Morning
Colloquium
Facilitated
by Ray Hill
We
are going to start today with a slideshow presentation of the graphics
created during yesterday's presentation. We will spend a little time looking
over the images and discussing any open issues from yesterday. Then we
will have some discussions around yesterday's presentations.
Cultural
Change in the Navy
Jeff
starts the conversation by talking about some work he did at the Naval
Undersea Warfare Center. He was helping them reorganize. They wanted to
capture more money and be prepared for the future. They took a schematic
of a sub and laid it beside their organizational map. It is a well known
fact that most organizations are organized similar to the product they
make. For example, Ford's organizational map looks a lot like a car. The
same was true for NUWC.
Click
on image for enlargement
The
Navy is based on three platforms. There are the ships, subs, and airplanes.
There are alignments and loyalties to each platform. In fact, you wear
a pin on your uniform to show your platform loyalty.
In
our work with the reorganization, we tried to get them the think about
a world that did not revolve around these platforms. This was a huge culture
shift.
We needed to break the current bias.
A
challenge in doing this was finding resources in the budget to work on
changing the culture. The thing we were talking about was "breaking
the welds". This felt like a threat to the Navy structure. This is
the type of change that will take a long time.
Compare
this with Search Theory. Search Theory is a concept. Again it is hard
to spend part of your budget on concepts, when there are physical and
technology needs that have to be met. There is no incentive to really
learn about Search Theory as a military officer. It will take about a
generation (10 years) for some new ideas in Search Theory to catch on.
Introducing
UAVs
New
ideas are always difficult to implement. A recent example is the introduction
of the the UAV concept. When you invent a new idea it always creates a
pandora's box. A few years ago we could not use the term UAVs. People
kept pressing the ideas and kept talking about UAVs. Thanks to them we
can now talk freely about UAVs.
Budget
Authority
Everything
comes down to the two most powerful words in the system: "Budget
Authority". When you present a new idea you have to explain them
how much money they save first. With Search Theory it is hard to sell
something that is not a thing.
There
is no limit to what you can do as long as you don't care who gets the
credit. An example of this is the Special Forces in Afghanistan. The Special
Forces are the breeding ground for innovation. They have more freedom
for innovation.
(Jeff
comments on how when he was in uniform there was a role that he had to
play within the service. He had to take that uniform off to be able to
do the work he wants to do.)
Snowball
Effect of New Ideas
There
are people listening to Search Theory out there in the world. We have
some believers out there who agree with Search Theory and Agent Based
Models. The ones we thought had the most powerful message have been destroyed
due to the threat of their message. This is common with any innovative
or disruptive technology.
There
is a snowball created. It will roll down the hill eventually. We have
to keep bringing people together to talk about this stuff. This
is not an easy fight. One way to create more success is to present this
as an additive technology and a not replacement technology.
Fear
of Change
It
is not just Search Theory but when you talk about change in general. Anything
that you want to get in to the congressional budget will take three years
of lobbying just to get your idea heard. But, if you present it as an
update to a program then you have a better chance of being heard.
You
have to think of your technology as an additive capability. If you give
other program managers the chance to talk, can help create internal communities.
Revolutionary
Events Bring Change
Something revolutionary has to happen in the world. An example is September
11, 2001. Because of that event the Special Forces have a much stronger
position. They are buying autonomous sensors now. They have a political
position and those that support them have a political position too.
Support
on the Inside or Outside
Do
you think there needs to be more acceptance on the outside before this
gets accepted inside?
There
are organizations that are desperate for technology to help them achieve
their missions. The desperate ones will listen. You have to show how things
are a small additional cost. Money for innovation is not always an additive.
People who innovate within an organization are not always viewed highly.
Sometimes you have a new person to come on board to take a job. Their
goal is to keep their job and to get a promotion. Innovation may not be
the way to get promoted. It is not always safe. Sometimes the safe route
is to not rock the boat.
Beginnings
of Acceptance
With Iraq and Afghanistan, there has begun to be acceptance for this type
of information. There are also a lot of people trying to move Search Theory
from the bottom up instead of top down. We do not have to hit all the
big organizations. A lot of great work in this area comes from the small
entrepreneurial businesses working on this topic. An entrepreneurial firm
can do a lot more with $300,000 than a big company.
Why
we are here?
We
are here to come together and build a community of interest. We want to
identify opportunities. As bad as it is in the world right now, there
is an opportunity driven by the current situation that could open the
door for Search Theory.
Recent
Innovations
There
have been a lot innovations over the last few years. For example,during
Gulf War I, getting lost in the desert was a very real and dangerous reality,
until there was the creation of a magic little box called GPS. This device
has since changed the world. In 1992, there was a request by the Special
Forces to make an electronic filmless camera system. This would enable
people to communicate intelligence without the use of film. This new way
of transmitting pictures made the way for digital cameras. The Special
Forces drove digital pictures for us.
There
are events going on right now that are driving innovation.
