AI Unplugged: Good Enough

Experts worry a lot about #AI reaching Artificial Generalized Intelligence (AGI), the moment when AI becomes so advanced it functionally acts as good as a human. But the bar is likely far lower: if we can't tell the difference, does it matter?

The Death and Rebirth of Audio

It's illustrative to look at previous technological disruption as a parallel for AI development. In that regard, MP3s tell a story not of a superior product, but one that was "good enough."

When the MP3 format came onto the music scene, it was a demonstrably inferior product. It had lower audio quality than CDs. And yet, it was far cheaper and more portable (1/12 the data footprint). Not surprisingly, the then music giants controlling distribution vastly underestimated the impact of MP3s on their business. Part of the reason being they didn't understand how people listened to music:

As with other disrupted industries described by Clay Christensen in Innovator’s Dilemma, the incumbent players in the music sector at the time (labels, CD manufacturers and distributors, CD player manufacturers) initially dismissed the innovation as a low quality hack that would never take off. When Karlheinz Brandenburg, the genial inventor of the new format, approached the big record companies to recommend that they adopt a new copy-protected version of the file, he was “informed, diplomatically, that the music industry did not believe in electronic music distribution”. Similarly, when his group showed their prototype of the first portable MP3 player at an engineering fair in 1995, an executive from Philips told them: “There will never be a commercial MP3 player.”

The truth was that while MP3s' audio quality was inferior to CDs, most users didn't care. We listen while we're jogging, while we're in the car, while we're making dinner. We listen when there's kids screaming, highway noise all around us, while we're doing something else. MP3s didn't have to be better audio quality -- that was no longer the game -- they just had to be "good enough."

But how did MP3s come to dominate the tightly-controlled music distribution industry?

Simple: they didn't compete at all.

The Only Way to Win is Not to Play

MP3s' strength was never its audio quality. So the format's creators didn't try. Instead, the German researches at the Fraunhofer Institute gave a desktop-based MP3 encoder away for free. They hedged their bets though, with a maximum limit of twenty songs. After that, you were required to send in a registration fee to unlock full functionality of the encoder.

We all know what came next: hackers, pirates, and the mass chaos of distribution channels in the seedier parts of the Internet. What was meant to be freemium became truly free, and now the MP3 beast was unleashed and rampaging around the Internet countryside.

But without some way to guide this amorphous creature with a low-quality roar, it would have had limited range and probably starved to death. Thanks to Person-to-Person (P2P) technology, it found pathways to reach everyone. That was the rise (1999) and fall (2001) of Napster. At its peak the platform had 80 million registered users.

This Sounds Familiar...

If this sounds familiar, it's because AI is running the same well-trod paths. Released for free to the public, AI has rapidly been adopted by an enormous number of users at a much lager scale. ChatGPT currently has over 180 million users and rising.

That's not the only parallel to MP3s. ChatGPT and other generative AI create convincing text that seems human-like. Whether or not the AI is actually intelligent is beside the point.

AI does a very good job of "cold reading," the same tactic used by psychics to convince their customers that they are reading your mind. You'll see this dynamic play out in articles where AI is declared intelligent, when in reality the human is leading the AI with subtle (and often unintentional) cues. It plays on a quirk of humanity known as "subjective validation," the idea that any information that is correct has a personal meaning or significance -- correlating two unrelated events because our personal beliefs knit them together into a convincing narrative. AI does this very well. Too well.

Detecting if AI isn't a person requires language proficiency, education, and technical tools that the majority of the world doesn't possess. AI doesn't need to better than a help desk, or a therapist that costs money, or an online friend who never texts you back. It just needs to be "good enough."

And in that regard AI is already disrupting industries. Data scientists will argue about what will happen when we finally get an AGI, and it's a legitimate concern. But in terms of the impact AI will have on humanity? It's already here, giving us pretty good answers, comforting us in times of need, and available 24 hours a day, seven days a week.

Does AI need to be smarter than us? Probably not. We can't tell the difference anyway. It just needs to be "good enough."

Please Note: The views and opinions expressed here are solely my own and do not necessarily represent those of my employer or any other organization.