314: Mike Schuster, Two Sigma AI Leader and Google Translate Pioneer, On AI in Finance, Data Challenges, Collaboration, and Future Trends
AI and the Future of Work
Deep Dive
What are the key differences between building AI technologies in tech companies like Google versus finance firms like Two Sigma?
The primary differences include company size, resource availability, and collaboration culture. Google, with over 180,000 employees, has vast resources, allowing for experiments on thousands of GPUs. In contrast, Two Sigma, with around 2,000 employees, operates on a smaller scale, requiring more cost-conscious decisions. Collaboration is also more critical in finance, where complex systems demand teamwork, code reviews, and rigorous testing, which may not be as ingrained in finance as in tech.
What challenges does Mike Schuster highlight regarding data quality in finance?
Data quality in finance is often poor due to multiple factors, including human errors like 'fat-fingering' trades, engineering issues in data recording, and inconsistencies during events like mergers. These issues make financial data noisy and messy, requiring significant effort to clean and standardize before it can be used effectively in AI models.
How does Mike Schuster view the ethical risks of using AI in finance compared to other industries?
In finance, ethical risks are less about disadvantaging specific groups (as in tech) and more about ensuring accurate decision-making. The primary concern is avoiding overpromising or misusing AI, which could lead to financial losses. Unlike tech, where bias might affect users directly, finance focuses on optimizing measures like risk and return without disadvantaging retail users.
What was the breakthrough moment for neural machine translation at Google?
The breakthrough came when a small team, including Mike Schuster, developed a prototype using neural techniques instead of statistical methods. This prototype, tested on English-to-French translation, outperformed existing systems significantly. The success led to scaling the model across multiple languages, eventually running on 20,000 servers globally, revolutionizing machine translation.
What does Mike Schuster predict for the future of AI over the next five years?
Schuster expects incremental improvements rather than revolutionary changes. He anticipates models becoming more efficient in terms of energy use and cost, with advancements in software rather than just hardware. Data quality will remain a challenge, but filters to separate good from bad data may improve. Overall, AI will become more integrated into daily life, similar to how speech recognition and calculators are now commonplace.
How does Mike Schuster address the risk of AI overpromising in finance?
Schuster warns against the hype and overpromise of AI advancements, which can mislead investors and decision-makers. He emphasizes the importance of grounding expectations in reality, as many doomsday scenarios predicted in AI have not materialized. Instead, he advocates for a balanced view, focusing on practical improvements and avoiding exaggerated claims.
What role does collaboration play in advancing AI at Two Sigma?
Collaboration is crucial at Two Sigma due to the complexity of AI systems. Schuster highlights the need for teamwork, code reviews, and rigorous testing, which are standard in tech but less common in finance. Building a culture of trust and constructive feedback is essential for tackling large-scale projects and ensuring the reliability of AI systems.
- Challenges of building collaborative teams in finance for complex AI systems.
- Differences between AI development in tech companies (Google) vs. finance firms (Two Sigma).
- Importance of code reviews and unit tests in collaborative AI development.
- The evolution of collaboration in finance towards team-based approaches for complex projects.
Shownotes Transcript