12.12.2012 Views

Who Needs Emotions? The Brain Meets the Robot

Who Needs Emotions? The Brain Meets the Robot

Who Needs Emotions? The Brain Meets the Robot

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

architectural basis of affect 239<br />

response to a perceived threat, in a robot whose processing speeds are so great<br />

that it needs no alarm mechanism. It is arguable, <strong>the</strong>n, that only linguistic expression<br />

(see Arbib, Chapter 12) is capable of conveying <strong>the</strong> vast majority of<br />

tertiary emotions, whereas most current research on detecting emotions focuses<br />

on such “peripheral” phenomena as facial expression, posture, and o<strong>the</strong>r<br />

easily measurable physiological states.<br />

THE NEXT STEPS<br />

<strong>Emotions</strong> in <strong>the</strong> sense we have defined <strong>the</strong>m are present in many control<br />

systems, where parts of <strong>the</strong> control mechanism can detect abnormal states<br />

and react to <strong>the</strong>m (causing a change in <strong>the</strong> normal processing of <strong>the</strong> control<br />

system, ei<strong>the</strong>r directly through interruption of <strong>the</strong> current processing or<br />

dispositionally through modification of processing parameters). <strong>Emotions</strong><br />

thus defined are not intrinsically connected to living creatures, nor are <strong>the</strong>y<br />

dependent on biological mechanisms; e.g., operating systems running on<br />

standard computers have several emotions in our technical sense, although<br />

<strong>the</strong>y lack many of <strong>the</strong> detailed features of <strong>the</strong> sorts of emotion to which our<br />

folk concepts are applied.<br />

What is special about at least a subset of emotions so defined (compared<br />

to o<strong>the</strong>r non-emotional control states) is that <strong>the</strong>y (1) form a class of useful<br />

control states that (2) are likely to evolve in certain resource-constrained<br />

environments and, hence, (3) may also prove useful for certain AI applications<br />

(e.g., robots that have only limited processing resources, which impose<br />

severe constraints on <strong>the</strong> kinds of control mechanism that can be implemented<br />

on <strong>the</strong>m).<br />

Useful affective control mechanisms are likely to evolve if <strong>the</strong>re are many<br />

evolutionary trajectories that, given various sets of well-specified initial conditions<br />

and fitness functions, will lead to those control systems (e.g., Scheutz,<br />

2001; Scheutz & Schermerhorn, 2002). A subset of those will be control<br />

mechanisms that can produce emotional states suited to coping with emergencies<br />

or unexpected situations as <strong>the</strong>y occur in dynamic, unpredictable,<br />

real-world environments.<br />

It is not yet clear which of <strong>the</strong> more subtle and complex long-term<br />

emotional states, such as grief, ambition, jealousy, infatuation, and obsession<br />

with a difficult problem, are merely side effects of desirable mechanisms and<br />

which are states that can be intrinsically useful in relation to ei<strong>the</strong>r <strong>the</strong> needs<br />

of individuals or <strong>the</strong> needs of a social group or species. Human aberrations<br />

make it clear, however, that machines containing useful mechanisms are<br />

capable of getting into highly dysfunctional states through <strong>the</strong> interactions<br />

of those mechanisms. As machines become more human-like, we can expect

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!