This is a re-post of an article that I wrote for a now defunct blog almost ten years ago. I feel that it has become more relevant given the recent rise of artificial intelligence as well as its rapid and widespread adoption across numerous industries.

Original post (February 20, 2016, via the Kindred Robot blog):

I went to see a the keynote address of Rise of the Robots. Technology and the Future of Work presented by Martin Ford at Mount Holyoke College this weekend. Ford is an entrepreneur and author who has brought to light a selection of very interesting trends in employments and their relation to automation.

---

---

Image: A sketch of a robotic arm built from a frame of interwoven oak branches

It's springtime and the frost of winter has yielded to the explosion of buds and blossoms across various species of plants and trees. Recently while cleaning, I came across an oak twig I had saved sometime last year, having whittled the bark off of it, the twig seems exceptionally strong. This left me thinking about the attributes of various materials, and wondering how something like oak compares to items such as aluminum or carbon fiber when it comes down to evaluating their strength vs their costs.

Relative Cost

Could variables such as availability, cost, and strength leave wood outperforming metal? The difficulty in answering this question is greatly increased by relative costs. For example, location often affects the price of materials - wood not native to an area costs more to ship it to that location. Likewise, scale plays a tremendous factor. It might be efficient for one person to harvest oak branches to build a structure, but less practical to do that at an industrial scale where farming methods and machinery become necessary.

In my case, I want to determine a rough cost evaluation that's just relative to me. I'm just getting twigs and branches from naturally planted trees using regular/sustainable trimming and pruning methods). Additionally, mining ore and refining it might require tools, energy, and materials not currently available to me (although I'm up for attempting this for a future blog post).

Flexural Strength

Given my lack of testing equipment, and materials, I've compiled the following table from a variety of sources. Not that many of these are approximations or can vary significantly depending on how the material is treated, tempered, etc. Additionally conditions such as temperature can impact the strength of materials and how/when they fail.

Table References:

1. https://www.atlasfibre.com/understanding-flexural-strength-guide-to-flexural-strength-in-materials/
2. https://www.ijert.org/research/processing-and-flexural-strength-of-carbon-fiber-and-glass-fiber-reinforced-epoxy-matrix-hybrid-composite-IJERTV3IS040781.pdf
3. https://www.smicomposites.com/carbon-fiber-cost-factors-that-influence-the-most/
4. https://alvibel.pl/en/what-determines-the-strength-weight-of-plywood/
5. https://workshopcompanion.com/know-how/design/nature-of-wood/wood-strength.html
6. https://blog.lostartpress.com/2021/03/21/buying-wood-by-the-pound/
7. https://pmc.ncbi.nlm.nih.gov/articles/PMC4233722/

The idea of machines expressing human traits is fascinating, and I think the challenges around this make some of the most compelling characters, at the very least in science fiction. Perhaps there is something inherently relatable about the challenges humanity faces that we recognize in robots and computer models when they begin to encounter and overcome similar obstacles. As these technologies continue to progress, I suspect we'll push the boundaries of what we know about ourselves even more than what we learn about designing computer models.

I've seen a few articles that cite a recent study on how large language models (specifically GTP-4 in the case of this study) begin to behave differently when presented with emotionally-charged, or stress inducing information. Specifically they become less likely to behave objectively and trend towards expressing biases latent in the data they've been trained with.

One suggested explanation is that this is in-part a "trait" of the LLMs. They've been trained using vast amounts of human generated data, and so they mimic the pattern that humans follow in similar situations. Somewhat humorously, the study showed evidence that using mindfulness meditation techniques on these LLMs helped to "relax" them and counteracted the trend towards biased behavior.

While the study focused on more of a black-box method of working with the GTP-4 model using prompts, I think there's something inherently interesting about the nature of the type problems that were presented to these LLMs.

