The emergence of competitive trading algorithms in financial markets marked the onset of the competitive AI era. This trend may soon extend beyond the realm of finance, with AI competing in various other domains of business and society.
The ongoing grand discussion on AI overlooks something concerning and immediate. Beyond the debates over AGI, the clash between AI doomsayers and enthusiasts, there's a crucial question that demands attention.
What occurs when we pit our AIs against each other?
This prospect—referred to as competitive AI (CAI for short)—is oddly missing from all the social media debates. However, if you believe it's mere fantasy, reconsider. At a recent hackathon, developers devised the LLM Colosseum, where top models were trained in Street Fighter 3 and compelled to engage in combat.
Using "colosseum" as a metaphor signifies an existing AI category that might be on the brink of a breakthrough. AI combat won't stay confined to screens, with virtual fireballs being exchanged, at least not for much longer. It won't be long before these battling AIs spill over into the real world, vying for dominance in crucial realms of business and society.
What will this entail? And why aren't we addressing it? True, it may present new opportunities and enhance efficiencies—that's the positive aspect. However, it could also lead to strange, unjust, and, in some instances, tragic outcomes.
DEFINING CAI
Simply put, CAI refers to AI bots or agents engaged in direct competition with each other or humans in various spheres of business or society.
Competitive AIs are already in existence. They operate in both mundane and remarkable capacities, from purchasing goods to trading stocks, and even referencing legal precedents. They might already be present in competitive environments.
Expanding beyond traditional beat 'em ups, CAI is notably surfacing in supply chains, as exemplified by startups like Pactum. In essence, Pactum specializes in negotiation services for procurement. Its AI facilitates the purchasing of resources and goods on behalf of major corporations such as Walmart, negotiating with smaller sellers.
While Pactum stands as just one startup, albeit an intriguing one, its trajectory underscores a broader trend of technological competition. It's improbable that Pactum will remain the sole buying agent in its field. Its presence necessitates a response from other players in the market.
Transitioning to AI-powered adversarial buying will undoubtedly enhance efficiency but may also give rise to unforeseen issues. Before the impending wave of generative CAI takes hold, it's beneficial to examine past precedents.
FLASH CRASHES AND SHOPPING BOTS: THE PRECURSORS TO CAI
In contemplating the potential outcomes of CAI, it's valuable to examine existing instances of competitive AI.
The emergence of competitive trading algorithms in financial markets heralded the era of competitive artificial intelligence, providing a foundational understanding of the complexities it presents. These algorithms, engineered to execute trades at speeds and in patterns beyond human capability, have fundamentally transformed financial trading in an irreversible manner.
The ascent of high-frequency trading marked a significant shift towards leveraging algorithms for competitive advantage. These cutting-edge trading strategies utilize intricate algorithms to analyze market data swiftly, executing trades multiple times per second.
Trading algorithms have undoubtedly introduced a level of hyper-efficiency to markets, yet they have also been implicated in new anomalies, most notably flash crashes. These events, well-documented but still unresolved, entail a rapid and uncontrolled plummet in share prices within an extraordinarily brief timeframe, often followed by a rapid recovery.
In essence, AI competition has already led to peculiar and unforeseen occurrences. Apart from these unusual effects, it has also erected a significant barrier to entry. Companies that excel in the stock market are heavily technologized, leaving little opportunity for those lacking in AI expertise.
Following high-frequency trading, the progression of competitive AI expanded into the consumer sector, particularly through the introduction of shopping bots. These tools are engineered to surpass human buyers and other bots in acquiring sought-after items, such as the latest gaming GPUs or limited-edition fashion pieces.
The proliferation of scalper bots has been distorting markets for goods like PlayStations for an extended period. These bots automate the process of locating and purchasing stock the moment it becomes available online, frequently reselling at inflated prices.
This trend has given rise to the emergence of super scalpers—individuals or groups utilizing advanced AI tools to assert dominance in the online resale market. The competitive landscape of these AI bots means that as they progress, the rivalry intensifies. They are continuously upgraded to purchase items at even greater speeds, evade anti-bot measures, and capitalize on any available means to outperform both other bots and humans.
The ramifications of this AI arms race for both stock trading and shopping are profound and enlightening. As these tools advance, they consolidate success among a select few well-resourced operators, to the detriment of the average consumer. This scenario could lead to a situation where highly sought-after goods become progressively out of reach for the ordinary person, widening the gap between those who possess competitive AI capabilities and those who do not.
