2017 Kanai The Case for Making Artificial Intelligence Conscious
<p>Jun 16, 2017 04:01 PM... in nautilus</p>
<p>
Ryota Kanai...</p>
<p>
inscrutability... and paras that follow that. in AlphaGo section..</p>
<p>
Propublica investigation: Florida... black defendants prone to recidivism by algo.</p>
<p>
and Euro Union demanding ... legal "right to explanation: ... why an AI made its decision.</p>
<p>
metacognition: confidence in decision.</p>
<p>
Philosophers: metacognition is the essence of consc...no... crap!</p>
<p>
but... when we know something... we know that we know it...</p>
<p>
RK and colleagues trying to build metacognition into NNs....
(interesting but fruitless approach)</p>
<p>
Endel Tulving: mental time travel...</p>
<p>
conscious when predicting consequences of actions or planning for future.</p>
<p>
think about waving hand in front of face or going to kitchen to make coffee.</p>
<p>
Agnosia... patients can't do reaching task in face of time delay......</p>
<p>
Pavlov conditioning only works with two stimuli overlap in time.</p>
<p>
if separated in time... must be conscious of the relationship.</p>
<p>
Awareness is needed to retain the memory of the stim after it stopped.</p>
<p>
Function of Conscious.... broaden our temporal window......
give present moment extended duration.</p>
<p>
Brain keeps gen. sensory repr even when there is NO direct sensory input.</p>
<p>
Common thread is counterfactual info generation. * (this is good!)</p>
<p>
memory of the past or predictions for unexecuted future actions...</p>
<p>
active process of hypothesis creation and testing...</p>
<p>
counterfactual... can perform non-reflexive behavior... eg waiting 3 seconds before acting.</p>
<p>
need to have an internal model that has learned the world.</p>
<p>
Models can be used for reasoning, motor control and mental sim.</p>
<p>
Introspection and imagination... 2 ingredients of consc...</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>