The Anatomy of AI is a great piece that examines the hidden costs associated with AI and personal assistant devices. Specifically, the diagram above “visualizes three central, extractive processes that are required to run a large-scale artificial intelligence system: material resources, human labor, and data.”
A few fun (and depressing) quotes indicating the disparity of resources between the persons at each end of the processes:
Amazon CEO Jeff Bezos, at the top of our fractal pyramid, made an average of $275 million a day during the first five months of 2018, according to the Bloomberg Billionaires Index. 17 A child working in a mine in the Congo would need more than 700,000 years of non-stop work to earn the same amount as a single day of Bezos’ income.
Regarding the methods used to ship components:
Even industry-friendly sources like the World Shipping Council admit that thousands of containers are lost each year, on the ocean floor or drifting loose. 29 Some carry toxic substances which leak into the oceans. Typically, workers spend 9 to 10 months in the sea, often with long working shifts and without access to external communications. Workers from the Philippines represent more than a third of the global shipping workforce. 30 The most severe costs of global logistics are born by the atmosphere, the oceanic ecosystem and all it contains, and the lowest paid workers.
Wow. I really loved The Future We Wanted—it’s by far my favorite piece we’ve read for our Unconventional Uses of Voice Technology class. I love science fiction in general, but I especially love scifi that explores the relationship of human beings to machines and raises deeper questions about what it means to be human. The Future We Wanted hits both of those themes, and was both vulnerable and emotionally provocative.
The protagonist, Polly, exists in a near-future world in which an AI is considered to have a “virtual identity” and “personhood”. This world is infused with oppressive sexist overtones: Polly is expected to perform all of the emotional labor in her family’s household, and is later told by her husband that she would like their household AI system, Augusta, more if she connected with the “womanhood” of it.
Her women’s therapy group—“women helping women”—isn’t much help either. Polly gets vulnerable about her struggles with Augusta—how it makes her feel like a failure, how she doesn’t see it as a person—and another group member is allowed to cross talk and hurl accusatory statements at her. When this escalates into a back and forth between the two women, the group facilitator’s response is to silence both of them and end the group…not a very safe space.
And, at the same time, this world (or at least the people in it) purports to be post-sexist. “Thinking of [Augusta] as sexist is a dated framework,” her husband tells her, later half-heartedly chiding their son for making “appearance-based judgments” about Augusta.
As time goes on, Polly starts missing group therapy to pay the bills while her husband and kids sits around and play games with Augusta. Poor Polly seems to have no in her corner. She is gaslighted by the people in her world and given useless platitudes. Ultimately, this dissonance leaves Polly feeling so alienated and disconnected from both the humans and AI she destroys Augusta. Bravo.
Like any wood project that comes off of a CNC router, the Asana bot needs a lot of clean up and finishing work to shine. The majority of this is sanding and oiling before finishing with wax (or another top coat).
My current process is:
Take job off 4-axis mill and cut parts from larger block
Sand off tabs
Sand the parts from 220 -> 400
Inlay brass powder on body with CA glue
Sand body + brass to even 400
Oil parts with tung oil
Repeat #4 every 24 hours until I like the color
Sand from 400 -> 600
3 step wax on lathe buffing wheel
So much sanding. So much tung oil. Starting at 220, going up to 600 prior to waxing.
So far, the biggest challenge has been getting the brass ohm inlay right. I’m using brass powder and cyanoacrylate (CA) glue rather than attempting to inlay a solid piece of brass. My current process is: clean up the inlay area with high grit sandpaper (400+), put multiple clear coats on top, add brass powder, add CA glue, sand. Ultimately, getting a great result seems to come down to getting 3 variables correct: cut depth for the inlay, amount of clear coat before attempting inlay, and amount of inlay powder used.
For me, 2 thin coats of clear coat has been a good place start. I’m still having issues with the powder getting deep into the open grains of the walnut, but it’s significantly better than early attempts with 1 coat. 1/16”, cut with a 1/32” bit, has proven to be a good depth for the inlay cut.
The Yoga Robot 9000 has official been renamed the Asana Bot (for now)—asana is a sanskrit word for posture that is used generically in yoga.
I choose walnut for the head and body and baltic birch ply for the arms and legs, to give a nice light-dark contrast—inspired by Grotesk’s “A Bigger Big Poppa”.
I experimented with gluing up baltic birch scraps at different angles, but ultimately ended up using single 3/4” scraps for the first few prototypes.
I definitely want to experiment with this more down the road to see what kinds of patterns I can produce in the legs and arms. I think it’s also worth trying some glue ups with acrylic inside.
Legs and arms were cut with round 1/8” bit, body and head were cut with flat bits from 1/32” up to 1/4”
Getting the CAM right on the body/head combo was super important, and I eventually cut job time from 2.5 hours to just over 1.25
For our first Dialogflow assignment, and made a silly bot that helps you decide which team to root for in the playoffs this year. Download a .zip of the bot here
Spoiler alert: it hates the Yankees and the answer is always the Astros.
