Spy catchers (counter-intelligence). An intelligence analyst competes against a researcher armed with AI. Who will catch the spy first?
This story illustrates the danger of overly relying on AI.
1250 words (4 pages) fiction plus commentary
It also shows the different mindsets of researchers and analysts.
The 4-page story is followed by a commentary on the differences between human intelligence and artificial intelligence, and the different roles of intelligence researchers and analysts. See commentary.
“A quick catch” – spy catcher story
Bhanuprakash studied his opponent. Athanasia had sharp features, a brash character and bright clothes. Her fingers moved so fast that they’d given her a reinforced keyboard and a touch-screen console. In the moments when she was reading she flexed her fingers as if squeezing imaginary stress balls.
This is a race, their shared manager in the Agency said. For his part, Bhanu sat rigidly on his chair with his hands poised above his keyboard like a piano player. His inputs were mostly limited to scrolling through the pages in front of him. Bhanu was immaculately dressed in neutral colours, with his facial hair trimmed so precisely that he stood out in the office. His weakness was his waistline, which had surrendered unconditionally to his love for his wife and her cooking.
“Are you actually trying?” Athanasia called out from her console opposite him. “I’ve already let two of my cats out, and Skimbleshanks is puffing and ready to go. They’re sniffing out our suspects.”
Cats. Athanasia had given each of her AI engines the name of a cat. Skimbleshanks was her railway engine, and it mapped out the network of contacts for a suspect.
“It’s all action over here,” Athanasia continued. She moved her arms like she was peddling a bicycle. “I’m already trawling through all our suspects and their comms, and creating profiles.”
Bhanu watched her animation. His own face was neutral – he was born that way. So he used his voice to project his thoughts. “I am reading, and I am thinking about the suspects.”
There were ten suspects, any one of which could have passed the secrets to the Chinese. It was trade secrets on mobile phone technology, not defence secrets, but everyone around Bhanu saw it is a national threat.
“Bhanu, you spend time thinking? That’s pre-hist-oric. There are quicker ways of achieving results.”
Bhanu considered the intelligence researchers he had known. She was on the hyper end of the scale. Perhaps it was her time working with artificial intelligence engineers that did it. Athanasia could coax results out of huge datasets using cleverly designed models of behaviour. The Android Spy Catcher. They called her that for her ability to find suspicious patterns in people’s behaviour – the kind of patterns that marked them as potential spies. In the last year she’d found a spy in a company building fighter aircraft, and another in the nuclear industry.
“Yes,” he said slowly, without taking his eyes from his computer monitor. “Your method is quicker, though perhaps not so good for global warming.” The Agency’s bank of computers became so hot that they were located by the river, and not far a power station.
The Android Spy Catcher ignored his environmental protest, as she did every time. “Hey, Bhanu, I’m not saying you’re redundant. You’re a great guy. You just need to get into machine thinking.”
Bhanu tilted his head gently to one side. His knowledge of artificial intelligence was not as great as Athanasia’s, but he knew its strengths and weaknesses.
By the end of the day he was still reading about their suspects. Unless he saw something special, he figured he needed two more days. It takes time to understand people.
Athanasia was finished next morning. “Hey, Bhanu,” she announced. “I’ve got the spy. It’s the systems architect who’s been leaking secrets to the Chinese. His numbers are off the scale compared to the other people around him. It’s a definite match. I’m passing the name up. As they say in tennis: game, set and match.”
Bhanu paused. Her interruption was a distraction to his concentration. “Congratulations,” he said. He contemplated what he’d read about the systems architect. “An interesting find,” he said. “I wondered if you’d say that.”
“You had suspicions about him? Well done. But it’s not the same as hard facts. Sorry about that. You didn’t have a chance against my vast data lakes and machine learning.” She raised an arm into a power salute. “The Android Spy Catcher wins again!”
“Congratulations,” he said. “I think I’ll spend another day or so building up detail about the systems architect. It will be useful for the surveillance team.”
“You do that, Bhanu. I’ll write up the report for the Big Wigs upstairs.”
Bhanu went home and sat in the little room he had used as a home office during the Covid-19 lockdowns. The walls were white with no decorations, there was a single window with shutters that he kept closed for security, and there were no distractions.
The next two days at work were difficult. There were increasing demands for him to give up and move to the next item on his worklist. He pretended to give in, but he was still reading and thinking. And at the opposite desk, Athanasia had completed her report and was getting excited about some technology improvements.
“Big news!” Athanasia said suddenly. “We’ve got an intrusive surveillance order on my spy. The surveillance guys are preparing to move in. They’ll be watching his wife and eldest son, too.”
Bhanu considered the situation. “This may not be good,” he said.
“What? Of course it is. We need evidence to convict him, and to find his handler.”
“Athanasia, it’s time you and I had a talk with the people upstairs.”
It required a meeting, of course. Nothing important happened without a meeting, preferably with a lot of posturing. This time they had a room with a view to the main road. At this time of day there were more delivery vehicles than cars. It was a distraction. Bhanu listened to Athanasia reiterating her case. Eventually it was his turn to talk.
“We should consider the possibility that the systems architect is a decoy,” he said slowly.
