"How AI/ML Can Support SDA".
Col. Roth
On the Importance of Space for National Security
-Our national interest in to and from space... space is now a warfighting domain.
-Number one operational imperative is the space order of battle because all those other operational imperatives need space to win.
On the Awareness and Understanding of Adversaries in Space
-Space domain awareness is fundamental for us to develop these plans, action systems and components.
-For space order or battle, space domain awareness is a key building block for that because you can't have a space order in a battle and know what to do without space domain awareness.
On Predictive and Reactive Capabilities
-Predictive Behavior was really important for us. We have this on our Ops floor.
-The next example I wanted to show was real world operations... we lost contact with the vehicle So Not Unusual to lose contact... we were able to bring back a satellite that was stuck in a restart 5 days later.
On the Role of Technology (AI and ML) in Enhancing Space Operations
-It took about 5 days to train the model and then within minutes it diagnosed exactly what happened on orbit.
-By and large we talked about AI/ML in general and that space domain awareness is the foundation to help us understand space.
On the Challenges Faced and Importance of Vigilance
-It even goes down into the Cyber realm so it is really important for us to have space domain awareness and it's not just about looking up in space as well.
Col. Raley
On the Role of AI in Enhancing Speed and Accuracy
-What we're looking for out of AI/ML is faster and more accurate analysis. A machine can think faster than a human.
-Timely manner improved object tracking...machines can keep track of all the details that are going on.
-The cyber interactions that we need to look at in the future having artificial intelligence systems.
-One of the best return on investments is that the Resource Management aspect of artificial intelligence allows us to move from an automated system.
On Space Domain Awareness and Environment Understanding
-It's more about creating a better information environment about what's going on in space.
-Understanding space weather is not something that's easily understood... granularity of our space.
-One very non-intuitive aspect of our space domain awareness is space environment
On Distinguishing AI and Autonomy
-I want to draw a little bit of a line... between artificial intelligence and autonomy.
-Understand the difference between autonomy and artificial intelligence autonomous systems.
On the Role of Humans in Collaboration with AI
-By us paying people largely offshore to label data... we have to come up with better ways to understand what happens in Space.
-I don't have time to have you just sit and click through on mundane tasks of labeling data so we have to come up with better ways to understand what happened in Space.
-If they fail to trust it when it's thinking faster than them and ahead of them then they're going to be suboptimal.
-Here's the things that I need help with, opportunities for my industry and academic partners.
On Potential Risks and Challenges of AI
-If you have an intelligent agent inside the machine... how to put that within the correct boundaries.
-We need training for humans to trust AI.
-Another way that's kind of simpler to put this is if you have an intelligent agent inside the machine it may decide...that's a risk we take when we go do something like that.
On Quantifying AI's Value and Effectiveness
-I need quantifiable data about what's the right thing to do about that.
-My job is to quantify how well that does whether it is worth the investment.
-That predictive Behavior was really important for us... it also shows that if we were paying attention enough to the Telemetry that was coming through that we would have known 30 days before it failed.
-It's a very straight train when I'm talking about creating a system that's supposed to be helpful to an operator. You know I'd like to make their lives easier but what I really want to do is make them more effective at doing their job day to day.
On Potential Risks of Not Adopting AI
-All that said and even with the warnings that we've put out here about using AI, if we fail to keep up with AI we're going to face down the risk of doing nothing.
Q & A
Col. Raley
On Human Intervention and AI Decisions
-Certain functions and critical things you need a man in the loop but you may need a blue Suitor in the loop as well.
-We don't want artificial intelligence machine learning deciding on when we defend ourselves with nuclear weapons.
-I don't know that we have things in the lab that are particularly about that but that would be one of the things we would look to quantify is when does it help to have that person there or not.
On AI's Assistive Role
-As far as a general principle especially as we look at the operational level of command and control that artificial intelligence should probably be more in that assistive role of the did you think of this were you paying attention to this piece of data.
On Human Emotion vs. AI Decision Making:
-In the same way that you can a human and that becomes very difficult for people to digest when we're talking about life and death decisions.
On Time Factor in AI Autonomy
-I think the key element though is going to be time we're a human is going to slow us down outside of our adversaries where it is no longer worth having that human in the loop.
On Investment in Talent Management and AI Research
-If you see a lot of the breakthroughs that are coming out from open AI Google Microsoft Etc students all over the past year.
-It's going to be that we have to give them some autonomy to run their own research programs and enjoy the success of that because we aren't going to compete on a purely monetary basis.
On Pace and Advancement of AI
-Well I was going to say as far as keeping up with the pace of AI we're actually and I kind of got to this when I was talking about how data sets are one of our critical needs we're going to have to learn to teach AI to train Ai.
On Integration of AI into Operations
-Perfect more capabilities faster and the operator is open to trans you know something that would have been very difficult to bring a contributing sensor on to the space satellite SDA network is now a lot faster.
Comentários