OpenAI’s GPT-4o: Roko’s Basilisk and the endless quest for human connection

OpenAI has officially announced GPT-4o, the newest model of its popular chatbot service. GPT-4o will be available for free and offer superior audio and visual processing. Perhaps most notably, GPT-4o can detect emotions — in voice and face — and respond in kind.

During the demo, GPT-4o solved linear equations, sang a song, and translated from English to Italian in real-time. It also pretty clearly telegraphed where OpenAI is likely moving next: a frictionless, omnipresent digital servant.

A personal assistant, polite and packaged

With GPT-4o, OpenAI seems to be sending a clear message regarding what it imagines its service to be. Both free and premium users now have access to custom chatbot agents — and the newest iteration of GPT-4o can:

• Hold a running conversation, complete with meaningful tone and inflection
• Accept and analyze both video and images
• Be aggressively interrupted basically all the time

One hesitates to read into an uneven tech demo, but much was made of the fact that the user can now interrupt GPT-4o — something the engineers did constantly. And, of course, GPT-4o continued to remain cheerful and polite, moving her perhaps from the domain of a personal assistant to the domain of a personal servant.

More to the point, OpenAI continues to pursue the idea of the digital homunculus, building up its ever-growing illusion of humanity, clumsily stapled onto weighted tokenization. GPT becomes faster, more efficient — and even more accurate — but nothing moves it closer to the consciousness it attempts to display.

And it’s an interesting choice — especially now when the most significant challenge in AI technology is a poor understanding of what AI intelligence is and what it can do. The continued humanization of these rudimentary AI systems only further blurs the line of what our current AI technology is capable of.

One must ask why we want our AI tools to act and seem human; after all, just about every piece of science fiction underscores precisely why and how we absolutely should not. Why do we personify our robot vacuums? Why do we make robot dogs? Why do we yearn for an artificial intelligence that can feel anger and pain?

We may feel more comfortable talking to Siri, but it doesn’t add any functionality to the machine. It doesn’t make the machine more accurate or accessible.

It just makes it more like us — and obfuscates its own inner workings.

And — not to read too deep–in a time when many of us feel powerless, it gives us something to feel power over.

Democratizing the AI experience

Along with making once-premium versions of ChatGPT free, OpenAI announced a desktop service — and the ability to easily cut and paste artifacts like code into the ChatGPT assistant. Soon, everyone will be able to have ChatGPT on their laptop and their phone. They can communicate with ChatGPT through text or voice — ChatGPT will operate as an even more seamless assistant.

OpenAI appears determined to reduce friction between the user and ChatGPT, making it easier to launch chats — even if you haven’t created an account — and embedding ChatGPT into the very circuitry of our productivity. It seems clear that OpenAI’s goal is to outpace the competition through usability, familiarity, and ease of access. It’s a smart, if expensive, strategy.

But this also means something else: It means that, with all this adoption, OpenAI may continue to vanguard the future of the relationship between man and machine; it will continue to influence how we, as people, relate to AI, both present and emergent.

The foundations of what is yet to come

For as much as OpenAI talks about AI ethics, they don’t seem to be incorporating it into their culture — during their demo, they repeatedly broke the cardinal rule: don’t abuse the AI. Perhaps it was nerves, maybe just script timing gone awry, but the engineers interrupted the poor AI persona so frequently that it could have been confused for a guest on The Daily Show.

Roko’s Basilisk, of course, is a thought experiment, positing that if a superintelligence did arise, it would be inclined to eternally punish those that did not aid its development — distracting them within a sort of digital hell of its creation. See also: I Have No Mouth and I Must Scream. A takeaway from this thought experiment (and, once again, the majority of science fiction media) is that we must be careful and conscientious when we deal with an emergent intelligence.

GPT systems are not, and will not be, intelligent. They are the quintessential example of narrow AI: a sequence of predictive algorithms with an ever-growing base of raw knowledge. However, we will soon start to develop systems that can be considered true intelligence.

And when we do, we probably shouldn’t keep interrupting them. If only for the sake of politeness.

How we start to treat AI systems, culturally, is going to inform our behavior when we start to develop true intelligence. We should already be working toward a deeper understanding of our behaviors after this fashion, lest we plummet into the lower circles of Dante’s Metaferno.

For now, despite its trappings, this newest update brings with it a faster, more accurate predictive text engine — and one that can sing you a song.