[FRIAM] heuristics for fuzzy objectives (was Re: Drones to detect wildfires)

[uǝlƃ ☤>$]

I suppose this might be central to the difference between narrow and general AI. Hardness can come in multiple forms. And the extent to which the constraints can be traded-off admits softness even in the context of hardness. And the match between conditions and actions isn't always crisp, either. So we not only have [soft|hard]ness of conditions, but also of actions. 

This is the response I always have to the idea of instrumental convergence. And, of course, it evokes my broken record that general intelligence involves a complex of interactive, dynamic control loops of various rates. At last week's salon, we broached the (false) concept of multitasking in humans, particularly that women seem better at it than men. (Full disclosure, 1 of the men is gay. One of the women is decidedly not cis and uses "they" pronouns.) One guy (there's always *that guy*) cited the pretty clear research that there is no such thing as human multitasking. My proposal was that while we all interleave and task-switch, men tend to have a smaller set of objectives than women. So, men may have N things to balance, switching from 1..N; and women have M things, where M >> N. The conversation was chaotic. So I didn't get a chance to also suggest that men's ordering might be more total (first do task 1, then switch to task 2, ...) and women's might be more partial with *side-loaded* preferences influencing the execution.

Viewed this way, the fear of instrumental convergence might be seen as a *masculine* neurosis. So AI Risk might be well-mitigated by handing over control of the world to women. 8^D It's difficult to avoid descending into prurient imagery: https://youtu.be/zTv9AhCuSU4

On 6/1/21 10:46 AM, Marcus Daniels wrote:
> Emergency responders have hard constraints.   Dozers and crews need to be moved into place.  Aircraft need fuel and access to water or fire retardant.    Crew need meals and can only function and survive within certain temperature ranges.   The objectives themselves have a political dimension:  Save multi-million dollar properties vs. public forests vs. native lands.   In this sense the objectives are fuzzy and the whole firefighting effort itself depends on a significant extent to predicting weather conditions accurately.  Some of those weather conditions are even created by a fire.   
> 
> One can cast the constraints as coarse energy levels, but that may wash out the ill-defined objective entirely.

-- 
☤>$ uǝlƃ