what ai was

Posted: April 8th, 2025 6:41 AM

The idea that machines can automate computation found genesis in the application of mathematics at scale. Everything can be tied to some number. The distance and speed at which a car travels. The number of customers who have X feature enabled and the likelihood that feature indicates a renewal. Everything devolves into compute.

Algebra, some Calculus, and Statistics join together to form modern AI. Reducing a program that can change the style of your picture into any conceivable artistic category you can think of with near perfect accuracy may be intellectually dishonest. However, these are essentially the underpinnings of the intelligence part of AI.

Use this feature set and do this linear algebra here with these weights and then take this bias and this activation function and do some calculations here and then do this vector multiplication here is the quick-and-dirty. Despite this complex process and X number of engineers working on it, there are still hidden internals about how the AI is really operating. Within these internals there are likely answers that point to how machines can think, but it is unclear what that future really looks like.

Alan Turing referred to the Lady Lovelace objection against thinking machines due to their inability in creating any novelty the way a human can. The difference between an isolated machine and a human researcher is perception. Unless a machine has a capacitive humidity sensor it can only know that the environment is humid if that data is supplied to it. Inversely, a human can know it is humid-even with awful precision-by simply existing. The machine is drastically limited by its lack of perception.

Despite coining the term 'perceptron', there is nothing a machine can quite perceive on it's own. The machines utility comes not from the algorithms, but from the data supplied to those algorithms. Thus, a machine is limited by its ability to gather information and not quite by its ability to process the information.

AI is a glorified universal calculator that ingests available recorded data and has no clear means to perceive it's non-digital environment-with even some gaps in perceiving it's digital environment. Augmenting research by shortening compute time and offering additional statistical insights that the user may not have known about before has a plethora of applications.

The novelty of invention comes, at least in part, from one's ability to notice nuance in environment. Visual, audible, and haptic feedback carry messages that bits and bytes simply cannot. A simple example of this type of inventing is Philo T. Farnsworth. Farnwsworth invented the Television.

Living in rural Idaho offered Farnsworth plenty of time to ponder over his ideas of creating a machine that could process images. A spark that guided the anchor leg of the process was Farnsworth observing the columnar structure of the crop fields. This gave Farnsworth the idea that images could be scanned in vertical lines and displayed.

His ability to visually see a pattern and apply it to what was being held in his context eventually led to the technology that has since evolved but still exists in almost every household throughout the world. This problem did not necessarily break down into a kind of computational puzzle to be solved.

Several experiments over a few hundred years led to the eventual creation of the modern airplane. Each of these experiments required gathering data, testing hypotheses, and retrying many times until the environment could be fully understood. The principles of drag, thrust, lift, and weight had to be tested and understood. There was no readily available data to perform a difficult, time consuming computation on.

In the near future there will be partnerships and collaboration between robotics companies and those who are creating the models of AI. At that point, we will look back on this time of using AI as a sort of calculator, either with regret or astonishment that advances were made so quickly.

There is likely a way that a visual system, an auditorial system, and a haptic system could be combined with an AGI/ASI brain system to create a machine that can gather data, access any corner of the internet, and perform intense computation and generative thoughts to invent some novel solution, invention, theorem, or resource. However, today, as of writing this post, we are trying to figure out alignment and interpretability; further prospects prove to be ahead of us.

The requirements of energy, money, and algorithms remain the most pressing bottlenecks to the growth of AI. If there is no energy to power the machine or model service ran via the cloud, no money to pay for the fees and materials, and no algorithms to continually improve the models then AI quickly reaches a plateau. This seems to be a likely scenario given that AI has historically hit several plateaus since the 1940's.

lesson

Learning about AI during a time where professionals are fearing for their future livelihoods and experts in tech are salivating over the potential superpowers they will unlock in the next 5 years creates a jarring experience. Being told that there is no meaning in pursuing a career in software engineering because of the near-term impact of AI while also being captivated by it's power today means that decisions need to be made.

One thing is abundantly clear to me today:

Deep technical expertise cannot be replaced and should be gained through relentless study. Sobriety safeguards perception and protects our essence of humanity.