After What Just Happened - Are You Concerned About an AI Apocalypse?
Next week, Apple is expected to announce its plans to rebuild Siri from the ground up. The tech giant plans to upgrade its chips to connect OpenAI to a core part of its hardware, and use AI to power more onboard applications.
Consider that OpenAI is already the backbone of countless technologies and applications, not the least of which is Microsoft co-pilot, a centerpiece of Microsoft’s operating system, currently used by 1.4 billion people. Thousands of AI businesses are now being built to power core applications for industries like finance, automotive, healthcare, energy and hospitality.
But, what happens when there’s a system-wide outage? What happens when these AI systems go down? It’s not too far-fetched to suggest that an extended outage could grind operations to a halt, threatening our economy and even public safety.
In fact, it happened this week.
On Tuesday, June 4th, at 10:33 AM ET, ChatGPT, Claude and Perplexity all went down simultaneously. The outage lasted for hours. I was actually using Perplexity at the time of the outage. For those unaware, Perplexity is an AI search tool that leverages ChatGPT as one of it's sources alongside LlaMa 3 and Claude 3 Opus. The thing was, they all went down.
When Perplexity finally came back up, I asked it what happened and it’s answer was somewhat disturbing:
"Perplexity, Anthropic’s Claude and OpenAI ChatGPT experienced simultaneous outages due to a widespread issue affecting multiple services. The root cause was a significant disruption in the infrastructure of a major cloud service provider, which each relies on for their operations. This disruption led to connectivity problems, resulting in downtime for various platforms that depend on this infrastructure.
Such incidents highlight the interconnected nature of modern web services and how a single point of failure in a cloud provider's network can impact numerous dependent services simultaneously."
That last line stopped me in my tracks.
As many are aware, the Internet is supported in part by central, redundant superservers, 7 of which are located here in the United States. This ensures continuity and access. As a result, there have been few widespread Internet outages in decades of commercial use. Some examples include outages to the Republic of Armenia in 2011 (due to a severed cable), a CenturyLink ISP outage in 2020, and an Internet blackout across West and Central Africa (due to failures in subsea cables).
OpenAI, Anthropic and Perplexity's cloud infrastructure is apparently not as dependable or redundant as the Internet.
As the business world begins to restructure itself upon foundations of AI, I believe Tuesday’s outages offer a stark warning of the potential risks of our dependence. According to Gartner, “More than 80% of enterprises will have used generative AI APIs or deployed generative AI-enabled applications by 2026.” So, It stands to reason that the more our economy's infrastructure and services depend on AI, the more prone we are to outages (and denial-of-service attacks). As such, it may be time to consider how your company is addressing emergency planning, failover, redundancy and security to mitigate the risks of AI.
