BuzzWord Bingo - AI/ML
Our social media feeds, TV ads and Billboards are full of them - BuzzWords. Although many of these have long, carefully researched and well cited articles, we will stick to the quick and brief description that is applicable and understandable by everyday people in the context of today's business scene.
AI (Artificial Intelligence)
Just as a quick heads up, we have been using ML/AI (and a number of other analytics methods) for a number of years in our security & other analytics, and if you are interested in exploring what more you can do with AI or your data, just let us know, we love this stuff. that said, here we go . . . .
AI is currently a much used marketing term to drive interest and sales in products or services. There is rarely a tech-based product that isn't spruiking AI in its make up.
In reality, AI, as a concept has been around since the old philosopher days and actually became more formalised (as a learning stream) back in the 1950's.
But I won't get into the history, or the debate about what Artificial Intelligence is or isn't, although I do have opinions on this :) wikipedia is a good source/start for going down that endless tunnel.
Instead, I will keep this short and practical - what this means for you and your business today.
AI uses a bunch of computing power (often thanks to newfound awesome cloud computing ability) to take an input, run a bunch probability maths and similar over it, and ultimately come back with something that is the most probable response.
There are different types for different purposes - one style is like giving it a bunch of animals and telling it which are cats and which are spiders, then it 'remembers' the pattern on each (cat has four distinct legs, furry, two eyes and tail, spider 8 legs, many tiny eyes etc). Then when you give it your pictures it compares to what are cats and spiders. This is called supervised learning (it is told right from wrong, or categories)
Simple Model - Input Layer=Pictures, hidden layers look at features (how many legs, whiskers, eyes, colour, shape, etc) and then output cat or spider
Unsupervised learning is when it is up to the maths to find its own patterns - often good for 'computery' type data. A user normally logs in 9-9:30, opens email, opens social media, opens browser, reads the news . . . . and they do this on mon-fri. Saturday and Sunday nothing.
So when security people (like we do) provide security monitoring, if it detects a change in your pattern it lets us know.
Make sense? It is the same principles, when it comes to making images, and turning Word docs into PowerPoints.
But here is the kicker . . . where does it get the sample data to recognise the patterns in the first place? Originally, a lot of information was taken from the internet- like a big vacuum cleaner, it was all swallowed up and analysed (often without the owner of the data knowing, and they have since locked a lot of this down to stop them taking their data). But that was originally . . . where do they get the data now?
Often the data comes from you, especially if you are submitting documents for analysis and output.
And here are two important things to understand out of this:
1. When you submit requests for AI services (like ChatGPT and similar), most of it runs on others people's computers, and they keep it to train it to be better, using your data. Always read the fine print. ChatGPT does have a warning, most others don't. And some companies are incredibly frivolous with AI and your data (and the marketing).
2. Also, What AI often gives you back is the most probable response based on the information THAT IT WAS FED IN THE TRAINING. Not facts, but what it thinks is the most likely or most popular. You must review its outputs.
So the key advice from Small Robot is this:
Treat AI tools as you would a stranger on the street. You wouldn't give them your bank account PIN/Password, or even your identity data, and anything they said, you wouldn't take as fact, you would check and verify it. Same with AI tools.
Should you use it in your business - absolutely, if it makes sense and you have use case for it. It can be a great time saver, and provide anlysis and out insights you might not otherwise have known. Many things in life have already had it built in for years - and we ourselves have used it over the 8 or 9 years.
Should you be extra mindful and follow the key advice above - absolutely. Protecting your data and reputation are paramount for a business, and the loss of either can undermine trust and lose customers.
Just a quick fire few terms:
ML/Machine Learning - basically, the building blocks of AI. Using machines/mathematical compute to do the probability analysis at depth/scale/timeframe. AI is ML on top of ML on top of ML (in basic terms). ML analyses a Chapter of a book for patterns, then paragraphs, then sentences, then individual characters and then across all of the learned patterns together before having a model that can provide answers
LLM's - Large Language Models, specifically trained to understand languages - like the example in ML above. Picking the patterns like, knowing that "It" is normally followed but "was" or "is" or "has" but never followed by "weren't". This is what most people relate to - it is what ChatGPT uses.
Generative AI - similar to the above. Give it some info to describe something and it provides its best probable reply to what is most commonly accepted for that. Often used for generating text - like, "write me an article on ransomware" . . and it looks for examples and generates the article (which you review and edit yourself to make sure it is right).