Generally, people can be classified as aggressive (Type A) or passive. Could the programming of AI systems cause aggressive or passive behavior in those AIs?
$\begingroup$
$\endgroup$
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
1
The Wikipedia entry on this personality theory says of Type A people:
The theory describes Type A individuals as ambitious, rigidly organized, highly status-conscious, sensitive, impatient, anxious, proactive, and concerned with time management. People with Type A personalities are often high-achieving "workaholics."
All of those attributes could conceivably be explicitly programmed in. Alternatively, most of them could arise from a basic goal of performing a certain task as efficiently as possible. After all, if you really want to carry out a task, you're going to get organized, you'll only do other things if they're asked of you by someone important, you won't want to get bogged down in irrelevant things, you'll actively pursue the necessary resources, and you'll want to use time as effectively as possible.
Note that this applies only to strong AIs, since weak AIs like image recognizers don't generally have personalities that we can interact with.
Now, just for fun, let's consider an overly aggressive personality, to the point of a disorder.
This Counselling Resource page seems helpful in describing what an aggressive person does. The page includes a bulleted list of common characteristics, which I distill into the following:
- They attempt to gain dominance and control
- They oppose to anything that places limits on them
- They take advantage of others to further their own goals
- They hide information from whose who would oppose them
- They rarely decide to stop pursuing their desires (even impulses)
This all seems like a characterization of an AI designed to be the best at its task: the best out of any other agent, and the best it by itself could possibly be. Ruthless pursuit of the highest performance would involve taking control of all relevant resources (including other agents), demolishing barriers to the goal, thwarting those who would interfere with progress, and carrying out each possibly-useful idea/desire to completion.
In summary, yes, an AI's behavior and personality are programmable, either explicitly or through some kind of emergence.
-
$\begingroup$ This is why I believe more work needs to be done to address the control problem. $\endgroup$ Commented Jan 7, 2018 at 6:02
$\begingroup$
$\endgroup$
As can be observed in the real world with creatures such as fighting fish, such things are possible even in very simple spatially-embedded systems. All one needs is the notion of 'territorial radius', i.e. the amount of 'personal space' that an entity need to be comfortable. Giving individuals in a species even slightly different values for this radius gives rise to different observable behaviours, which one might choose to label as 'aggressive' or 'passive'.
See the fantastic book 'Vehicles' by Valentino Braitenberg for an explanation of how natural it is to ascribe complex behaviours to simple mechanisms.
