When it comes to artificial intelligence, there’s almost no telling what machines will be capable of in the next month, year, or decade. I say “almost” because there’s a definite ceiling on AI potential.
But, before I get into that, let us bow our heads and intro this issue:
- No fate but what we make (404 words)
- Try-all and error (132 words)
- A prayer for the living (33 words)
No fate but what we make
Yesterday I nonchalantly mentioned how cool Boston Dynamics’ latest choreography routine was. And, don’t get me wrong, those robots are awesome. But it’s always bittersweet whenever the company releases a new video because the general public is really, really, REALLY dumb about AI technology.
If I had a training token for every time someone supposedly working in the STEM industries said something along the lines of “if you put a GPT brain inside of a Boston Dynamics robot it’d be game over for humans!” … I’d have enough tokens to train a new LLM model to tell all those people to stop hyperventilating.
The fact of the matter is that AI can’t do anything that a human being, given sufficient time and access, could not do. Cramming GPT into a robot would just give you a hunk of metal capable of outputting language. We already have that; it’s called “whatever hardware you currently use to access ChatGPT or other language generators.”
People get so excited over progress that they seem to lose their basic common sense abilities. ChatGPT is awesome. Atlas (the BD robot) is awesome. But that doesn’t mean we can conflate the two technologies into something greater than their parts.
Confusing AI’s predicted future capabilities with humankind’s fears and fantasies is a path that only leads to hyperbole and the kind of ‘technological winter weather’ that the self-driving car industry is currently dealing with.
I could list off some of the myriad ways people (even some experts) mistake machine potential for human fantasy (gaydar and predictive policing come to mind).
But, instead, I’m just going to issue an open, public challenge. Tag me on Twitter (@mrgreene1977) if you think you have an example of something an AI can do that a human, given sufficient time and access to tools, could not.
- Speed isn’t a sign of intelligence. I don’t care if an AI can sort through a billion files in two seconds. The question is, is the model doing something with that data that a human, given enough time to perform the same functions, could not?
- Submissions must be scientifically testable. Don’t @ me with some crap about Gaydar, for example, unless you have a scientific method for determining queerness independent of self-reporting.
- Don’t confuse artificial intelligence with physical robots.
Prizes: Winners, if any, will have their tweets featured in a mea culpa issue of More Than Meets the AI.
Try-all and error
This is the dawning of the age of AIquarius
Age of AIquarius
Harmony and understanding
Sympathy and trust abounding
No more falsehoods or derisions
Golden living dreams of visions
Mystic crystal revelation
And the mind’s true liberation
That, folks, is the kind of random tangent that AI doesn’t go off on in the middle of a news column for no reason. The 5th Dimension is a human thing, AI wouldn’t understand.
Anyhoo… if you’re reading this, you’re probably interested in learning more about large language models. And, if you’re interested in LLMs, you should sign up for Cameron Wolfe’s newsletter. They take deep dives into various models and their papers in order to explain what’s really going on.
And, this is shameless self-promotion, but you should also sign up for my wife’s newsletter, Love Yourself!.
A prayer for the living
Dear DNA coursing through our chromosomes,
Oh great calories powering our minds,
I beseech thee, deliver us from…
oh nevermind, I just skipped breakfast
that’s something I can take care of myself