(Modulating Epsilon)

I’m Sam Hiatt (he, him), a software engineer since 2006, studying AI / machine learning since 2016. I spent six years at NASA Ames, and four years at Weather Underground/IBM, implementing basic machine learning algorithms to study the natural world.

I was initially drawn to the field of artificial neural networks by an innate fascination with the underlying technology. I wanted to understand how we (humans) program machines that can “think” like we do, and I found the concept of artificial neural networks naturally compelling, particularly because of how they are (loosely) modeled after the structure of our own brains. I had a sense that if I could understand the way those algorithms worked that it might help me understand a bit more about myself.

I could never foresee how true that might become.

Large Language Models blend my special interest in natural languages with my special interest in computer programming, and the advent of ChatGPT changed the game for me (as it did for many people, I suppose).

I have experienced the potential that AI tools hold to assist with executive functioning challenges and to aid in addressing communication challenges. But the tool is a double-edged sword. The same things that make it so useful also make it very dangerous.1 (#meta it seems like I need to provide some kind of concrete example of the danger)

I created Modulating Epsilon Consulting to help therapists, coaches, and teachers understand how AI tools can be used to build shared understanding, to improve human connection, while steering clear of the risks.

Modulating Epsilon is mindful exploration.

It’s body awareness, as a prerequisite for understanding.

If understanding isn’t shared, then it leads to isolation.
I aim for AI interactions to bridge human connection, to help us find common ground with each other.

def epsilon(ε):

noun:

the fifth letter of the Greek alphabet ( Ε, ε ), transliterated as ‘e.’2

symbol:

permittivity — the ability of a substance to store electrical energy in an electric field.3

symbol:

in reinforcement learning — a subfield of AI that studies how systems learn through trial and error — a parameter that controls how often a system explores the unknown instead of repeating what has worked before.
A small ε means relying primarily on current knowledge, whereas a larger ε means more willingness to try something new.4

“A simple alternative is to behave greedily most of the time, but every once in a while, say with small probability ε, instead select randomly from among all the actions with equal probability, independently of the action-value estimates.” — Sutton & Barto5

def 米(mǐ):

Chinese character.

noun:

a grain of rice; a shelled seed.6

symbol:

an idea — the fireworks in my mind.

def 眯(mī):

Chinese character. Components: 目 (eye) + 米 (rice/seed).

verb:

to narrow the eyes; to squint.6

symbol:

what happens when attention meets an idea. Narrowed perception. Focus.

def 撒(sǎ):

Chinese character. Components: 扌 (hand) + 散 (scatter).

verb:

to scatter; to spread; to let go.6

symbol:

the act of sending ideas outward — from the mind, through the hand, into the world.

散 — Components: 龷 (split/divide) + 月 (moon) + 攵 (strike/tap).
sǎn: free and unfettered; natural and at ease; loose; scattered; careless, inattentive.
sàn: to break up; to disperse; to scatter; to distribute; to disseminate; to dissipate; to find diversion; to walk around, to stroll.

References

  1. [[Dohnány et. al. - Technological folie à deux]] ( #TODO replace with external reference/citation) ↩

  2. “Epsilon, N.” Oxford English Dictionary, Oxford UP, July 2023, https://doi.org/10.1093/OED/5727044399. ↩

  3. “Permittivity, N.” Oxford English Dictionary, Oxford UP, July 2023, https://doi.org/10.1093/OED/1098496860. ↩

  4. This definition was drafted by Claude (Anthropic) and edited by Sam Hiatt. ↩

  5. Sutton, R. S., & Barto, A. G. (2018). Reinforcement Learning: An Introduction (2nd ed.). MIT Press. Section 2.2, “Action-value Methods.” ↩

  6. Definitions sourced from Pleco Chinese Dictionary and Wiktionary. ↩ ↩2 ↩3