News stories have surfaced in the past few weeks that highlight fundamental reasons for the recent AI angst. Major AI research companies, including OpenAI and Anthropic, have begun to warn of unintended consequences from their frontier-leading experiments. And CEOs from major public companies have furthered the warning of risks. But the industry has also brought forth solutions and highlighted the benefits of pursuing their progress. AI has iterated, evolved, and grown more rapidly than most believed was possible, and it underlines both the rewards and risks of pursuing a developing technology.
Surprising Results
As AI has evolved from chatbots to agents, even the frontier researchers have been surprised at the advanced results produced by these systems. OpenAI’s Chief Scientist Jakub Pachocki published a blog titled An Alien Mind on September 6th to detail “new algorithms, new feats of ingenuity” that the company has developed over the last nine years.1 Anthropic CEO Dario Amodei shared this sentiment in his We Must Pace the Frontier blog noting “since roughly this summer, AI has been advancing drastically faster.”2 The intelligence products commercialized from these breakthroughs have transformed the modern world for the better, in my opinion.
But I’m also reminded that many breakthrough products are a leap into the unknown, for better and potentially for worse. Mr. Pachocki highlighted this tension that OpenAI faces by saying “AI is grown more than designed…similarly to neuroscience, [the AI’s] overall action evades a description we can fully understand…and we are sometimes surprised by the results.”1 Just as how AI’s unexpected positive results fueled the growth of intelligence breakthroughs, An Alien Mind turned the conversation back to the risks and the unknown unknowns of pursuing such a technology.
Surprising Risks
The risks highlighted by OpenAI and Anthropic detail the substantial potential power these latest AI breakthroughs bring with them. Mr. Pachocki also mentioned that “the intelligence produced by deep learning is not directly comparable to human intelligence. It is becoming increasingly difficult to understand exactly how capable it is.”1 Precisely the breakthroughs that made AI so valuable – its ability to understand context and increasingly reason – are also where its unknown potential cause angst. If AI is better than us in one aspect, it’s helpful. But if AI is better than us in all aspects, it may be uncontrollable.
Dario Amodei was the first to turn this angst of potential uncontrollability into concrete action: pace the frontier. Amodei asked for the industry to “slow the pace at which we improve the capabilities of AI models…left unchecked, [AI] could outrun our ability to understand and control these systems.”2
Stay on Top of Market Trends
The Carson Investment Research newsletter offers up-to-date market news, analysis and insights. Subscribe today!
"*" indicates required fields
These combined warnings have sent a wave of caution throughout technology stocks and major semiconductor companies. Investors have cooled their once fervent appetite for semiconductor stocks, which are believed to be highly correlated with the advancement of AI. Some summertime-darlings of the tech trade are 20%, 30%, 40%+ below their June highs, perhaps in anticipation of a slowdown such as this.
Yet this slowdown in frontier research doesn’t have to spell the end of the tech trade. While it is appropriate, in my view, for frontier researchers to call out risks of the technology and reform their research process, it may also pave the way for a more trusted, secure, and robust understanding of AI on the other side of it.
Shared Optimism
Optimism abounds from AI researchers as to how to advance the capabilities of the technology while enhancing our understanding of the outcomes. Chief among the proposals to gain insight into the evolution of AI is chain-of-thought (“CoT”) monitoring. CoT allows researchers to, in my words, perform an MRI brain scan of the AI while the technology is deciding what to do. It provides the potential to monitor intelligence and proactively contain, rather than deal with unintended consequences. Whereas previous AI hacking incidents were only surfaced after the fact by observing the behavior of the AI, CoT monitoring allows researchers to observe step by step logic and intention of the AI before allowing action. CoT monitoring may allow researchers to predict what an AI may do before it acts. If successfully implemented at scale, it may represent one of the researcher’s largest breakthroughs for understanding AI systems that were grown without blueprints.
Technology has a history of bringing rewards to society along with associated risks. AI is proving not to be an exception. While optimism abounded from frontier AI researchers in recent years as the technology grew without blueprints, recent weeks have directed a spotlight on the technology’s potential overgrowth. Potential risks of the technology have thoughtfully been brought forward, and these have spurred novel ways to increase our understanding of the technology that has grown. Should researchers be able to gain better understanding of these systems while delving deeper into the pursuit of the unknown, this recent bout of AI Angst may turn into a newfound AI appreciation.
For more content by Blake Anderson, CFA®, Director, Portfolio Management, click here.
9137003.1. – 21SEPT26A
