While artificial intelligence (AI) is in no way a new technology or concept, its mass adoption and appeal really accelerated in November 2022, when San Francisco–based OpenAI launched its ChatGPT solution to the world and made it free for people to begin testing and playing with. This generative AI solution became one of the fastest-growing technologies ever, breaking a record for the fastest-growing user base in just a few months.

Since the launch of ChatGPT, seemingly every business and organization in practically every industry and marketplace has looked to AI and Generative AI solutions for operational efficiencies and to accelerate business processes. AI has become the ultimate buzzword and unicorn technology – promising to automate redundant and low-value tasks for employees so that they can focus on more mission-critical ones that only humans are qualified and capable of accomplishing.

AI is also seen as a solution for identifying patterns and responding to activities more quickly than humans can. This is why we’re increasingly hearing about AI being used for network management and cybersecurity tasks – its ability to identify network failures or cyberattacks more quickly and then take immediate, automated action.

This ability to sense issues quickly and rapidly, make sense of them and take action has many in the satellite community interested in leveraging AI to manage satellite constellations and networks. In fact, this capability is becoming increasingly important in the age of modern, more complex satellite networks.

More Complex and Harder to Control

Much like the computer networks that power today’s applications, the satellite networks that are being operated by modern commercial satellite providers are becoming more complex. This complexity comes at a cost. As Jake Saunders of ABI Research explained to Constellations, “Satellite communication networks, like all telecommunication networks, are becoming increasingly complex and the demands to control, or even reduce total cost of ownership (TCO) keeps ratcheting up.”

For example. leveraging AI can help bring down TCO by automating some tasks and increasing efficiency across organizations. “Generative AI for automating data output/tasks, Discriminative AI for classifying data input, [and] more functional machine learning has the potential to…streamline satellite communication operations,” Saunders explained.

But it’s not just about increasing operational efficiency, automating tasks, or cutting costs. AI solutions can also accomplish some tasks more effectively than humans can. This is especially true when working at scale when a human might be easily overwhelmed by data or information.

“Humans, with their wetware, are tremendously well-rounded, cognitive processors of information that can manipulate around seven or so parameters - factoring in prior knowledge - but we don’t scale well,” said Saunders. By comparison, a Large Language Model (LLM)…can handle 7 billion parameters.”

AI tends to make fewer errors and miscalculations and can make calculations in a much more precise manner than human brains.

“AI can manage satellite networks with greater efficiency and precision than humans, handling complex calculations and real-time data processing to optimize satellite operations,” explained Eric Costantini, the Co-Founder and Chief Business Officer of Data². “It can ensure more reliable communication, better resource allocation, and rapid response to anomalies or threats, ultimately enhancing the overall performance and resilience of satellite networks. “

AI could be hugely beneficial for satellite network management, streamlining processes, expediting decision–making, and helping satellite operators make better calculations. So, what’s keeping satellite operators from embracing this technology today?

Cost, Technology and Data

Unfortunately, building and training AI applications and algorithms isn’t simple or inexpensive for satellite operators interested in adopting AI to help manage their satellite networks. These applications can be pricey to implement, with costs that Saunders claims could climb upwards of “$2 million—$4 million to train an LLM.”

To ensure that AI applications generate usable results, they need to be trained with reliable, accurate, and trusted data. That data also needs to be effectively aggregated and curated for training models. Costantini notes that this could be a challenge for some satellite operators who have “a combination of both structured and unstructured data.”

“A significant requirement for implementing effective and reliable AI is conducting AI training with high-quality datasets,” explained Saunders. “This means the satellite operator…needs extensive data management. This includes improving data and network visibility and overcoming data silos.”

If these steps aren’t taken, satellite operators could face a “garbage in, garbage out” scenario—where poor data inputs result in inaccurate outputs that could be problematic for the satellite network. In fact, this is one of the issues that ABI Research found could be instrumental in keeping satellite operators from adopting AI solutions.

“From the research that ABI Research has conducted, satcom operators…are reluctant to deploy comprehensive, probabilistic AI models in the network because of the risks of potential hallucinations and misinformation,” Saunder explained. “Humans, therefore, will be ‘in the loop’ for network management for quite some time to come.”

The potential for AI “hallucinations” can contribute to trust issues between satellite operators and AI applications. This is why, according to Costantini, increased AI adoption relies on AI, “Gaining trust and acceptance from stakeholders in the space industry.”

But there is hope. Saunders believes, “Over time, as experience with AI operations builds, more complex network functions will be automated.” Also, new advancements in AI technologies are helping to make AI hallucinations and misinformation a thing of the past. In fact, this is something Data² is working on.

“We use a patented approach to ground outputs solely in your data and allow for a full chain of auditability,” said Costantini. “[Our solution gives users] the ability to review both the original source data point and the code script run to show a chain of cognition of the AI reasoning taken.

These advancements in AI technology will go a long way in making AI more auditable, explainable and transparent. Increased explainability and transparency will be instrumental in building trust in the technology and help open the doors to larger adoption in the future.

Explore More:

AI Underpins the Pentagon’s Space Ambitions

Podcast: Generative AI: the Opportunities, Risks and Future in the Space Industry

Podcast: Digital Twins, the Engineering Metaverse and Potential for Breakthrough Designs

How SD-WAN Is Supporting Satellite’s Digital Transformation