Concert tickets, anyone?
THE UPCOMING WAVE OF CAI
That's the historical context; now, returning to the present day. Companies such as Pactum and similar startups signify the imminent surge of generative, multimodal CAI. Many are currently in a nascent developmental phase, awaiting opportunities to interact with counterparts. Nevertheless, this transitional phase is dwindling. Without swift regulatory measures, the proliferation of CAI appears inevitable.
Among all the sectors poised to be impacted by CAI, the legal field appears particularly susceptible. Litigation inherently involves adversarial processes, and we are witnessing the swift emergence of foundational models and copilots tailored for this sector. Harvey, supported by OpenAI, stands out as the most advanced legal copilot and is already being implemented by law firms.
The concept of the "augmented lawyer" represents CAI at a slight remove, and is poised to reach a pinnacle during the anticipated Elon Musk versus OpenAI case. However, it appears that AI legal agents could soon directly engage in competition over legal matters.
Harvey is indicating its potential utilization in direct negotiations. This form of automated law is already somewhat unconventional, but takes on a surreal quality reminiscent of the movie "Inception" when one considers that all agents could be driven by the same foundational models. In such a scenario, who emerges victorious? The prospect of the same underlying model operating sock-puppet auto-lawyers and potentially even judges within the same dispute presents an unusual plot twist, one that is conspicuously absent from mainstream AI discourse.
The potential for CAI to be wielded as a weapon in reputation battles appears to be an inevitable progression in a society entrenched in culture wars. While startups like Signal AI may serve as the defense, the offense, at least for now, could be improvised, or multifaceted agents like Devin may be repurposed for more aggressive tasks.
The notion of sophisticated agents unleashed to roam social media and online archives solely for the purpose of personal takedowns is indeed disheartening. Unfortunately, we are already witnessing the effects of reputational CAI on a broader scale. In elections and geopolitics, clever trolls and propaganda farms are inundating the digital landscape with information pollution, frequently engaging in direct, automated responses to one another.
This inundation of CAI-generated content will undoubtedly affect more commonplace activities. With the recent collaboration between Copilot and Adobe, CAI could soon permeate marketing, with advertising agents engaging in adversarial conversations across social media platforms. Adobe is now focusing on the content supply chain, developing technology to generate endless imagery and messaging. When paired with CAI, it's only a matter of time until this content becomes dynamic, directly responding to competitors.
Indeed, the prospect of spam engaging in its own conversations is becoming increasingly likely.
Certainly, the military domain presents the most apparent and distressing use case for CAI. Battlefield chatbots, like those experimented with by the U.S. Army, along with advancements from companies like Palantir, are reshaping the landscape of traditional warfare. This hints at a future where AI-driven militaries can contend over terrains like mud and trenches, manifested through drones and advanced surveillance technology.
The transition towards autonomous defense raises longstanding but chilling questions about the role of AI in warfare. Recent reports have surfaced regarding the deployment of an AI system named Lavender in the Israeli-Palestinian conflict.
The looming question is, how long until it encounters an adversary?
THE "SMART CAPITAL" DILEMMA
The most significant concern posed by CAI is the potential for deepening inequality among companies, nations, and individuals. The substantial enhancement in capabilities that CAI offers to compute-rich and AI-powered entities will not be uniformly distributed. We might be on the brink of forging a fractured world.
In his optimistic essay "Moore's Law for Everything," Sam Altman acknowledges this shift towards an AI advantage. As artificial intelligence becomes increasingly prevalent, he believes that one of the primary wealth holders will be "companies, particularly those that leverage AI."
The essay outlines how an AI-driven world may mitigate conflicts over resources. However, it does not address the potential consequences in a scenario where CAIs are specifically engineered to compete for and acquire those resources through conflict.
CAI is an impending force that will further consolidate power towards capital and away from labor. In essence, CAI could give rise to AI-powered "smart" capital, a formidable force that will outperform traditional "dumb" capital, inevitably overshadowing human labor.
CAI agents will inevitably engage in competition for global resources. The societal challenge lies in ensuring that the tangible benefits of AI do not come at the expense of fairness and accessibility for those who have limited or slower access to AI technology.
we urgently need to initiate the CAI conversation. The pertinent question is not when CAI will arrive, but rather what steps to take in response to its emergence. Addressing this issue proactively is essential for navigating the potential challenges and opportunities that CAI presents.