#23: AI - From "Caution and Concern" To "Supernova Exhilaration"

SEC Warns of AI's Threat to Financial Stability

Gary Gensler, the chair of the US Securities and Exchange Commission (SEC), has expressed concerns about the potential financial instability resulting from the widespread use of artificial intelligence (AI). He has stressed the urgent need for regulators to address risks from the concentration of power in AI platforms.

Key points:

1. Gensler believes that without timely intervention, a financial crisis induced by AI is "nearly unavoidable" within the coming decade.

2. A significant challenge for U.S. regulators lies in the cross-sectional nature of AI risks. Existing regulations primarily target individual financial institutions, but the AI issue is more horizontal. Many institutions could rely on a single AI model or data aggregator, often hosted by tech giants outside of regular financial oversight.

“It’s frankly a hard challenge, It’s a hard financial stability issue to address because most of our regulation is about individual institutions, individual banks, individual money market funds, individual brokers; it’s just in the nature of what we do. And this is about a horizontal [matter whereby] many institutions might be relying on the same underlying base model or underlying data aggregator.”?

3. In July, the SEC proposed a rule concerning conflicts of interest in predictive data analytics. However, the rule centered on models used by individual broker-dealers and advisers and didn't address the broader, horizontal challenge posed by AI.

4. Gensler pointed out the limited number of cloud providers in the U.S., which often provide AI services, underlining the concentration risk.

5. The EU is ahead in the regulatory curve, drafting strict rules on AI use. Meanwhile, the U.S. is still assessing which facets of AI need new regulations versus what falls under existing laws.

6. While AI is increasingly being incorporated into Wall Street operations, there's apprehension that relying on a single data model can lead to "herd behavior," potentially sparking a financial crisis. Gensler foresees such a crisis possibly occurring in the late 2020s or early 2030s.

7. The broader regulatory community in the U.S. is also focusing on AI, with concerns spanning market stability, data protection, and potential tech monopolies due to AI's scalability.

“I’ve raised this at the Financial Stability Board. I’ve raised it at the Financial Stability Oversight Council. I think it’s really a cross-regulatory challenge”.

8. Gensler, known for his efforts to promote market efficiency by tackling concentration, also flagged potential competition issues with AI, especially around market-making.

Overall, the rapid integration of AI in the financial sector and its potential systemic risks have raised alarms, urging regulators to take swift and comprehensive action.

AI Revolutionizes Drug Discovery and Scientific Research

* Susana Vazquez-Torres, a graduate student from the University of Washington, is using AI to advance drug discovery for ailments like snake bites.

* Snake bites kill around 100,000 people annually, and the current treatments are costly and unsafe. With AI's intervention, promising compounds can be developed within months instead of years.

* The U.S. National Academies discussed AI's potential to revolutionize scientific disciplines. Yolanda Gil believes AI can collaborate with humans to conduct experiments systematically.

* At the University of Washington, AI is helping design proteins. AI can predictively design a protein structure from scratch, bypassing the traditional tedious testing process.

* The AI technique used is 'diffusion modeling,' similar to popular AI image generators. It tailors proteins with a specific design using accumulated protein data.

* In fields with less centralized data, like material research for renewable energy, AI application is challenging but considered essential.

* Researchers envision AI playing a vital role in scientific discovery, formulating new ideas by analyzing vast scientific literature. The hope is to detect patterns and connections that humans might overlook.

* Gil envisions AI systems planning, conducting, and continuously updating scientific experiments, increasing accuracy and systematic approach.

* Despite AI advancements, researchers like Vazquez-Torres remain optimistic about the continued collaboration between humans and AI. She believes that the potential solutions AI can help develop outweigh concerns of job displacement.

ChatGPT: Not Just a Chatbot Anymore

Nvidia researchers have showcased the versatility of the technology behind the popular chatbot ChatGPT. In just a few weeks, they trained it to play Minecraft, where the chatbot learned various in-game skills such as mining, swimming, and building houses. Jim (Linxi Fan), a senior research scientist at Nvidia, commented on the bot's ability to navigate and improve within the Minecraft world autonomously.

This experiment is indicative of a broader shift in AI research, transforming chatbots into autonomous A.I. agents that can engage with online tools such as software apps, websites, and spreadsheets. The long-term vision is that these agents will evolve and might replace white-collar jobs, automating numerous tasks.

Jeff Clune, a computer science professor with prior experience at OpenAI, highlighted the vast commercial prospects of such technologies, which could be worth trillions. Current A.I. agents, like ChatGPT, offer limited capabilities like searching on Expedia, but the future promises more sophisticated digital personal assistants.

The underlying technology, GPT-4, is remarkable not just for language processing but also for its ability to write computer programs. This capability was key in Nvidia's Minecraft experiment. While chatbots are still at the beginning stages of accessing various internet Application Programming Interfaces (A.P.I.s), advancements are being made.

Recently, OpenAI expanded ChatGPT's functionalities, enabling it to, for instance, fetch maps or generate visual charts, by running the code it writes. Similarly, companies like Google and Microsoft are investing in such innovations.

The dream for researchers is to build A.I. agents capable of setting and achieving complex goals online, like starting a business. While today's systems aren't fully there, they're progressing rapidly.

Dr. Clune's work in 2022 with OpenAI aimed to teach A.I. agents to use software tools mimicking human actions, like clicking and typing. The process involved the A.I. learning from online videos of people playing Minecraft. Start-ups and other businesses are working on similar projects, aiming to make digital assistants that can use various web tools and platforms.

Dr. Clune believes that when A.I. agents can perform any online task, it will greatly simplify life but also poses the risk of replacing many jobs, emphasizing,

“It replaces all the tasks.”

AI in Astronomy: Robots and Algorithms Revolutionize Supernova Discovery

A breakthrough in the world of astronomy sees artificial intelligence (AI) simplifying previously time-consuming tasks. Bright Transient Survey Bot (BTSbot), a new tool developed by researchers, has seamlessly integrated robotic telescopes and machine learning to detect, confirm, and classify a supernova - all within a matter of days.

"For the first time ever, a series of robots and AI algorithms has observed, then identified, then communicated with another telescope to finally confirm the discovery of a supernova ultimately freeing human researchers for deeper analytical tasks."

Highlights:

1. The Process: On October 3, the robotic observatory, Zwicky Transient Facility (ZTF), spotted a new celestial source, named SN2023tyk. Within two days, BTSbot identified this as a potential supernova, a task typically performed by human astronomers.

2. Automated Spectrum Collection: Following the detection, BTSbot signaled another robot, the SED Machine at the Palomar Observatory, to gather the spectrum of this event. This spectrum, akin to a cosmic fingerprint, reveals the supernova type.

3. Analysis & Classification: Another algorithm, SNIascore, assessed the spectrum and confirmed the discovery as a Type Ia supernova, caused by a white dwarf collapsing upon drawing too much mass from its partner.

4. Announcement: BTSbot went a step further by broadcasting the discovery on October 7 via the International Astronomical Union’s Supernova working group website.

This transformative system was created by a joint effort from Northwestern University, Caltech, the University of Minnesota, Liverpool John Moores University, and Stockholm University.

SIGNOFF

"Why did the AI go to therapy? It had too many neural issues!"

Thanks for tuning into this week's DEEPakAI: AI Demystified newsletter. Stay connected and keep learning! ????

p.s. - The newsletter includes smart prompt based LLM generated content.