The Genie Is Out. Pandora’s Box Is Open. [Thoughts on Generative AI]
A paradigm shift like no other. There is no going back.
I have this pet theory that mega-successful tech products have to be one of these three:
- Genie — grants your wishes without any effort. Just say it, and it will come. Example: Uber —Take me to my desired destination, or Tinder — Match me with a hot date.
- Superpower — amplifies your abilities to superhuman levels. Example: Grammarly — never make a grammar mistake again, or Evernote — remember everything.
- Guru — teaches you stuff that you never knew. Example: Youtube — get access to the best teachers, or Stack Overflow — learn to write better code.
ChatGPT is a revolution because it is all of the above. A true unicorn that combines genie, superpower, and guru in one tool. (Is tool even the right word for it?) It grants your wishes, amplifies your abilities, and teaches you stuff that you never knew.
Small wonder that it is the fastest anything in history to reach 100 million users. I haven’t met anyone who has used it without coming away with admiration, awe, and even a bit of fear. Almost like an encounter with an alien intelligence, or a divine oracle.
A Copernican Moment
At its core, the large language model powering ChatGPT sounds simple enough. It uses statistics to infer and respond with the most plausible words based on a training data set that basically spans the entire web. It just puts one word after another and glues them together. Autocomplete on steroids. Miming.
That’s it? But why does it seem so damn smart? Well, GPT ingested so much of our writing that it somehow uncovered the hidden shapes and contours of language — underlying representations of text that we were never able to perceive. That’s it. (The law of large numbers strikes again!)
There is a more downbeat implication to this. So much of what we consider deep thinking may in fact simply be putting one word after another and gluing them together. Mixing, matching, and remixing. The coveted human skill of expanding and contracting text is turning out to be terribly easy for a computer. Creativity and understanding may be more mechanical than we thought. From Scott Aaronson’s blog:
The discovery of just how much of humans’ language understanding can be simulated by a probabilistic autocomplete is entirely comparable in magnitude to the discovery of the moons of Jupiter — so no surprise that it engenders a similar opposition.
This is a true Copernican moment for humanity. We who consider inspiration as whispers from the ancients, who deem creative pursuits as a sacred space beyond the cold metallic reach of computers… we may not be that special after all. If the sun does not revolve around the earth, why would the muses bother circling around our minds? Our species met its match with a giant corpus of text, good old statistics, and farms of graphics chips going brrr.
Back to the Command Line
The giant corpus of text that we handed it as an offering to the AI gods? We are all complicit. Everything we published online fed the machine. The public web from its inception in the early 90s to 2021 was effectively a training phase, a bootloader to generative AI. So the web was the welcome party, and we are only now getting to the main event.
For all that we gave up, what do we get in return? Generative AI is paving a whole new way of interacting with computers, a leap akin to the emergence of graphical user interfaces after the command line. From Bill Gates’ The Age of AI Has Begun:
In my lifetime, I’ve seen two demonstrations of technology that struck me as revolutionary. The first time was in 1980, when I was introduced to a graphical user interface — the forerunner of every modern operating system, including Windows. […] The second big surprise came just last year. […] It (ChatGPT) wrote a thoughtful answer that was probably better than most of us in the room would have given. The whole experience was stunning. I knew I had just seen the most important advance in technology since the graphical user interface.
What’s funny is that we are kind of going full circle back to the command line. We interact with these super-AIs like ChatGPT or Midjourney through chatboxes. We type a bunch of text, and the computer responds. The difference is that unlike the command line, generative AI does not respond in a deterministic way. The prompts act like incantations that uncover magical new capabilities in probabilistic fashion.
Now you may argue that conversational interfaces are nothing new. Siri came out in 2011, after which came a deluge of chatbots. But there is a reason we called those assistants, and we now call GPT and its ilk as co-pilots. They are less parlor tricks, and more colleagues and companions who could make sense of anything we say and give responses ranging from intelligible (at the low end) to mind-shattering (at the high end). That impact cannot be overstated.
A Gutenberg Imprint?
Let’s stretch it even further. Can generative AI have a similar impact to the printing press? In the world of software creation, it may have a shot. Coding until now was an intense, laborious activity, similar to how the old scribes hand-wrote the holy verses. Limited access to books meant that few learned how to read. The printing press obliterated all those barriers and changed history.
With generative AI, natural language is code. English is the programming language. Programming is such a killer use case because it plays to the AI’s strengths — grammatically structured, with clear testable outcomes. Non-coders can become coders, and trained coders can become 10x-100x coders. This will lower barriers to software creation to an extent yet unseen, and this is where we see imprints of Gutenberg. Software will eat more of the world, unshackling those previously saddled with technical debt and prohibitive developer costs.
From SKVenture’s Society’s Technical Debt and Software’s Gutenberg Moment:
A software industry where anyone can write software, can do it for pennies, and can do it as easily as speaking or writing text, is a transformative moment. It is an exaggeration, but only a modest one, to say that it is a kind of Gutenberg moment, one where previous barriers to creation — scholarly, creative, economic, etc — are going to fall away, as people are freed to do things only limited by their imagination, or, more practically, by the old costs of producing software.
Every era of computing so far led to an exponential increase in the number of software applications — from a few dozen with mainframes, to tens of thousands with PCs, to millions with mobile, and now potentially hundreds of millions with AI. We live in heady, exhilarating times. Let a billion flowers bloom.
To be continued: What impact will generative AI have on jobs, on society, on our relationship with technology and other human beings?
