Picture of the logo

No BS Overview of AI

My own personal approach to what AI means and what applications it realistically can be best applied for.

I've had a couple of friends and family reach out to get some answers on what this AI stuff is all about, and I figured that I'd write it all down and share it all at once. I think that AI is a broad and super fascinating subject, and I fancy myself as something of an AI expert myself.

What is 'AI'?

The term 'AI' is short for Artificial Intelligence and has become a blanket term for a wide range of technologies. I'm not going to spend a whole lot of time gatekeeping what defines a technology as 'AI', if it should be anything from 'if' statements to humanoid robots doing flips. The discussion recently of AI tends to focus on the following areas:

  • Machine Learning
  • Generative AI
  • Big Data I am making some very large generalizations here, and I encourage you to go and look up these and other tangential topics for yourself, I'm just focusing on the topics that seem to be most talked about on Linkedin when the generic term 'AI' is used.

Machine Learning

Of the subject areas mentioned, Machine learning is the oldest of them. Machine learning, in it's most oversimplified definition, is a function that is given an input, analyzes it and then produces an output that is not defined by any explicit instruction.

An example of this in action would be to imagine a neural network that is being trained to play rock, paper, scissors. Traditional code might have the function produce one of the 3 by selecting it at random, or selecting whatever the opponent used last turn. A neural network (ML) would 'learn' by playing several rounds, and after a given amount of rounds of rock, paper, scissors, it would give outputs based on what is most likely to win.

In the real world, Machine learning is used in applications from dynamic product pricing, facial recognition, self-driving vehicles, handwriting readers, personalized ad placement and social media algorithms.

Big Data

Big data is a broad term that applies to taking large, often unorganized and unstructured data, and processing useable information out of it. An example of big data might be taking thousands of Tweets about a given topic and producing a sentiment analysis from that topic. Big data tends to find odd relationships when the dataset grows so diversely. A company might determine that ads for their product become more effective at a time of day that seems random. They could have determined that all of their users have a similar daily routine and because of that, their users become to receptive to the product at a similar time.

Big Data is similar to strip mining minerals. Giant machines crunch and scrape millions of tons of rock and dirt, then ship it to a factory where it can be separated and refined to a useable resource.

Generative AI

Gen AI has been the bell of the ball recently. This term covers any AI model that takes in an input and produces a output based off of it's training data. Gen AI models are growing very diverse, with the most popular applications being image, text, video and audio generation.

Open AI's ChatGPT is the best example of this subject area. It's LLM (Large Language Model) has nearly become synonymous with the blanket term 'AI' and became one of the most successful consumer applications in history. ChatGPT is a multi-model AI. This means that it can take in more than one input (audio, text, image) and produce more than one output (image, text). Multi-model applications are rising in popularity and offer a very generalized way to interact with AI. Another practical application of Gen AI is Github CoPilot. CoPilot offers complex and dynamic code completion for developers, saving us a trip to stack overflow in most occasions.

The two examples above (as well as the majority of Gen AI services out there) utilize LLMs. Large Language Models can be dismissively described as a glorified auto-complete. Given a string of words, an LLM determines what the next most likely word or phrase should come next. If I say the phrase:

"Twinkle, Twinkle Little Star, how I wonder..."

Can you guess what comes next? Given that you've probably heard or seen this nursery rhyme, you would assume the rest of the text should be "what you are." That's what LLMs do.

Applications of AI

Browsing Linkedin these days will have no shortage of ways that AI is used in the business world. I've seen everything from using ChatGPT to run an entire business, Personal Assistants, AI clones, Customer Service bots, business analytics, financial advisors, and even fully fledged application development. This is my opinion, but most products and services that utilize Generative AI are complete horseshit. We are in the wild west of technologies like Gen AI, and a lot of products and services using this will become obsolete or bankrupt in a few years.

Big Data and Machine learning have thrived in the business and tech world and are used just about everywhere now. Machine learnings more 'flashy' applications include computer vision, like Tesla's self driving functionality and facial recognition as well as advances in robotics, like Boston Dynamic's Spot rover.

When should you use AI?

Abraham Maslow's infamous quote

If the only tool you have is a hammer, you tend to see every problem as a nail

Applies so well to the broad use of AI. Gen AI, big data and machine learning are all incredible powerful tools, but are not always the best solution for every single problem.

Take a customer service bot for instance. Sometimes, all a user might need is a generic traditional script based flowchart to get to the information that they need. Companies should consider the amount of time, energy, training, cost and possibility for error when scoping out a project like this. A simple if statement tree is going to be magnitudes less expensive than a custom pretrained LLM customer service model.

To determine if AI is a good fit for a given problem, consider all of the options, consider what a given technology can bring, and be aware of the costs. AI runs about as efficiently on computer hardware as an engine full of cheese. Each image that you generate for your new company logo costs nearly 20 cents in computer runtime.

Best uses for Gen AI

I have found that the generic, all-purpose apps like ChatGPT are great to throw ideas off to, or proofread a draft. However, in a business application, specialization is key. If you are looking to create an app that writes tweets in perfect Iambic Pentameter , It would be more scalable and clean to train a specialized Large Language model on basic English and Shakespeare's writings, instead of using the massive, expensive, generalized models like Chat GPT.

Best uses for Big Data/ Machine Learning

Applications of big data and machine learning often go hand-in-hand. large quantities of data are often simplified to a format that a machine learning model can interpret. From here, any sort of business application can be augmented. Machine learning thrives in situations that structured input is not always available, or multiple decisions with enormous amounts of variables need to be made in short periods of time.

In conclusion, Use the right tool for the job (it isn't always an AI solution), Use specialization where you can, and be intentional with your data and outcomes.

If you like our content, please consider subscribing to our weekly newsletter. I'm biased, but it's pretty great. Sign Up

Beecon LLC, All rights reserved.