“A decoy? No!” It was Athanasia, interrupting. “His profile is full-on spy. He has contacts, he has opportunities for dead letter drops, he’s got communications, he has unexplained finance, and he’s had holidays in places where he could have been recruited.”
She hadn’t been listening, Bhanu decided. “His profile is too good to be true – it’s as if it’s been artificially improved to please the machines.”
“So? Just supposing your hunch is right, we still need surveillance to eliminate the suspect.”
“By watching him, we’ll alert the real spy and the spy’s handler. They will go silent, and we’ll have no evidence against the real traitor.”
Athanasia frowned so deeply it wrinkled her face. “Then what do you suggest?”
“We should look for the handler.”
“Bhanu, you’re talking nonsense: if we don’t have the spy, we don’t have the handler.”
“Not exactly,” he stood and moved to the whiteboard. “A decoy is only useful if the handler knows he or she’s come under surveillance. There has to be someone close to the decoy who is trained to watched for anything suspicious.” Bhanu wrote a name on the board. “Try this woman. She’s their child carer. I’ll show you her profile, it’s very indicative of a cut-out between a spy and their handler.”
“A cut-out? How do you know it’s her?”
“She also has contacts with another person on our surveillance
list. Didn’t your AI engine name her?”
Athanasia eyes widened. “The name was on the list. Skimbleshanks ranked it as coincidental.”
Bhanu didn’t react to the mistakes of AI. “We need to follow the cut-out,” he said. “Then we have the real traitor and we may get the agent behind them.” He looked at Athanasia. “There’s also the matter of your suspect. The surveillance needs to be called off. We shouldn’t be spying on innocent people.”
Athanasia’s eyes narrowed as she stared back at him. Her fingers curled in like a cat’s claws.
Commentary on artificial intelligence, human intelligence and spy catchers
“Counter-intelligence is an art. We start with an empty canvas and use the CI tools on our palette to paint our picture.”
James, M. Olson, former chief of CIA Counter Intelligence
On artificial intelligence vs human intelligence
The original purpose for this story was to explore how artificial intelligence can be tricked into giving false results. It’s a weakness that can be exploited by people with mischief on their mind. It’s used to defraud insurance and financial systems, and to confuse security and defence systems, and to queue jump.
In adding a twist to the end of the story, I found myself illustrating another important difference between humans and machine learning: most artificial intelligence relies on having a single measurable purpose. It has to be, because it’s calculations are based on comparing against the current case to large numbers of previous ones. Humans see the bigger picture, and some can make a jump in logic with seemingly random connections.
(I’ve also got other short stories on artificial intelligence – see all.)
On intelligence research vs intelligence analysis
As background to intelligence researchers, they are trained to find relevant information for a single purpose – something that can be measured. In large intelligence agencies, researchers tend to be very specialist, working within narrow areas with highly specialised tools, and without visibility of the bigger picture – Athanasia is an example of that. Another characteristic of researchers is that they are trained to avoid making personal judgements.
Like Bhanu, analysts make judgements. They handle the “big” questions and they’re trained to see context and connections. Analysts bring together research from many areas, in order to explain why things are happening, and to explore possibilities of what may happen in the future. Within national intelligence agencies, there is a preoccupation with estimating the probability of different outcomes, so that politicians and leaders can make informed decisions.
Analysis takes time, even when it begins at the same time as the research work. Before the analysis is complete the “customers” may already have access to the research reports, and they may have made decisions. (The customers include politicians, civil servants, military, other intelligence agencies, and more.) Using the story above as an example, a decision to start surveillance was made on the basis of the research, before Bhanu’s analysis was completed.
A caveat: there are notable differences for intelligence research in small units and the public sector – that’s covered in “Intelligence research specialist jobs – 14 tips for survival“.
On spy catchers and counter intelligence
The counter-intelligence (spy catcher) theme was inspired by James, M. Olson’s book, “To Catch a Spy”, Georgetown University Press, 2019. http://press.georgetown.edu/book/georgetown/catch-spy The author was chief of CIA Counter Intelligence, and then trained others entering the same profession.
It’s an extraordinary book that distils his experience of catching spies, including his “ten commandments” for practitioners. I found it a good read, with 12 real cases and anecdotes from many others.
The book is also an eye-opener to the almost obsessive mindset of CI people, in their hunt for spies. Perhaps they have to be like that to avoid the feelings of guilt about suspecting even close friends, and to have the persistence to keep going when there are so many obstacles.
I was intrigued by Olson’s repeated reiterations of his anger against people who spy against the US.
- Review in Washington Post https://www.washingtontimes.com/news/2019/may/7/book-review-to-catch-a-spy-by-james-m-olsen/
- Collection of review comments
https://www.hfsbooks.com/books/to-catch-a-spy-olson/
If you want a study of the morality of counter intelligence, this is the wrong book. The author’s mind is decided. Try something like “Principled Spying: The Ethics of Secret Intelligence”, by David Omand and Mark Phythian, Oxford University Press. https://global.oup.com/academic/product/principled-spying-9780198785590
This story was originally published 27th September 2021. It was lightly revised on 2nd June 2024 to remove minor errors. Since the original writing, there’s been a massive explosion of use of Generative AI, but curiously the story is as valid now as it was when written.
