No, the CIA won’t be reversing time with generative AI. It’s a reference to George Orwell’s dystopian novel, 1984, which details a future where society is devoid of personal freedom and overrun by surveillance due to the totalitarian regime led by Big Brother. The novel illuminates the terrifying potential of a totalitarian regime to control thought, rewrite history, and obliterate dissent, reflecting Orwell's fears of the consequences of unchecked political power on human rights and the fabric of society. But why refer to this novel?
A few days ago, Randy Nixon, the director of artificial intelligence endeavors at the US Central Intelligence Agency (where, as legend would have it, secret agents are minted), shared with Bloomberg that the agency is in the midst of crafting a chatbot, reminiscent of the ChatGPT ilk, to lend a helping hand to analysts navigating through the expansive seas of internal and external data for enhanced investigative ventures. This digital aide is also honed to distill and encapsulate a plethora of web-based information proficiently, delivering well-cited responses to the curious, thereby facilitating a more enlightened intelligence analysis and decision-making saga. Nixon conveyed that this generative AI sprang largely from the spur of China's brisk strides in the AI arena, notably within the corridors of government and defense.
While one might harbor the notion that this generative AI will serve as a harbinger of good, fortifying the shores against terrorism and misdemeanors on US soil, the annals of US intelligence outfits are inscribed with instances of surveillance excesses. And this newfound tool could, perchance, nudge surveillance pursuits to a realm mirroring the Orwellian vision of 1984.
US Intelligence Agencies Have Always Been Spying on You
Perhaps you may recall the name Edward Snowden, the US National Security Agency (NSA) whistleblower who leaked classified files a while back. Now, while the NSA and CIA aren’t exactly siblings, they do share the occasional clandestine cup of coffee, hinting at a probability that they might swap shiny AI tools amidst murmurs of intelligence.
Snowden revealed the NSA’s hobby of collecting digital whispers from the pockets and desktops of the unsuspecting. With each text, call, and ‘like’, the NSA knitted a cozy “pattern of life” quilt around not just the purported baddies, but every soul unfortunate enough to be caught in the wide, indiscriminate net of association.
Types of Data Collected by NSA
Now, one needn’t be pen pals with a terror suspect to have their digital trail scrutinized by the NSA. The agency has the liberty to hop, skip, and jump “three hops” from its targets. A hop here, being a connection from one person to another. So, if you’ve ever exchanged digital nods with someone, who nodded at someone, who nodded at someone who nodded at you - congratulations, you're part of a network vast enough to dwarf the population of Colorado, should you boast a social circle of 190.
The NSA utilized several surveillance programs: PRISM accessed data from tech companies; XKeyscore analyzed global internet data; Boundless Informant organized metadata; Upstream tapped into global communications; BULLRUN decrypted secure communications; MUSCULAR intercepted data between tech giants' data centers; and the EvilOlive Metadata Collection program analyzed connections between individuals through phone records and other communication metadata.
The Snowden reveal was like a plot twist in a spy novel, showing the NSA had a knack for eavesdropping that made the existing rules seem like a joke. The Foreign Intelligence Surveillance Act (FISA) of 1978 was supposed to keep a lid on the NSA's nosiness at home, but over time, it got tweaked to allow more snooping without needing a warrant, especially if it was about non-US folks, despite how this violated the Fourth Amendment of the US Constitution which protects against unreasonable searches and seizures. Even under the eyes of Congress, these laws started to operate more in the shadows, and many lawmakers acted surprised about the NSA's shenanigans, and thus there has not been meaningful progress towards reforming the NSA even today.
Generative AI Can Take Surveillance to Orwellian Levels
In the not-so-distant past, intelligence agencies were quite adept at carrying out surveillance operations even without the aid of generative AI. Now, with the advent of generative AI, the amount of data they can process has skyrocketed. This scenario has its pros and cons. On one hand, generative AI could help CIA agents swiftly identify real threats like terrorism or major criminal activities and act on them. However, on the flip side, it presents a tempting tool for the CIA or other intelligence agencies to ramp up surveillance on citizens, masking this overreach under the guise of preventing terrorism and crime, while fishing for potential political dissenters.
The plot thickens with concerns regarding the reliability of the foundational model that the CIA is likely employing to power this tool. It's speculated that they might be using an on-premise version of Scale AI’s Donovan solution, bolstered by Cohere’s large language model (LLM). The CIA even has a testimonial on Scale AI Donovan solution page, adding a twist of intrigue.
However, the hiccup in this narrative is that Cohere’s LLM was recently found to be the underdog among popular LLMs like GPT and Claude, as per a detailed analysis conducted by Vellum.ai. Beyond just lagging in performance, LLMs have a tendency to hallucinate and if such a tool becomes a part of the CIA's daily toolkit, there's a chance it might misidentify individuals, linking them erroneously to suspects.
While it seems inevitable for the US government to hop on the generative AI bandwagon as other global counterparts are doing, the stakes are high. If the Senate and, by extension, the US citizens, take lightly the amplified power generative AI could hand over to US intelligence agencies in terms of surveillance, the dystopian narrative of George Orwell’s 1984 could inch closer to reality as soon as 2024. The phrase “Big Brother is Watching You” could transition from the pages of a fictional narrative to the stark reality of modern-day governance.
Komentarze