediction but it is not actively used for testing
the event, I'm not checking whether, exploring.
The information simply arrives (I was not specifically searching for it) but its
consistency should be tested in anyway,
first of all, with anticipatory representation in the background, off attention.
>> Attention is moved on this discrepancy.
Intensity laws look similar to the previous ones.
In both cases there is this additional principle:
15
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
Additional principle:
>> Two kinds of ‘unexpectedness’
Given the 'belief' (Prediction) about a future event P when P really
happens we can have
3 different cases:
1. P & prediction that P ==> No surprise
2. P & No prediction that P (& No prediction that Not P) ==> weak surprise
3. P & prediction that Not P ==> strong surprise
16
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE?
iii) Surprise in Integration or Implausibility-based Surprise
”I wouldn't have expected that..." (if, on the basis of my beliefs, by
means of some inferential processes, I would have produced an anticipation I wouldn't
have anticipated this).
Incredulity comes later from the deep control, the attempt to integrate with previous
knowledge, not from a mismatch; and not necessarily is preceded by a mismatchsurprise.
Prediction:
The more "incredible" the more "surprising"
“Incredible” means that from my previous integrated-coherent believed
beliefs I would derive(Not P) the opposite of the candidate belief that P.
The stronger the inferred opponent (Bel% Not P) the more incredible P for me, and thus the
more surprising.
17
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE FOR ?
several functions
the immediate response and short-term functions
 It is not only related to redirecting attention on the mismatching facts.
 Signaling the crisis of a automatic, unconscious, routine process and to
go back to higher level intentional, conscious processing.
 Discovering that we were “ignorant”: “careful! there is gap of
knowledge”
 Concentrating cognitive processing resources on facts,
- trying to better interpreting them etc. (epistemic reaction).
 ‘Expressive’ response
 Activating resources for possible practical (non-epistemic) activity;
physical arousal, bodily preparation for fast reaction.
18
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE FOR ?
long term effects and functions
of the perceived surprise due to a bad prediction;
for example:
Revising that prediction, reducing its strength (a learning
feedback);
Becoming more cautious next time in the same
circumstances; less confident; increasing controls
before and during the actions;
Reducing self-confidence as predictor; acquire more
evidences, be careful in reasoning, etc.
19
MindRACES, First Review Meeting, Lund, 11/01/2006
WHAT IS SURPRISE FOR ?
A very relevant function is the other way around:
 The lack of surprise, the confirmation on predictions and
expectations leans to reduce control, attention, packing in
automatic execution
 Eliciting emotions: surprise  appraisal of  emotion: fear,
disappointment, relief, exultance, …
Surprise-based Emotions
 Impact on memory? The surprising events are better
memorized? and more accessible?
Others??
20
MindRACES, First Review Meeting, Lund, 11/01/2006
Some EMERGING CHALLENGES
for example: 4
From MindRaces discussion and from AAAI ws (Deb Roy)
What is the use of “mental” simulation ?
- Learning by simulating
- Anticipating input and comparing
- for learning
- for adjusting the action
- for surprise and emotional responses
- “Mentally” solving problems
- …
21
MindRACES, First Review Meeting, Lund, 11/01/2006
Some EMERGING CHALLENGES
for example: 3
Curiosity
& Epistemic Activity
- beyond Attention
- or being Coutious
22
MindRACES, First Review Meeting, Lund, 11/01/2006