Paradox
A
paradox is figuring out how you get around the engineering of these systems.
How do we use a technology surrogate? We can take random objects and combine
them to make a product with zero engineering. We have to find easier ways
to test concepts. Everything does not have to be done at high cost. We
need more small companies who are performing simple tests with given rule
sets. We need to be able to have physical abstracts. In order to make
to UAV's better, we don't just build it better, we make the behaviors
better. The behaviors are as important as the engineering.
What
cognitive aspects of the search problem deserve further study?
One
cognitive thing we are looking at is sonar operators. We have observed
that there are two kinds of sonar operators. Those that do well and those
that don't. One of the schemes that we are looking at is having an expert
operator in one place so the other operators can call on the expert. They
can share screens. Can the expert do more than one job? How many customers
can he handle? How do you determine how well the expert is doing?
This
gets to the issues of distributed intelligence. How do people feel about
doing their job in a group setting instead of as an individual? If you
want to change the way people do things you have to change the reward
system. You
can do anything as long as you don't care who gets the credit.
How
do we extract information from an expert without overwhelming them. How
can the expert publish his thought process? One way is through the use
of weblogs.
People can publish in a market place their thought chains. These thought
chains needs to be publish so they can be mined. People can then put out
smart queries to access his experience.
How
do you overcome the publish and pride issues? Where
are the incentives in a shared model? It is a dichotomy?
Click
on image for enlargement
What
are some of the human integration issues associated with distributed systems?
This
is a horrible problem for UAVs. Getting the output from UAVs to fit into
existing command and control systems. We have attempted several systems.
The problem is that people keep changing how they manage the UAV on the
fly. Literally, they make changes while the UAV is in the air. The usual
solution was that they needed more people. While the number of people
involved increased and the utility did not.
Someone
needs to fly the UAV and decided where it goes. Someone always has an
idea of where it should go and someone else has another idea. This is
a serious problem.
A
possible solution to this problem would be automatic target recognition
(ATR). As the ATR gets better, this will be a powerful step and will reduce
stress on the carbon based life forms.
Still,
deciding where the UAV goes–and what it does with the information
that it collects–is the major problem. As the systems matures and
artificial agents become more common, they will handle more of the administration
problems. Such as their routes and their interactions
Human
Integration
You
can not afford to not have the human integrated in the process. The human
aspect is the most innovative aspect and the edge that we have to have.
It is not about control. It is that the human is the kernel of success.
Will
there be a day when we do not need the carbon base life form? How much
more insightful is the human if they can see what a whole swarm sees and
not just 45 degrees to either side out of the window of an aircraft.
There
is only so much ROI that we can have. Automation will make the system
more efficient. But still it will not always be right. The Automation
Technology is out there and other countries have access to this technology.
We constantly strive to be 10 years ahead of our competition in terms
of technology.
Could there be a time when systems compete? It is a mandate that humans
be involved? The human mind is incomplete and this leads to new thoughts,
solutions and innovations. Nothing can be as novel as the human mind.
At what scale do we inject human cognition.
Where
does the human integration not belong ?
You
can not take the human out of the sensor-shooter situation. It is hard
to trust combat to automation. Will we ever have technology that understands
itself and its functions–is "self-aware"–and knows
when to use its powers and when not to get involved. Will UAVs able to
fly themselves? It may be shown that if the guys on the ground can just
let them fly then the fleet could perform better than if we interfere.
If
you look at technology, we are becoming managers of portfolios. Look at
all the portfolios that we currently manage in our life (home stereo systems,
computers, etc.). The Army will take the UAV to the next level. Ground
soldiers do not have time to fly and operate on the ground. They need
to have voice commands because they do not have time to manage the UAV.
Boid
Simulation
There
is a great simulation out on the Internet called the Combat Boids (www.combatboids.com).
In this simulation based on the flocking behavior, the birds chase you,
they go your course and have unlimited fire power. When you kill them
a genetic algorithm replaces them with the surviving birds' fitter
characteristics (speed, accuracy, agility, etc.). You are outgunned 20
to 1. Now I know the rules, I can play the game and make it a competition.
There is an algorithm and I have to try to discern it. In life, the real
game is our networked forces against their networked forces.
Distributed
Forces
What
if you have distributed forces against swarms. What if you take out communications?
Can we develop a way for people to not need to communicate to each other.
If they know the rule sets, can they execute their plan?
Is
there a way to communicate once you loose your leader and communications?
What if there were two leaders now? How do people know what speed to operate
at? We need something like a pace car (like a software) and everyone follows
the pace car. When we think about cognition, control and command, we need
to think about the rules. How do you define and find the fall back hierarchy?
What is the horizon that people should look at?
You
need a sparse network and two links for every node. There are many misconception
we have about how collectives work. We want to make sure that we are working
the right way. We do not need everyone to be connected. If they are, then
you have people who can not get past certain points.
