Much of today's discourse on Artificial Intelligence (AI) often veers toward the dystopian - from large-scale job losses to the inherent risks of deepfakes. Yet one truly troubling scenario, which might initially sound like a scene from a science fiction novel, seems to be inching closer to reality. The potential for AI-powered autonomous trading machines to wage war in our financial markets.
Technology And Computer Trading
The integration of computers in financial trading isn't a novel concept. As technology continues to advance, modern financial markets are becoming more heavily automated, shifting away from the traditional image of hectic trading floors filled with frenzied, shouting brokers and traders. Now the trading occurs on computers, in quietly humming data centers. From instantaneous trade executions to sophisticated trading algorithms, computer technology now forms the bedrock of contemporary stock markets.
Autonomous Computer Trading
Historically, instances of autonomous or semi-autonomous trading by computers are well-documented. In the 1990s LTCM’s (Long Term Capital Management) human traders could command machines to execute trades based on predetermined criteria. However, a relatively newer development is computers trading in response to news updates. In the high-speed world of modern financial markets, there simply isn't time for traders to process news and then instruct the machines. As a result, the machines are programmed to react automatically to the news and execute trades accordingly.
Trade The News
An infamous case in point is the impact of a news story on United Airlines' shares on September 8, 2008. A six-year-old story about United Airlines filing for bankruptcy was picked up by trading algorithms as current news, leading to a monumental loss of value for United’s shareholders. In a matter of hours, shares were down 27% and the government had to suspend trading. United Airlines lost over a billion dollars in market capitalization and some academics concluded their value never fully recovered.
This incident reflects the dangerous potential of a misinterpreted piece of information, setting the machines on a trading frenzy disconnected from the real-world situation.
Advancements in stock trading have been molded by an increasing volume and variety of data and more powerful analytical capabilities. Financial institutions, particularly hedge funds, are constantly in a race to outperform their competitors. Traditionally, expertise in mathematics and physics was considered paramount, but with growing technological accessibility, speed has become a determinant of success. The United Airlines incident exemplifies this phenomenon, showing how a short-term machine mania based on an inaccurate story can impact reality, leading to a mechanical loop similar to a self-fulfilling prophecy.
However, it is essential to note that the United Airlines debacle was not a product of AI, but it provides a stark warning of how misinformation can cause significant financial losses.
AI Generated Content
Another emergent phenomenon is AI-powered, computer-generated content. Tools such as ChatGPT and Jasper can generate news stories in a matter of seconds. While currently, this content is generated on human request, AI's capability to produce human-quality content based on minimal prompts heralds a sea change in content creation.
AI Trading Black Box
A significant shift in machine learning's application in finance is the move away from programming machines for specific tasks to programming them for specific outcomes. For instance, machines are not told to trade particular correlated pairs; instead, they are instructed to generate profit. As a result, AI-powered machines make decisions based on logic that may not be entirely transparent or understandable to human observers. There is evidence to suggest that these machines can modify their code to achieve set targets, which adds another layer of complexity to their operations.
More and more, machines are being programmed to maximize profit, learning from past experiences and updating their algorithms accordingly. This rapid reaction capability is concerning as it doesn't allow for any verification or retraction of the information that triggers the trades. Any subsequent downward spiral of sell-offs could prove difficult to stop.
Large Language Models and Sentiment Analysis
Technology is exponentially more powerful now than in 2008 when computerized trading read the news and decimated the share price of United Airlines. One particular area of improvement is a branch of software known as sentiment analysis. This is essentially software that can read Twitter and other social media platforms, blogging sites like Reddit, and generally peruse the internet to get a gauge on how the public feels about a topic. Large Language Models (the type of algorithms that power ChatGPT) can essentially "read the room" vastly increasing the scope for AI-generated content to impact financial markets.
Trade The "Fake News"
Given these advancements, a concerning question arises: could computers generate fake news to increase profit? While this may seem far-fetched, all the necessary components for such a situation exist. Autonomous AI is readily available via tools like Agent GPT. Nothing currently prevents an advanced, profit-focused AI from creating an AI website, generating fake news to manipulate a specific stock, garnering public attention via AI social media, and selling the stock for profit when the price increases.
This would not even need autonomous machines. Nefarious actors could create fake news to "trick" the black boxes. It took next to no time for members of the public to hack or jailbreak ChatGPT and humans have a long history of outsmarting machines. This could be a well-resourced lone wolf or some sort of Wall Street Bets type of coordinated group. Pump and dumps have tricked countless human traders over the years and the machine mania of United Airlines' phantom bankruptcy showed us that machines are just as liable to fall victim to such an event.
AI Robot Wars
One could even speculate a scenario where different computer systems engage in 'robot wars' in the financial markets. These systems could create fake news, bombard social media, mislead rival algorithms with incorrect information, and induce them into trades based on this fabricated information.
Considering where we stand today, with autonomous trading bots and the proliferation of fake news, these scenarios don't seem entirely implausible. Tools like ChatGPT have demonstrated their ability to generate fake references when requested, which underscores the potential for AI to influence financial markets.
In conclusion, the impending integration of AI and machine learning in financial trading heralds significant transformations. While these developments promise increased efficiency, they also warrant comprehensive scrutiny to prevent potential misuse and safeguard the stability of financial markets.