AI: Threat and Mitigation

I try to resist commenting much on artificial intelligence (AI) simply because though I am somewhat knowledgable I am far from an expert, and the following discussion should be taken as offered. Nevertheless, with AI dominating the news I feel compelled to say a few things with regards to AI and its relevance to the resilience of space (and other) systems. In some sense this is a new dawn, and it is best to be prepared for what is to come. Forewarned is forearmed, after all. I do not believe what I’m about to state is in any way controversial in a factual sense.

To start, I’m not a fan of the term AI and the manner in which it is used in public discourse, primarily in such things as press releases and the press in general. “AI” has been and currently is used in the broadest sense possible to describe any number of complex algorithms that can be used to provide any number of capabilities. These algorithms are differentiated from others by the fact that they can be trained on an input data set and then operationally can learn and adapt to their environment. Many times this is denoted as AI/ML, where ML is shorthand for “machine learning.”

Early examples of AI include some versions of chess programs, such as IBM’s Watson, which became more than a brute force supercomputer, eventually refining its own play over time. More recently the term “AI” tends to mean neural-net based large language models, or LLMs, such as ChatGPT and Claude. Backed by huge supercomputers in giant data centers, these AI programs have demonstrated certain capabilities far beyond what humans can accomplish. It’s worth noting, though, that this is not a first. Over 20 years ago the best chess engines could outplay the best human grandmasters. This is a broadening and extension of past accomplishments.

Today’s AI is not strictly speaking, intelligent, at least not by laymen’s terms. What it excels at is data reduction and fusion: Big Data problems, taking disparate bits of information and formulating a hopefully-correct response. And it largely does this very, very well. The second thing that AI services do is democratize access to supercomputing resources. Fifty years ago the average person could not get access to a Cray computer. Now they can. Even for free under certain constraints. That is a game changer. Not unlike putting a weapon of mass destruction into the hands of one person.

So what does this all mean when it comes to resilient space systems design? It means that the barrier to entry has been significantly lowered for the development and application of some types of threats. What before might only be realized by a nation-state can now be achieved by a much smaller group of people. Maybe even a single person. Since AI exists in the digital domain, the emerging near-term threats tend to be cyber threats. Numerous examples already exist in the public domain, so I won’t recount them here. In particular, an AI-driven supercomputer now can execute what is eesentially a system penetration test on steroids in very little time, discovering vulnerabilities that previously might have been left unexploited for years. This trend is only expected to grow as hackers and other motivated parties learn how to better apply AI for their own gain. Space systems will not be spared. And older, established networks may be more vulnerable due to the lack of cyber security maturity at the time of their introduction and a false sense of security based upon many years of undisturbed operation. And as more networks are converged, the attack surfaces will only grow.

The other side of the coin is the use of AI systems by designers to increase the resilience of future space systems. AI can aid in solving very complex problems. One of those problems is working through the multi-dimensional web of determining the space system responses for a combination of capabilities, threats, and mitigations. And furthermore, to aid in postulating not only courses of actions, but the devising of new or enhanced mitigations to combat the emerging threats in support of new capabilities. Again, this is a Big Data problem: a space system consisting of many satellites, communications links (beams, channels), coverage areas, missions, capabilities, users, ground sites, networks, frequencies, waveforms, and on and on creates a very large trade space. System designers already employ a wide range of computing-based resources and tools to help develop optimal solutions. This is another tool in the tool belt. It is not, as of this writing, a replacement for the designer.

As always, it is important to keep an eye towards the future. AI is one of a number of technologies in development with the potential to greatly impact engineering and science as a whole, and this includes the development of space systems.. AI, quantum technologies, renewable energy,…all could revolutionize our world. A good engineer is always aware of the tools available to them. As with other tools, this tool has both positive and negative uses and implications. Once again, it’s a case of staying ahead of the curve to ensure that the positive impacts outweigh the negative.

Next
Next

As Simple as Possible, But No Simpler