On Tuesday night, The Austin Forum on Technology & Society hosted a community event titled “The Great AI Debate.”

The event featured a panel of 4 AI experts from the ATX community debating AI and its expected impact on society, in front of an audience of 100+ attendees.

Addressing topics like AGI (Artificial General Intelligence), risks of powerful AI, jobs, long-term views in 200+ years, and more, it proved to be an intriguing discussion where the panelists at times did not agree.

One point of contention was regarding risk of AI negatively impacting society. The full spectrum of responses was represented with the extremes being “no negative impact whatsoever” to “potential for catastrophic impact.”

While I hope the former is right, I cannot reasonably ignore the possibility of the latter.

That same day, Anthropic shared news that they would be holding back their latest model, Mythos, from public release. Why? The model’s capability to find and exploit software vulnerabilities has surpassed that of most humans. During testing it uncovered 1000s of vulnerabilities across the tech we all use today, including a 27-year-old vulnerability that no human or automated review had ever caught.

Incredible. But, also a bit alarming.

I share this not because I am a doomer. I am definitely not a doomer. Quite the opposite. I believe the potential impact of AI is truly exceptional and will have the ability to transform the world in a positive way.

However, I also recognize that AI is advancing at lightning speed and powerful models like Mythos are here right now. If a model like this could string together a few vulnerabilities to take control of the infrastructure that our global society is built on — that’s a problem. And in the hands of bad actors, potentially catastrophic.

Society must be able to catch up.

Holding models like this back, at least initially, and instead making them available to the cybersecurity industry immediately buys us some time. Time to catch up. Time to build a more modern and more secure infrastructure. And then when models with these capabilities are deployed to the masses, it can be done safely.

And if this is possible, I believe sophisticated AI will have the opportunity to transform the world into something far more amazing than we could ever imagine today.

Cure disease. Solve hunger.

That is a future I want.

Thank you to The Austin Forum on Technology & Society for hosting this event for our community and calling attention to this important topic!