5. Satisficing (Suffice + Satisfactory)
"decision makers can satisfice
either by finding optimum solutions
for a simplified world, or by finding
satisfactory solutions for a more
realistic world.
Neither approach, in general,
dominates the other, and both have
continued to co-exist in the world of
management science."
https://www.wikiberal.org/wiki/Herbert_A._Simon
Simon, 1979
7. AI・MLの仕事は経営の意思決定の仕事
Jobs of AI/ML are decision making in management.
Kazuo Yano, Hitachi
Cassie Kozyrkov, Google
Data are beautiful, but it s decisions that are
important.
It s through our decisions ̶ our actions ̶ that we
affect the world around us.
https://towardsdatascience.com/introduction-to-decision-intelligence-5d147ddab767
10. Homo Economicus
• Rational
• Unbounded Capacity
• No Cognitive Bias
• Only Self Interest
• No Interest in Others https://mises.org/blog/homo-economicus-straw-man
12. Rational, Optimal & Adaptive
• Rational
• optimal for a purpose of achieving a certain specific goal.
• Optimal
• maximally efficient for achieving a certain specific goal
• Adaptive
• compatible and effective (to a certain extent) for a
purpose of achieving a certain specific goal.
13.
14.
15.
16. ol and Grit
Goal
Goal Goal Goal
GoalGoalGoal GoalGoal Goal
Action Action Action Action Action Action Action
Fig. 1. Hierarchical goal framework. Goals are typically organized hierarchically, with fewer high-level
goals and more numerous low-level goals; the latter are associated with action tendencies, here broadly
construed to include attention, emotion, and behavior.
Duckworth & Gross, 2014
18. Summary
• Goals are important for any types of decision
making.
• Without goal, rationality, optimality, and
adaptiveness cannot exist.
• It is important to set right set of goals for both
biological and artificial systems.
• Humans are good at setting (hierarchical) goals.
21. Skill, Habit and Control
Initial state Estimates drift with noise Heading estimate updated
Distance vector updated
Lane error grows
Speed error grows
A
B
5
4
3
2
1
0
–1
–2
–3
2.0
1.5
–1.5
0.5
–0.5
1.0
–1.0
0.0
SpeedFollow
3
2
ble and un-
odules.
Follow mod-
e necessary
auses them
ward uncer-
w module is
mproves its
odules’ state
ogression of
odules: con-
er following
nes indicate
odule’s rele-
hus, for the
ts the car’s
depicts the
he lead car,
the angle to
ates overlap
s low uncer-
iverge, mak-
uncertainty.
mate updated
Distance vector updated
Lane error grows
Speed error grows
Hayhoe & Ballard, 2014
22. Hierarchical Reinforcement
Learning
A CB
An illustration of how options can facilitate search. (A) A search tree with arrows indicating the pathway to a goal state. A specific sequence
ndently selected actions is required to reach the goal. (B) The same tree and trajectory, the colors indicating that the first four and the la
have been aggregated into options. Here, the goal state is reached after only two independent choices (selection of the options). (C) Illustr
using option models, which allow the ultimate consequences of an option to be forecast without requiring consideration of the lower-level st
be involved in executing the option. (For interpretation of the references to color in this figure legend, the reader is referred to the web versio
)
M.M. Botvinick et al. / Cognition 113 (2009) 262–280
Botvinick & Niv et al., 2009
24. Value and
opportunity cost in a
current environment
Value and opportunity
cost in another
environment
Kolling et al., Science, 2012
Hayden et al., Nature Neuroscience, 2011
Kolling et al., Neuron, 2014
Wittmann, Kolling, Akaishi et al., Nature Communications, 2016
25. Summary
• Everyday behaviors like driving a car consist of
individual actions and a chunk of these individual
actions.
• The behavioral hierarchy is critical to organize our
actions.
• Recent developments in decision neuroscience,
hierarchical reinforcement learning and foraging
decision, capture this hierarchical organization of
human behavior.
27. • Dunbar number suggests that
a single person can have
100-200 people of personal
connections in her/his social
network.
• Yet, our society consists of
much larger number of people.
• There has to be something in
human cognition that can
overcome the limitation of
Dunbar number.
Dunbar
Harari
28. • Cooperative relationships and social norms can be created by the
human beings living in groups through the means of punishment
and establishing trust.
Hardin
Ostrom
Yamagishi (山岸)
• The ideal decision makers, Homo Economicus, creates the situation
‘Trajedy of Commons’.
• If this is true for real human beings, a human community cannot
survive by sharing common resources.
• But actual human beings and their community (at least for those
who have actually survived) can share goals and run organizations.
29. George Mason
They (western states) will have
the same pride, and other
passions, which we (eastern
states) have; and will either not
unite with, or will speedily revolt
from, the Union, if they are not
in all respects placed on an
equal footing with their
brethren.
—at 1787 constitutional
convention.
30.
31. Inequality Aversion
• One player, the dictator, has a sum
of money which he can allocate
between himself and another
player, the recipient.
• The Dictator game measures a
positive concern for the recipient s
material payoff that is independent
of the recipient s behavior, because
the recipient has no actions to
take.
• Dictator allocations are found to be
a mixture of 50% offers and 0%
offers, and a few offers in between.
left to right, arcs represent the possible actions that the
player could take, and dotted lines represent one exam-
ple choice. The numbers represent the material payoffs
for the players and they are color coded (along with the
actions) to match the players. For example, in the depic-
tion of the Dictator game (see Figure Box 11.1), the dicta-
tor (red) has $10 and can send any of that $10 to the
recipient (blue). In this particular example, the dictator
sends $3 to the recipient, keeping $7 for himself.
Fischbacher, 2004). In Figure Box 11.3, we assume that
the third party has an endowment of $5 and for every
dollar spent on punishment the dictator loses a dollar.
In the example, the third party spends $3 on punish-
ment, reducing his payoff from $5 to $2 and reducing
Send 3
Dictator Recipient
7, 3
10, 0
0, 10
FIGURE BOX 11.1 Example of a Dictator game.
Offer 3
Proposer
Offer 0
Offer 10
Responder
Accept
Reject
7,
0,
FIGURE BOX 11.2 Example of an Ultimatum game.
NEUROECONOMICS
Fehr
5
Ui(x) = xi - αi
1
n −1
max
j ≠ i
∑ {xj - xi,0} -βi
1
n − 1
max
j ≠ i
∑ {xi - xj,0} ,
we assume βi ≤ αi and 0 ≤ βi < 1. In the two-player case (1) simplifies to
Fehr & Schmidt, 1999
32. Eat Like Locals (Social Influence)
https://www.youtube.com/
watch?v=FldlObA4Cdw
https://www.youtube.com/
watch?v=sMZR-YGz_Gc
33.
34. Envy & Schadenfreude
• Schadenfreude: a positive
emotional state in the face of
someone else s misfortune
• Envy: a negative emotional state
in the face of another s fortune
http://comments.bmartin.cc/2014/11/12/feeling-
good-when-others-suffer/
42. Summary
• There are increasing demands for understanding humans
in the context of social networks.
• Humans usually do no like inequality. Successful
civilizations have considered this value systems of
humans well.
• Trust is the major issue in both industries and societies
globally.
• But the nature of the issues regarding trust may differ
across different cultures and countries.