How to Become a Centaur (2018)
jods.mitpress.mit.edu
jods.mitpress.mit.edu
Hah. I injured my hand a couple of years ago, and had to type with one hand for a couple of months. I could code review, I could participate in meetings, I could identify the source of a bug. But I could not write any but the simplest code. I was mystified. I could write emails and to some extent documents (typing of course slowed down a lot). But I became basically incapable of writing code. Using too much brain power on the physical act makes you have less left over for the CONTENT. Cool to see that this problem was understood much earlier.
We have such a fine grained control over our body, we should be able to find a signal stream or two to interface with some computer system.
cat /dev/bra1n > /dev/null 2>&1Is there any research being done into more efficient typing? I mean beyond different keyboard layouts (I've never tried dvorak and co) and keyboards (I like my wireless mac keyboard well enough).
I know visual languages don’t get much love around here (largely for good reasons), but I wonder would that be a good solution to what you describe, especially if using a touchscreen rather than a mouse.
> At first, Garry wasn’t surprised when a human grandmaster with a weak laptop could beat a world-class supercomputer. But what stunned Garry was who won at the end of the tournament — not a human grandmaster with a powerful computer, but rather, a team of two amateur humans and three weak computers! The three computers were running three different chess-playing AIs, and when they disagreed on the next move, the humans “coached” the computers to investigate those moves further.
> [..]
> When you create a Human+AI team, the hard part isn’t the “AI”. It isn’t even the “Human”.
> It’s the “+”.
"the human chooses the questions, in the form of setting goals and constraints — while the AI generates answers, usually showing multiple possibilities at once, and in real-time to the humans’ questions. But it’s not just a one-way conversation: the human can then respond to the AI’s answers, by asking deeper questions, picking and combining answers, and guiding the AI using human intuition."
Example:
"In 2016, Zhu et al created a painting tool where you draw in the rough outlines, and an AI photo-realistically fills in the gaps. The human and the AI have an artistic conversation through pictures. The human can draw some green lines on the bottom, and the AI replies with several possible photo-realistic grassy fields to choose from. Then, the human can draw a black triangle above that, and the AI replies with several pictures of a mountain behind a grassy field. Through this push and pull between human & machine, art is made."
For games like Go, where computers are not yet capable of viewing the full depth of the possibility tree, I imagine that human+computer is still a viable strategy.
The top Go player in the world retired saying that he thought he was almost perfect, but then a "superior being" showed him he wasn't even close. I doubt there is a human on the planet that can improve the strength of the best AI at Go.
Perfect solutions to games is a 40 year old approach. It's something you learn in school, so you can quickly see that brute-force searching is not feasible on even small problem.
What it's doing is using the computer's ability to randomly test various lines of inquiry many, many times faster than a human being can, and then use statistical methods to help rule out the lines that cannot produce anything of value.
That's why it's Monte Carlo Tree Search.
That said, any sufficiently advanced technology is indistinguishable from magic [0], and with modern Machine Learning techniques we are getting closer to crossing the line over to "magic" in certain fields, such as Chess.
[0] Clarke's Third Law, see https://en.wikipedia.org/wiki/Clarke%27s_three_laws
One wonders whether the inherent approach in AlphaStar (supervised learning) can only ever hope to imitate the best, but surpassing the previous best seems to still be a human domain for now, at least in SC2.
> Unfortunately for the article's point, despite a brief moment of viability, even human+computer chess teams are now far outclassed by computers alone.
and it then continued:
> The top Go player in the world retired saying that he thought he was almost perfect, but then a "superior being" showed him he wasn't even close. I doubt there is a human on the planet that can improve the strength of the best AI at Go.
I was making the point that, even though there have been AI advances since the article was written about chess, defeating its immediate point about chess, these advances have also come with their own caveats. ("Oh this technique that beats chess perfectly can't hope to beat Go". "Oh this technique that beats Go perfectly can't hope to beat StarCraft".) These caveats serve to reinforce this article's broader point - that AI has limitations, that humans can complement, and vice versa.
The key piece missing in chess was how to know what positions were worth searching with a computer. A human had that intuition, and so a 'centaur' could be better. But that hiccup was solved by AlphaGo, using the expectation neural network.
About the only thing left for humans to do is deciding which problems to try to solve.
Best line I've heard this month.
It felt amazing the day I started wearing a Pebble watch and my friend got his Google Glass -- by that evening we were no longer just two people walking along the streets of San Francisco.
He was there, tapping on the device around his head, taking pictures and posting them into our friends' Google Hangout without even having to pause and pull out a phone. A quick buzz later, I glanced down at my wrist and told him which of our friends had responded in the room and relayed the message, and within seconds we were doubled over in laughter.
What previously required opening up a laptop, or at least a phone, in order to take a picture, select that picture, send that picture, and wait for a response, and we were now able to have that entire loop of feedback in under a minute, without becoming distracted by the rest of the device.
It was the first time I felt like a cyborg, and one of those moments I've been trying to recreate since.
Take FB ads. The best way to use it is to take full advantage of FB's AI - set a digital conversion event, and tell it to optimize for it. It will then paperclip optimize towards that.
The marketers job is to give the right conversion event + creative for the fb algorithm to do its job best.
For more typing, I carry around my own standard 104-key keyboard.
Brick + pencil is the perfect metaphor for any kind of frustration-causing depowered productivity tool