The Hanson Robotics / University of Texas paper Upending the Uncanny Valley was an interesting and somewhat frustrating read. While I ultimately agree that the pursuit of humanlike robots is a worthwhile endeavor (and welcome the advances these groups are making toward that goal), I had a hard time focusing on the majority of the content in the article because I found myself angry at the psuedo-scientific manner in which the concepts were portrayed. The article begins with a statement that "the myth that robots should not look or act very humanlike is a pernicious one in robotics research, one commonly known by the term 'Uncanny Valley'." I'm not sure how they came to this definition, but I would argue that it is both wrong and misleading.
As far as I'm aware, the uncanny valley refers to depictions of humanoids that are somewhat realistic, but not realistic enough to be truly convincing--leading to cognitive dissonance and a negative emotional response from most human beings. This is different from the claim that "robots should not look or act very humanlike". Essentially, humanoid robots should be purposely cartoonish or convincingly lifelike, otherwise they are creepy.
I do appreciate the fact that "the robots of David Hanson do not tiptoe around the uncanny valley, but dip in and out of the uncanny in attempt to chart the territory and its boundaries", as I think this is the best way to make progress in building lifelike androids.
The paper further notes that "the Uncanny Valley has never been studied with real people", so, they wanted to "put the theory to the test with human subjects." However, their approach to doing this is to run a set of web surveys, the results of which they then interpret to mean that "there does not appear to be an inherent valley."
Unfortunately, the article makes no mention of the survey design or methodology, which left a bad taste in my mouth. It seems to provide a very convenient vehicle for a robotics company that builds realistic-ish robots to push their agenda.
In order to determine if their findings are worthwhile, it would be helpful to know several things, such as: How large was their sample size? How were these people selected--was it randomized? Were the people chosen already familiar with Hanson Robotics' work? If so, were they more likely to find a psuedo-realistic robot agreeable versus a random person who would find it uncanny? How does their familiarity with robotics and/or the robotics industry prejudice their viewpoint? Do these same people react differently to psuedo-realistic robots in person versus on a computer screen?
Without the answers to those questions, and a more in depth analysis of the results, it seems disingenuous to claim that no valley is inherent and "a new theory is called for."
Continuing work on my original character for Art Toy Design, current named Yoga Bot 9000. Bringing him into 3D was a fun and interesting challenge--how does a robot with no elbow/knee joints do yoga?
I worked in CAD/3D printing and modeling clay this week, which worked out pretty well. Modeling clay is a nice addition to 3D printing for rapid prototyping.
Mocked up in CAD, testing various leg designs
Initial head print came out looking a bit too much like the reddit alien, Snoo
Rescaled to get a better head to body ratio and less reddit-ness
Coming together with modeling clay
Trying out various yoga poses, up first: down dog
One legged tree
And finally, shivasana (corpse pose)
I initially planned to build a "misfortune generator", where a user would click a button and be given a bad fortune ("It is too late for you to do anything about climate change..."). But, after spending thirty minutes trying and failing to build a large collection of mis-fortunes, I decided to change directions.
From a technical perspective, I grabbed a collection of some of the most popular definitions from the book, did some basic data munging in excel, converted the data to JSON, and used a simple sorting algorithm to "randomize" the data. It probably would've been easier to just use a Math.floor() / .random() and pass that into the array index, but I didn't think of that until after the fact, and I'm not sure if the result would have felt as random to the user.
The Problem My stereo receiver is from the 70s and has a single AUX input. When we have people over, there is no simple way to let people play music without unplugging TV connection, adding an RCA -> 1/8 Stereo cable, and making people plug in directly/leave their phone by the receiver.
Add an intermediary device that people can stream to via Bluetooth, which outputs via RCA to the receiver.
Old HiFi receivers! Hardwood (or veneer) + polished metal.
Electronics Fabrication Process
For my final project for API of You, I designed and developed a system for tracking climbs I have done across the world. Development was done in Node, JQuery, MongoDB, and Bootstrap.
See it live here.
[fac_button icon="github" name="smart-starter" text="code on Github" link="https://github.com/ntwallace/climbTracker" target="_self"]
For my 4-axis skill builder I did a prototype version of my final project.
I started with a foam yoga block from REI, but managed to destroy it VERY quickly. Unfortunately, I zeroed the Z axis out incorrectly--I tried to manually make it touch the grounding pole because I was clicking the wrong button in Roland's software.
I went into panic mode after I buried the bit an inch into the material and ripped a huge chunk out of it, and completely forgot to take photos of it. Fortunately, I managed to not break a bit (soft material + sharp bit = still alive) or damage the machine.
I went back the following week with a cork yoga block and was able to run the job correctly.
The majority was cut with a 1/4" upcut bit, the small holes were done with a 1/16" bit.
Compared to my wood cut done for my final project: