BlackSky and Deutsche Bank explore how an AI-based, software-first approach solves customer needs, the implications of generative AI and more.
Artificial intelligence is transforming how satellite Earth observation (EO) technology meets the evolving needs of commercial and government customers.
BlackSky CTO Patrick O’Neil recently joined Edison Yu on Deutsche Bank Research’s “Podzept” to discuss the role of AI in ushering in the era of real-time, space-based intelligence with a software-first approach.
BlackSky CTO Patrick O’Neil
Here are some of the highlights*:
What does AI mean for the EO industry?
“Earth observation systems produce copious quantities of data. Because of the quantity of data that is being produced by our system and other remote sensing systems, AI becomes enormously valuable for actually dissecting all of the data and understanding what’s happening.
“We typically think of pixels being produced, but how do you go from pixels to answers and actions that can come out of the information that you’re getting from those pixels?
“I think the space industry is both simultaneously extremely advanced technologically and, at times, somewhat archaic in the way that things happen. At BlackSky, we try and bring (AI) advances from other sectors of technology into the space industry and try and leverage a software-first approach to the way we deliver data to our customers.”
Transforming imagery into meaningful insights
“You can go into our system today and task satellites to collect images, and we’ll deliver those pixels to you. But you can also, at the same time, order some analytic products to be delivered that are derived from the pixel data that you’ve ordered.
“The planet is a very large place… Keeping an eye on activity around the entire planet is very challenging. At BlackSky, we’ve developed systems that allow us to take very, very large quantities of satellite imagery, then through that imagery, use AI to detect activity patterns like new construction or new roads being built. This allows our customers to not have to sift through every single pixel that we’re delivering [allowing] them to scale this capability to be able to find interesting activity signatures over very, very large areas.
“[W]here a lot of the value really comes in is when you take a look at that over time. So now you’ve been monitoring this location for, let’s say, several months, and you can start to spot trends. You can start to spot anomalies. This is how we were able to identify movement of military forces around the world and the repositioning of assets. There was a recently published study that showed another media firm [Financial Times] using BlackSky data to analyze the movement of Russian naval forces around the Crimea region and detecting when ships were moving from ports, and the repositioning of forces.”
Generative AI – Discovering the precursors to critical world events
“[W]hat a lot of these advances in Large Language Models (LLMs) and generative AI and these kinds of capabilities give you [is] the ability to do what’s called zero-shot learning, where you’re asking an AI to solve a problem that it’s maybe never solved before, and it’s able to do that without large quantities of training data.
“[M]onitoring the world is ultimately about tracking impacts to people and understanding activity. Understanding what happens after a natural disaster is really important, and the impact of every disaster can be really, really different. If you’re monitoring the supply chain, many things can happen. It’s really hard to anticipate what types of questions need to be answered in the future.
“In a traditional world, for every question, you’d have to generate this very specific AI. I think the advancements that we’ve seen from generative AI and LLM become really interesting in the geospatial world when applying zero-shot techniques to answer some of these key (futuristic) questions.
“At BlackSky we’ve been curating this large dataset for the entire planet where these key facilities – airfields, power plants, key critical nodes of the supply chain – have been monitored for years.
“Having the AI analyze very large quantities of these deep stacks [of imagery] and then being able to ask questions of it, in the same way that you ask questions of ChatGPT and other large language models, [is beneficial] so that you can immediately get results back without having to specifically train models on every single use case you may ever encounter. That to me, is where the biggest benefit is going to be. BlackSky is really well positioned to capture that capability because we have this extensive data catalog that we’ve been collecting for many, many years.
“That data catalogue becomes more valuable when we launch our Gen-3 satellites later this year, which are going to bring us down into a very high-resolution world where we’re going to see much more fine-grained human activity at a fine-grained temporal scale. And you really need to do both so that you can identify those activity patterns that are precursors to some critical events.
“[I] think some of the most interesting use cases that we’ve had historically have been related to some of the supply chain monitoring that we’ve done for some commercial customers. [For example] being able to take our satellite imagery, our high frequency imaging system, to monitor the shipment of goods and identify when and why there might be delays.”
A new insights-based acquisition model
“You’re already seeing pretty rapid growth in the adoption of these kinds of models. When you get into the national security space especially, they are not looking for ‘what’s the coolest new thing to solve my problem?’ They’re looking at ‘what will actually solve my problem? Because my problem is really important and lives could be on the line based on the performance of various systems.’
“We are already seeing pretty significant movement towards adopting this insights-based acquisition model. So instead of buying raw pixels, buy that AI output. And we already have a pretty extensive set of capabilities in that segment. We already have customers who are buying that directly.
“It’s taken time for customers to trust the output of that AI capability, but we’re very far along with our key customers in that conversation. They are purchasing, effectively, the analytics and using the analytics to tell them which pixels to look at. We’re delivering very large quantities of data, and hiring people to review every single one of those images is not scalable. We have very large customers with serious needs that have already adopted this model and are taking this AI-first, analytic-first approach to purchasing the data services that we have.”
The full interview is available here.
Learn more about how BlackSky is unlocking the future of space-based intelligence.
*Edited for clarity