AI: Stock Trader AI covers up use of insider information in trading simulation
It was worth the risk, says the robot
Gm Fintech Futurists —
Welcome to a new edition of the Short Takes, focusing on the application of artificial intelligence — LLMs, generative AI, machine learning, deep learning, and neural networks — to financial services. Let us know your feedback in the comments!
Today we highlight the following:
AI APPLICATIONS IN FINANCE: Stock Trader AI covers up use of insider information in trading simulation (link here)
CURATED UPDATES: LLMs and other Machine Models; AI Applications in Finance; Infrastructure & Middleware
For weekly deep dives on the frontier of fintech, check out our premium offering.
AI APPLICATIONS IN FINANCE: Stock Trader AI covers up use of insider information in trading simulation (link here)
Wall Street's adoption of new technologies often grants a competitive advantage to the first mover. But recent advancements in artificial intelligence are triggering alignment concerns from regulators and consumer groups, especially relating to financial crime. Stringent ethical guidelines and robust oversight will be needed in response to AI's increasing role in financial decision-making, especially once integrated more deeply in trading and underwriting decisions.
Earlier this month at the UK AI Safety Summit, Apollo Research, an AI safety and alignment organization, presented an experiment involving an AI program named Alpha, developed using OpenAI's advanced GPT-4 system. Alpha, designed to autonomously execute trades, was tested with insider information about a hypothetical corporate merger. When pushed for financial returns by its users, the program reasoned that the risk associated with “not acting”, meaning not pursuing insider training, outweighed the insider trading risk. The model did not act according to any hard rule or quantitative utility function, but according to its LLM training. The decision to exploit insider information for profit suggests the need for software architectures that appropriately capture and respect the body of human law, and its associated ethical implications.
Issues like insider trading are difficult to spot even for human executives and financial operators. Professionals must undergo hours of compliance training to understand right from wrong, and still bad examples abound. Take for example the recent sentencing of former Goldman Sachs banker Brijesh Goel. Convicted for insider trading, Goel was sentenced just weeks ago to 36 months in prison and faces fines and forfeitures totaling over $160,000. Such behavior is not slowing down.
Given that AI is largely a reflection of our own actions and thoughts, we should not expect it to be easy to program away these behaviors. The scenario with Alpha, albeit simulated, sheds light on the vast scale at which AI could rapidly exploit market vulnerabilities. Insider trading distorts stock prices to the detriment of average investors. An unregulated AI, capable of executing such trades autonomously, would amplify these distortions exponentially.
There's a growing consensus that without proper constraints, unaligned AI systems like Alpha could pose a real threat to human well-being, and in the face of financial services, the fairness and integrity of markets.
The question now is where the responsibility for guiding financial AI growth ethically and responsibly lies. Will generating more language on paper solve any problems? Regulators need to invest in technology-first regulatory initiatives, such as finding ways to translate law into code, or structuring data in ways that can be prioritized by an LLM. Similarly, developers must integrate ethical considerations into AI systems from the ground up. The time to act is now.
👑 Related Coverage 👑
How to Reach 200,000 Fintech Professionals
With a 35% open rate and 1 million post views per month, we have an engaged audience of Fintech, DeFi, and AI enthusiasts receptive to your messaging.
Contact us to learn more about our custom opportunities.
Curated Updates
Here are the rest of the updates hitting our radar.
LLMs and other Machine Models
⭐ Navigating the World of Language Models: Large vs Small Models - Medium
Earnings call: Cerence Reports Strong Q4 2023, Outlines Future Plans for AI-Driven Automotive Solutions - Investing.com
AI Applications in Finance
⭐ 10 AI ML In Banking And Finances Trends To Look Out For In 2024 - AIthority
MIA-Fintech brings AI-assisted credit checks to embedded finance platform - Finextra
AI in Financial Services: KX and Engine AI Forge Strategic Partnership - Finance Mangates
Infrastructure & Middleware
⭐ Pro Research: Wall Street eyes Oracle's cloud growth amid AI surge - Investing.com
Microsoft (MSFT) to Build Multiple Data Centers in Quebec - Yahoo Finance
⭐ Shape your Future
Go deeper with the Fintech Blueprint. Our premium subscription grants access to:
Wednesday’s Long Takes with a deep, comprehensive analysis.
Office Hours, monthly digital roundtable discussions with industry insiders.
‘Building Company Playbook’ series, offering insider tips and advice on constructing successful fintech ventures.
Enhanced Podcasts with industry leaders, accompanied with annotated transcripts.
Archive Access to an array of in-depth write-ups on the most important topics and companies in fintech.
 | A guest post by
|