Recently, I've been working on adding support for vector databases to the chatterbot Python library, and so naturally I've been giving a lot of thought to the idea of vector representations of information. But what are vectors used for in LLMs? In a general sense, "words" as the LLM understands them are represented as vectors, or rather a number that represents that word's relationship to other known words. As a small example, we can consider the relationship between "speaking" and "writing" alongside actions that could be synonymous with either:

        (Speaking)
            ●
           / \
          /   \
(Poetry) ●-----● (Rhyming)
          \   /
           \ /
            ●
        (Writing)

Speaking and writing have very close contextual relationships but on the other hand something like "running" or "battling" would be a significant distance away from these terms. 

Vector relationships like these may be relevant to the responses of LLMs becoming less optimal when presented with stressful information when we start to think about the attributes of those types of situations. Our own brains excel when faced with information and scenarios that we're familiar with, but begin to experience stress when faced with fundamentally challenging data. Much like the vector spaces, stressful scenarios involve points of information that are further apart, they are challenging because it takes greater effort to try to reconcile it issue at hand.

The distance between vectors is still a bit speculative on my part, but might be worth some experimentation to better quantify. More likely than not, the landscape of information shared between humans and machines is highly nuanced, and shaped by an extensive range of both strong and subtle forces.

Its been a minute, hasn't it? Recently I noticed the sunlight streaming in through a cellar window, perfectly gleamed off the collections of spiderwebs that have accumulated around Salvius. It's been about eight years since I've really focused heavily on anything robotics related. School, work, and just life in general took me in more directions than I can reasonably enumerate in a casual blog post like this one. Within the last few months my main projects have been a photo booth (complete with its own app, and hosting its own wireless hotspot complete with local DNS), as well as a separate photo sharing app for events (check it out at nimbusqr.com if you're interested). Both of the aforementioned projects are going to have their debut at my wife and I's upcoming wedding.

This weekend I finished up the new neck joint for Salvius. After traveling around last summer with the robot it quickly became apparent that a few modifications needed to be made. The main problem that I was facing was that the neck was too unstable to support the robot's head. As a result the head wobbled around quite a bit during transport and this led to some of the joints and connectors becoming damaged.

The new neck design I needed to create had two main requirements. First, it needed to fully support the weight of the robot's head. Second, I the new neck should allow the head to be easily detachable. I added the second design parameter for the purpose of making it easier for my self to disassemble the robot for transport and development when needed.

To accomplish this, I moved the universal joint in the robot's neck, which allows for left, right, forward and backwards tilting of the head, closer to the back of the robot. By increasing the distance between the universal joint and the linear supports that move the head I was able to improve its stability. The second major change I made was that I moved the neck rotation servo from the base of the neck, up into the head itself. This alteration was to help ensure a more modular design of the head and neck. Last, a metal axle with new bearings was added to the top of the universal joint to ensure a smooth movement as the neck servo rotates the head from side to side.

The Intel Edison comes ready to work with Python, but most Python developers would prefer to have access to PIP for managing their installed python packages. I've recently gone through the process of installing PIP on the Intel Edison and wanted to post the instructions for anyone who wants to do the same.

Connect to the Edison

Start by connecting to the Edison through the serial port. I'm running Ubuntu, hence the sudo screen.

sudo screen /dev/ttyUSB0 115200

Then you will want to make sure that your Edison is connected to the internet.

configure_edison --wifi

You will have to wait a few seconds, then select your network and enter your password.

Add Third Party Libraries

Intel Edison is powered by Yocto but pip is not in the official repositories, therefore we need to add the unofficial repositories. We began getting the latest Intel IoT Developer Kit libraries.

echo "src intel-iotdk http://iotdk.intel.com/repos/1.1/intelgalactic" > /etc/opkg/intel-iotdk.conf

Begin by adding the repositories using vi.

vi /etc/opkg/base-feeds.conf

Add the following three lines:

src/gz all http://repo.opkg.net/edison/repo/all
src/gz edison http://repo.opkg.net/edison/repo/edison
src/gz core2-32 http://repo.opkg.net/edison/repo/core2-32

Save and close the file. Then update the package list.

opkg update

PIP is now installed, however it requires setuptools to be installed before PIP can install any Python packages. The setuptools package is now included with the latest version of pip. To upgrade pip to the latest version you can now run

pip install --upgrade pip