Via Satellite

Space 2.0: Taking AI Far Out

Artificial Intelligence (AI) on ground and on orbit may forge a new era of information, with only the most relevant data increasingly brought forward.

Albert Einstein’s presumed words were: “Only two things are infinite, the universe and human stupidity, and I'm not sure about the former.” Jest or not, the universe’s being endless is an axiom all around the world. But perhaps also deserving this universal acceptance is data; it is an ever-expanding cosmos, and the more we know, the more we still have to discover. It is a never-ending realm of exponential proliferation. So rapidly expansive is data, that is just may be the only thing bigger than the universe itself.

The enormous volume of data being produced and added to this cosmos on a regular basis is empowering. However, while potential presented in the data is overwhelming, the rate at which we can transform it into actionable insight is underwhelming. This is not because of the “stupidity” referenced by Einstein, but simply because of the “human.” It is here that Artificial Intelligence (AI) overcomes this to unleash this potential.

Applying AI to the space industry is revolutionary. Far from an industry buzzword, it is already playing a significant role, and this can be easily understood when looking at remote sensing satellites. Ordinarily, these collect terabytes of data that must be downlinked to a ground station before being processed and reviewed. But now, enabled satellites could carry mission applications on board, including AI that would conduct that processing on the satellite. This means that only the most relevant data would be transmitted, not only saving on downlink costs but also allowing ground analysts to focus on the data that matters most.

Making this real, Lockheed Martin Space developed SmartSat, a software-defined satellite architecture that enables AI on orbit. A SmartSat-enabled satellite, the first of which is scheduled to lift off this year, could also carry — and constantly upgrade — its own cyber defense applications, improving security and resilience in orbit.

In full support of realizing AI’s technological revolution, Lockheed Martin has upped the ante. The aerospace manufacturer doubled its commitment to Research and Development (R&D) during the past five years, investing $1.2 billion in R&D in 2018 alone.

Lockheed Martin’s AI developments encompass both on-ground application and on-orbit. These span predictive maintenance of spacecraft, anomaly detection, human-machine augmentation assistance, adaptive cyber protection, and space modeling and simulation.

“We’re increasing our investment even more to continue bringing forward innovations in AI and autonomy that can be adopted and scaled to tackle complex, far-reaching and rapidly evolving challenges,” says Akash Patel, chief of AI strategy and ventures, Lockheed Martin Space Mission Solutions. “We are building AI and autonomy technology that will keep people in control while enabling them to be safer, more effective and able to focus on higher-level tasks.”

Trust is Critical

Having achieved AI breakthroughs is just the start though; trusting the behavior and outcomes of these systems is critical to collective success. The challenge here is correctly identifying the right entry point of humans in the loop in order to mitigate any potential risks and threats.

“AI will never replace human intelligence, but it will augment and enrich it. Our brains have much more advanced reasoning, logic and are faster than any computer. You’d know this if you’ve ever stumped Siri or your Google Assistant. What AI does is continuously enable the algorithm to learn, and it learns at much faster pace than human brains can. Eventually, the dangers will run out,” says Patel.

Speeding up this trust-building process are instances where AI solutions are essential, says Patel, adding that we have to be just as strategic about trust as we are about missions. In space, the systems are thousands of miles away, it’s not possible to send a repair crew to fix something. Similarly, astronauts on the International Space Station (ISS), or the first to land on Mars will rely on systems that can predict, self-diagnose problems, and fix themselves while continuing to perform without failing. In these instances, AI is the only viable solution and, trust or no trust, human lives depend on it.

“No technology is without risks if it is not properly addressed and governed. Therefore, AI also presents enormous risks if the ability to understand, analyze and act is not used to improve the quality of what is produced,” says Gaetano Volpe, co-founder and CEO of Latitudo40, a startup that applies AI algorithms to satellite Earth Observation (EO) images. “I believe that some form of regulation on what can be done and what cannot be done with these applications will have to be defined.”

Unlimited Applications

Volpe, whose expertise also extend to satellite telecommunications in the maritime market, sees AI impacting all sectors of the space industry, from launch to constellation control and satellite performance analysis. In a fast-approaching future, most operations involving satellite services will be based on some form of AI, he says.

“There will be a great impact on the two great families of applications: telecommunications and Earth observation. The possibility of applications is certainly unlimited, from predicting failures on a satellite through careful analysis of telemetry, to optimizing the routing of data packets for smallsats dedicated to telecommunications services on a global scale. The wave of technological innovation is just beginning, and in terms of identifying AI application, we can now only see a fraction of what will be uncovered in coming years,” says Volpe.

The application of new technology, ranging from automatic learning to autonomous or semi-autonomous decision-making, provides for a profound paradigm shift in the value chain. Today, we are witnessing the commercial application of AI in the downstream sector, but we will likely have satellites fully designed to work with AI logic directly on board the payload. This means the satellite becomes an increasingly intelligent object with ever greater capacity for autonomous processing, reducing the amount of data to be transferred to the ground. “Autonomous navigation and robotics can be applied in deep space applications, while image classification and predictive analysis of phenomena, based on neural networks, can be applied in a commercial context,” says Volpe. “In Earth observation, AI has the power to bring the use of geo information to sectors that have not yet been explored, making it within everyone's reach. A consequence of this new technology is that there will be no need for domain experts to extract the information contained in the data cubes. But an expert system, properly configured, can do so in a totally autonomous way, presenting the end user, which can be an urban planner, a site manager, an infrastructure manager or an agronomist, with the data needed to make decisions.”

Enter the Military’s Journey to AI

In addition to these end users, there are many others across numerous verticals, if not all, that require data for apt decision making. This is especially true in the military, and a reason for the massive amount of data that the military possesses. Processing these lakes of data into actionable insight at the speeds required to best support the soldiers out in field is achievable because of AI. And Raytheon is playing a crucial role in enabling exactly this. The U.S. defense contractor is developing the weapons systems, Command and Control (C2) systems, information systems and satellite systems for the U.S. Department of Defense (DOD) to transition to AI.

“AI is not simply about finding an algorithm; it is a process and a journey. It starts with data,” says Christopher Worley, Director of Digital Innovation at Raytheon Intelligence, Information and Services.

The DOD is very large with numerous repositories, or what is referred to as data lakes. These first need to be identified and then a process of data curation and cleaning begins, followed by data labelling — all of which are necessary steps to getting it ready for development. Only after Raytheon has prepared all the data, can it begin applying the technology required based on the needs of the DOD. And this is varied, says Worley.

“With C2 systems, we ask: How can we accelerate decision making from the intelligence, surveillance and reconnaissance side of the house? How can we fuse information from the data that’s collected to assist analysts deriving conclusions? There are many things that are often not thought about, such as medical records. How do we expedite readiness in order to give the medical teams the information needed to help diagnoses? Then there’s maintenance; we’re involved with the platforms, architectures, information systems, or even aircraft and ground vehicles. How do we apply AI to alert the maintainer before an accident or incident occurs? How can we apply AI to accelerate the logistics process to enable readiness on the battlefield at the technical edge?”

Challenges and a Generation of Distrust

While identifying where the data is poses a challenge due to the sheer volume, another obstacle dressed in red tape also needs to be overcome. A lot of the DOD works in a classified domain, but most of the advanced AI capabilities within the U.S. are in the commercial sector, and these players don’t necessarily have the clearances.

Ensuring that AI is sustaining is another challenge, explains Worley, noting that just because you build an algorithm off of data, doesn’t mean that it’s going to remain effective. You have to continually update, upgrade and retrain the algorithm based on the new information that comes in.

“There is no room for complacency. Once you commit to AI, you’ve got to be committed to sustaining it. It’s a long-term investment, but you’re ROI is man-hours savings and higher effectiveness in operations. There is no question on the ROI, it’s just about making the decision to go forward,” says Worley.

Even after surmounting these hurdles, at times there is still an issue of distrust, which is challenging because confidence between the operator and the machine is imperative.

“Adopting AI leads to a new culture; it’s a different approach to doing business. Whereas the younger generations of the military understand AI and have worked with it in private life, there are other generations within the DOD who are less familiar with this technology. The question is: How do you build that trust, so that what you’re used to doing on your own can now be done by a machine? This trust needs to be built and this requires modelling and simulation, exercises and development. This is also why we’re committed to human-machine teaming,” he says.

AI is a disruptive technology and there are always going to be challenges. But even if this poses a barrier, it doesn’t mean we can’t overcome it. It seems only time separates is from our collective success, but when considering the rate at which developments are happening today, this is fast approaching. VS