[Robotgroup] Fwd: Robot Questions

[Edwin Wise]

On 28 Jul 2007, at 9:17 PM, Eric Lundquist wrote:

> I was approached by a person writing an article about creating a  
> "Robot
> Army".   Anyone who wants to respond can send their responses to me  
> and I
> will consolidate all the answers.  I can keep your responses  
> anonymous if
> you want.

Oh dear, this article is likely to be terrible, given the nature of  
the questions.

But hey!  Perhaps interesting answers will make his article  
interesting.  No time like the present to inform those who would  
speak to the populace!

> I am writing an article on the practicality of creating a robot army.

Aha!  You have discovered my secret plan!  An army of ruthless,  
soulless, mechanical creatures, beholden to my every command, to  
march over every obstacle in my quest for world conquest.  And they  
called me mad!  All it will take is solving one or two tiny little  
problems...  mentioned within.

> Q: Recent developments in the world of robotics seem to suggest that
> the dream of a privately owned robot army may be closer to reality
> than ever before. Do you think that such a thing is possible? How much
> would it cost?

No. A lot of money. Way too much... far cheaper to use humans.   
Because really, there are two aspects to consider; cost and  
effectiveness.  Humans are both cheap and resourceful, especially  
when their lives are on the line.  And they are self replicating,  
though they _are_ working on the tools needed to make self- 
replicating robots.  But they are far from reality.

> Q: Robots such as the Packbot aren't truly autonomous. How close are
> we to being able to field a robot that can execute simple orders with
> a limited amount of feedback from its human masters?

Far far away indeed.  We can barely get robots to navigate across a  
well-mapped desert along a well-planned path.  To execute even a  
simple order like "go to Town Lake and sink the blue boat" would take  
skills in AI that we've not seen yet.  The classic example that  
computer geeks use to embarrass the layfolk is to get them to  
describe, in precise detail, exactly the steps it takes to create a  
peanut butter sandwich.  Ah! A simple task!  A child could do it!   
Try it sometime.  Write down the steps.  Send them to me.  It will  
be ... illuminating.

The task of AI, then, is to convert common-sense instructions into  
detailed plans of action.  Alas, the depths of the human mind are  
poorly understood (though we are getting closer!  Always closer!), so  
getting the artificial mind to understand the directives of the  
natural mind is a tricky task indeed.

> Q: Until recently, the cosmetic aspect of robot design seemed to be
> largely ignored. A robot army should be a terrifying sight.  Are there
> any designs that you have seen that show the kind of flair that the
> underlying technology deserves?

Oh yes, fearsome we can do; there are all kinds of artists and  
special effects people who have envisioned some mighty fine robots.   
Poke around the Robot Wars, Battle Bots, and similar arenas... or  
look up the work of Mark Setrakian (umm, if I remembered that right,  
it's late).  Fearsome looking is easy.  Fearsome acting is harder.   
There's a reason there are people on remote controls behind almost  
all robots.

> Q: Tracked and multilegged robots have significant advantages but the
> holy grail has always been bipedal locomotion. Despite the huge
> advances that have been made there are still no military or
> paramilitary robots that we are aware of. Do you think a reliable
> bipedal robot could be designed specifically for use in the
> battlefield?

Bipedal robots suffer from small size; it's easier to put a big robot  
on tracks.  Legs are also complicated, difficult mechanisms; you  
really only want a human-shaped robot if said robot is going to  
navigate human-shaped environments (or to be familiar and comfortable  
shapes when dealing with humans).  That is why NASA has a lovely  
humanoid robot they are working on for space; and that is why Japan  
is keen on human-shaped robots as caretaker robots for their aging  
population.

No, you want to have something like Bolo, smart war tanks; those  
would be very useful.  Fiction, of course, has covered this territory  
over the last few generations.

But for war, no!  Not even humans work in human-shaped packages  
during war; we put on wings and fly, or blades and hover; we don  
armored vehicles with giant cannons in them and rumble through the  
landscape, terrorizing all we see.  Only when it comes down to the  
small; down to the personal touch of invading human spaces to find  
individual humans, do our troops take to their two legs and enter the  
arena.  Certainly, bipedal robots would be handy strolling through a  
living room and ordering the inhabitants to lay down their arms, but  
a tracked vehicle can do the same thing more cheaply and more  
reliably -- both valuable criteria.  Think "Number Five" from Short  
Circuit, or that horrible RoboCop machine from episode, umm, whichever.

One of the most serious reasons you want a larger robot, with tracks,  
is power use.  Power supply is the biggest non-AI problem in robots;  
a small robot just can't go far or do much, because it has a low  
energy density.  Drop a diesel engine in, though, and you are talking  
fun... and longevity.

> Q: You can't really discuss artificial intelligence without mentioning
> Azimov's laws of robotics. What would be the simplest way to modify
> these laws to allow for the creation of a design that can be sent into
> battle against human opponents and will obey their master without
> question?

Asimov's laws are crap and nonsense.  There. I've said it.  Total  
bullshit.  I know, I'm being heretical here, but really, I hate  
those.  Don't mention them except to ridicule them.  Implementing ANY  
of the laws would take AI technology far far behind what will be  
released in the field at first.  In Asimov's books, they created  
analogs to the human brain, with full human understanding, and  
through MAGIC they baked in these totally abstract laws that are, in  
fact, nearly impossible to code into a real AI.  It's just not going  
to happen.

Think about it; you have a machine in hand that, through some amazing  
breakthrough in computing, power sources, and technology, is able to  
understand reasonably concrete and clear commands about which  
direction to go and which color uniforms to shoot at, but only when  
expressed in very grounded, physical terms... how are you going to  
put in a controlling overlay of abstract thought, such as, "thou  
shalt not, through action or inaction, allow non-enemy humans to come  
to harm."  Inaction?  How on earth is the robot going to have the  
abstract thinking and predictive capabilities to work that out?  It's  
really darned lucky to be able to recognize the difference between a  
door and a shadow on the wall.  Really.

The robot will acknowledge whomever has the IFF code and/or secret  
handshake, and it will ignore everyone else... except, perhaps, for  
the Chinese hackers or that really really bored 15-year-old in his  
basement.


> Q: Battle ready robots like Lockheed's MULE are prohibitively
> expensive. The designers of the Roomba have recently teamed up with
> Taser to manufacture nonlethal units for local law enforcement that
> seem like they would be more reasonably priced. Do you think it would
> be possible to add weapons to an inexpensive system like a Roomba or
> Robosapien?

Absolutely!  Easily done.  And the guy with the remote control will  
even be able to aim and fire them.  They had a remote-controlled gun  
for a while, out in the redneck wilderness, where people could log in  
by the magic of the inter-tubes and shoot bambi.  This kind of thing,  
we can do.  To get the Roomba or Robosapien to control them  
intelligently, autonomously?  No.  Not gonna happen.  Plus, these  
inexpensive machines are toys, not weapons.  They will break...  
immediately.

Anyway, good luck, have fun, and don't mind the cranky guy over here  
in the corner who's spent years of his life researching robotics, AI,  
and stuff.

Edwin!