This correspondence is featured in Australian Foreign Affairs 10: Friends, Allies and Enemies.
To read the full issue, log in, subscribe or buy the issue.
“Data Driven” by Danielle Cave
Danielle Cave’s essay thoughtfully examines the pain points that are driving intelligence agencies towards a new kind of tradecraft. In a world of big data, the intelligence community needs to be faster and more persuasive, agile enough to pivot to urgent “black swans” while maintaining coverage of long-term priorities and more mundane, predictable events.
While Cave is spot-on about spycraft becoming increasingly data-driven, it is worth unpacking what this means in practice for Australian intelligence agencies, especially as they balance the roles of humans versus advanced machines.
Spies may benefit from having more data at their disposal, but data can also add layers of complexity and difficulty to intelligence collection and analysis. How do you parse increasingly large amounts of information to find insight? As cryptographer and security expert Bruce Schneier puts it, when your mission is to find the needle in the haystack, the solution is not adding more hay.
Our key intelligence competitor, China, might disagree. China has effectively industrialised mass surveillance and data mining. With security cameras on every street corner, consumer data collected through every app, and cybersecurity laws giving pervasive powers to the state, China vacuums up enormous caches of data daily. Based on estimates from its Academy of Sciences, China currently holds 20 per cent of global data, or forty-four billion terabytes.
The sheer scale of this data collection raises concerns about “techno-authoritarianism” and the impact of data-driven technologies on privacy and human rights. China’s government has already weaponised data against the Uighur and Kazakh populations in Xinjiang Province. Worryingly, China is increasingly expanding or exporting these tools of social control.
Australian intelligence agencies cannot perform such wholesale data collection. Our agencies are subject to more stringent accountability standards and legal constraints. Even if Australia’s collection capabilities were to reach the same level as China’s, it is hard to see us trampling our values to copy China’s playbook.
If intelligence agencies cannot best China on data collection, they might still gain an edge through better analysis. Cave predicts that over the next decade, agencies will spend much of their time figuring out how to host, process and use data, which only offers value if it can be properly analysed.
China has invested heavily in artificial intelligence for data analysis, boasting that it has built machine-learning algorithms capable of accurately sensing and predicting criminal behaviours and national security threats. AI has the added benefit of being able to process information at lightning speeds without fatigue, unlike the human intelligence collector or analyst.
However, the deployment of AI for intelligence analysis has its own pitfalls. AI’s outputs are not always as accurate as they purport to be. From racist facial recognition to sexist credit scores, examples of algorithmic bias have been widely reported in recent years. Often these biases stem from flawed data used in the training of AI systems. It is an important reminder that data needs to be fit for purpose – it must be reliable and relevant for the specific dilemma that the AI is trying to solve. Complex AI systems, such as deep neural networks, can also present a “black box problem” where the AI’s outputs cannot be fully dissected and explained. Biased or unexplainable AI is unlikely to deliver intelligence that convinces decision-makers.
We ought to imagine a future intelligence community that combines the best elements of both human and machine to wrangle data and make better-informed decisions. This will entail greater investment in tools and technologies, especially AI, but also the workforce capabilities to harness data effectively and avoid AI pitfalls. So-called “soft” sciences – linguistics, sociology, anthropology, history – are critical for transforming data into insight. Take languages as the most basic example. There is a joke among China watchers: “What’s China’s first layer of encryption? Write it in Chinese.”
Spycraft will never be solely data-driven. Having the right human in the driver’s seat is more important than ever.
Olivia Shen is a policy adviser, a Fulbright scholar and a former visiting scholar at the Center for Strategic and International Studies in Washington, DC.
Read Danielle Cave’s response here.