Click
on image for enlargement
Blame
In
situations where an error has been made, would you rather blame a machine
or a person? In fifty years it will be battle of machines. People want
to know "WHO" is responsible for mistakes. An interesting observation
on this topic is the recent release of the NASA report on the Columbia
Space Shuttle. In the report they are not blaming particular people. Humans
like blame. Do
we produce scape goat agents so we can delete them?
We
talk about this a lot. If we can't get past this cultural hurtle we should
stop now. There are a lot of people who think this way. Robust systems
will fail in some cases and it is a distributed system. You may not hit
the target at first. The notion of failure and trust in your system will
have to be understood.
Use
of Robots
As
long as we are using this technology for Search only, we will not have
this problem. What if the robots are able to kill? How do robots fit in
the supply chain? Do we use them to bring meals and carry the injured
off the field? Are we creating Terminators?
Culture
question for SSG. The culture does not buy into the kill only the sense.
The autonomous sensors are everywhere. Only sensors can bridge. It is
only worth it if we can do the kill chain.
If
you think that sensing is not followed by a bomb or a bullet, your are
mistaken. Building sensors for sensing is worth it. And then they will
go to the next level.
Trust
One
thing that made ebay rise out of the soup is that they raised their price
and developed trust between the buyer and seller. Indirect connections
contribute more to tipping point behaviors. The notion of trust needs
to be developed more fully. We have to learn how to trust collectives
that do indirect things for you.
Loyalty
Sensing
is very important. It helps us understand what is and is not a target.
A problem with agents is that they do not have loyalty. There is no human
control and things can be reprogramed. There is no loyaltyamong agents,
and nothing is 100% secure. How do we ensure that agents will do what
they are supposed to and not be altered?
One
way to do this is make sure that if one thing fails that there are overlapping
sensors to make sure that networks catch the failure. You can also build
a system and a counter measure to your system. Things will be broken into
and will be changed. You have to build your counter measure.
It
is hard to guarantee a program will do what it is supposed to do. There
needs to be ways to make sure that parts of the system are not dependent
on the other parts of the system.
System
Failures
The
research needs to be about compromises in the system. Not just where the
system creates improvements but also shows the new dangers it presents
and how we mitigate those risks. We need more computer viruses to understand
the weaknesses that are in our systems. We need to break the systems more
to be able to make them stronger. We need to understand how systems will
fail. Understand the counter measures to your systems. Just like you never
send out a chemical weapon without a vaccine.
Click
on image for enlargement
How
might some of the new analytical techniques be employed to juggle/analyze
the multivariate trade space faced by the various services?
The
value of models is that beforehand you do not know all that you need to
know. Models give you the effects of the various outcomes.
An
example is an agent based model for a Marine Expeditionary Project. Everyone
had an idea to add more realism. One suggestion was to introduce noise
and measurement errors. What were the consequences of introducing biased
noise? By the end, we had a hundred different tunable parameters. Once
you do that, you have a complicated tool that you are trying to understand
and can generate enormous amounts of data. There are very systematic ways
to do this.
If
you have a model with 10 parameters, you can study the models by varying
the key parameters to get the outcome desired. This enables you to evaluate
the system. What are parameters that cause blue to win?
Efficient
learning systems are good at honing in on what is important. What you
are doing is making a simplified model of the model. This way you can
capture the important dynamics of the model and how the outcome is influenced.
The outcomes show you what is happening to make blue win or loose.
With
these numbers you can understand the complicated dynamics of the model.
It is important to push towards realism. You can not be frightened by
what you are doing or your outcomes. You must truly
analyze in detail the behaviors of the agent based models. Push
decision trees as a painless way to test a model. This
also helps you know the information you need for the model to be able
to understand the variables. You
have to figure out what you need and do not need.
Socialization
in Models
We
ran a big model in Korea. We then invited the high-ranking officers from
the United States and Korea to analyze the campaign. We knew we over-represented
someone if they were not complaining. There are techniques to discern
where the contributions come from and very few processes are not biased
to a certain aspect of the military. Many of the tools that we use have
been socialized.
Speed
of Navy Ships
In
the Navy, they created a model to see how fast ships can go. They did
not give the model speed constraints based on current state of technology.
We wanted to see what the break points were. Was it more efficient for
us to travel to certain destinations by air or sea. Is it worth investing
in something that can go faster than our current ships? Maybe we want
a ship that can go 145 knots.
Constraint
in Models
In
some models, you want to put constraints and in others you want to leave
them out to see if you come up with something new. If you understand cause
and effect, you will know the relationships that work. If
you use the right tools you will not look at 400 variables all together
but you will see that 10 variables are making a difference.
The
model says what should be happening and you can see what strange behavior
is going on that effects the model. This is a way to understand the model
and keep debugging the system.
We
have to think about all the possibilities and take them into account.
We do not want to limit ourselves to not make mistakes. Models
can help you identify tradeoffs or multiple objectives. There are techniques
that that you can use to maximize the trade offs and evaluate what is
important.
|
