top of page

Artificial Intelligence: it's time to demystify the obvious

Updated: Jul 14, 2023

Artificial intelligence. Machine learning. Neural networks. Deep learning. Generative AI. Large language models. Human-level AI. Intelligence as a commodity (this one caused a few raised eyebrows at TerraManta).

The collection of terms and acronyms continues to increase, creating more confusion among consumers of this technology and especially investors.

It's time to demystify artificial intelligence.

Let's start with perfect software in a self-driving car

The software can do it all: stay in the lane, safely change lanes, anticipate traffic, plan new routes, brake when pedestrians are in front, learn how to adjust braking distance in different weather conditions, wake you up when you fall asleep at the wheel.

The software is perfect and every single label has been attached to the software: powered by AI, machine learning, deep learning, neural networks.

Since the software is perfect and powered by artificial intelligence, why not install the software in an airplane so the airplane can fly itself?

Except it's not possible and so far only humans can correctly determine that it is not possible.

Software which uses machine learning excels at learning within a specific domain of knowledge.

It is not capable of understanding that a different domain requires different knowledge to be acquired.

Self-driving car software has never (and will never) discover concepts which are essential for flying an aircraft: altitude, takeoff speed (rotation speed in aviation language), flaps, stabilizer, ailerons, landing gear, communication with air traffic controllers, angle of attack.

Even for concepts which are similar - for an engine stall in a car and an engine stall in an aircraft - the response is completely different. Self driving software in a vehicle can never learn how to manage the aircraft if one or - worse - two engines stall.

No matter how much better and more advanced the self driving software can be, it can never ask a question, "should I learn more about extending flaps when landing"? The knowledge domain of driving cannot generate data about flaps or landing. There is no data to be learned from.

Intelligence is the ability to identify a path from one domain of knowledge to another domain of knowledge. So far only humans are able to identify paths from one domain to another

Artificial intelligence: a poorly applied and overused term

Regardless of the term being applied, most solutions marketed today and using artificial intelligence somewhere in the description can be best described as ...

Accelerated knowledge acquisition >>> Autonomous dynamic observations >>> Autonomous dynamic decisions. Where dynamic means both data and rules are employed to perform an action such as an observation or a decision.

Time to demystify AI: it's not AI

Self driving software uses a the same basic approach which hasn't changed. The technology did change and that's why the approach can be implemented in a more effective manner today.

Revisiting the approach in the self driving vehicle software:

- Build knowledge domain

- Train algorithms

- Expand knowledge domain (do not brake for a small animal crossing the road; brake for a human crossing the road); does anyone remember Feedback Loop?

- Continuously re-train algorithms while using newly acquired data

- Rinse and repeat

This is machine learning at it's best. Rapidly acquire knowledge in a deep domain and implement behaviors that humans cannot safely manage concurrently, for example monitoring speed, brakes, road conditions, unexpected objects on the road.

This is not artificial intelligence and that's why this term is so poorly applied.

Time to apply intelligence

At some point, humans decided that based on market conditions, technology readiness, commercial potential, safety regulations a car can and should fly.

This process - determine market conditions, commercial potential - is real intelligence and can only be performed by a human. Artificial intelligence cannot perform the same tasks if this task is being performed for the first time.

Flying car is the convergence of two separate knowledge domains: driving and flying. Only humans can use what we commonly define as intelligence to make the decision and ...

identify a path from driving domain to flying domain: let's build a car which drives by itself and flies by itself if needed

TerraManta is a machine learning platform

That's what the reader will find on our web site.

TerraManta welcomes serious inquiries from investors who want to see machine learning solve real problems in today's world.

TerraManta is featured in this book.

99 views0 comments

Recent Posts

See All


bottom of page