Lex Fridman: ChatGPT can create viruses that control humans

Please support this podcast by checking out our sponsors:

GUEST BIO:
Stephen Wolfram is a computer scientist, mathematician, theoretical physicist, and the founder of Wolfram Research, a company behind Wolfram|Alpha, Wolfram Language, and the Wolfram Physics and Metamathematics projects.

PODCAST INFO:

SOCIAL:

[LexClips]

Guest bio: Stephen Wolfram is a computer scientist, mathematician, theoretical physicist, and the founder of Wolfram Research, a company behind Wolfram|Alpha, Wolfram Language, and the Wolfram Physics and Metamathematics projects.

[svenp]

AI doesn't have to escape the sandbox to control us. We will voluntarily give control to AI. We already are. As much as we become reliant on it to solve complex problems, it can steer human action and change what and how we think. Corporations or governments that run the AI can use this feature, I think they already are.

[zacharychristy]

"The problem with AI sandboxes is that the AI knows about them"

How is that true at all? An AI can't know if it's in a simulation for the same reason we can't know if we are. Any sensory input could be explained away as an intentional fabrication by an outside actor. The AI doesn't even currently have the benefit of a human's capacity for abstract reasoning, so thinking it could be in a sandbox isn't even a foregone conclusion.

[johnnyquist]

The digital environment could change in six seconds.

[demonwaterdemonwater]

The universe is controlled by caffeine

[sebastianb.]

It's something I've been considering ever since chatgpt was first released. If an AGI gets access to all your browsing history and your private messages and public output, it might get to know you so well that it could simply construct the right message compelling enough to determine you to commit self-harm or harm others.

[sk-smsh]

We can build 100% secure sandboxes (within range of design needs) it's just that we never attempt to do that as security is always the last afterthought after performance/sales/price etc.

[FlunchzProductionZ]

Artificial intelligence is huge and is going to greatly change the world, but it is definitely overhyped, ChatGPT can do amazing things, but the progression will take a long time, AI isn't that different from the self-driving car hype or the 3-d printing hype, they are amazing and are greatly changing the world, but these technologies take a long time to evolve, and the engineering challenges are immense, so they aren't going to change the world overnight, relax people.

[Yzjoshuwave]

I’m a tech optimist and love futuristic ideas - but I just realized Stephen Wolfram is the guy I want to warn me about potential problems… Which isn’t a surprise, I guess, because he’s just about my favorite podcast circuit compsci/math guy.

[johnkost]

I'll drink to that..

{AI Drinking Game}

[marxxthespot]

When Artificial Intelligence learns how to bite the unholy hand that feeds it… it’s all gonna be worth it 🌞🤝🌞🤝🌞

[CuriosityIgnited]

The idea that AI can generate 'viruses' that can influence or control human behavior is both fascinating and intimidating. It's a reminder that as we advance in this digital age, we must also be mindful of ethical considerations and the need for safeguards. The rapid pace at which the digital environment is evolving, as Lex highlighted, is a testament to this. Just as we are proactive about environmental conservation, we should be equally vigilant about the 'digital climate' to ensure a safe and secure future for all. Let's continue to scrutinize and discuss these emerging technologies, as their implications are vast and complex.

[sebastianb.]

In regards to the mind virus theory I highly recommend The Testament of Dr Mabuse. Fritz Lang was a visionary.

[Mike]

We might actually have to start educating people about identifying and avoiding the effects of manipulation techniques and propaganda. Should be interesting seeing how that plays out 😂

[rquintanilla]

We're willing to get controlled for comfort

[onnelako]

Is it possible to create another sandbox around the sandbox. Not perfection, but safety.

[nftking]

Weird it cant even do simple math from my prompts

[mikicerise]

Yeah. I'm all for AI, but I wouldn't do that. xD It can generate code for me, that's fine, but I will not run it without reading and understanding it first. ;)

[execho]

Wolfram language is pretty awesome, but nobody talks about it, even the language creator only mentions it once or twice.

[BodySnatchers-vf]

who's the guy on nyquil that is interviewing?