Tulip Mania of AI

I am learning AI/ML for some time now. We're all bombarded with headlines about AI capturing market share, boosting margins, and creating revolutionary products. It's like the Wild West of AI out there! Companies are throwing AI buzzwords on everything, hoping it sticks like spaghetti to a wall.

So, I've been spending way too much time lately pondering this burning question: how can companies shove AI into everything without it exploding? :D

One topic that really caught my attention while learning about AI/ML was P vs. NP problems. It turns out, this seemingly theoretical concept has a big impact on how AI can be applied in businesses. One can find all sorts of definitions for P|NP problems. My understanding is this.

Imagine you are trying to find if there is a needle in a haystack.

Finding the needle (after someone points to a specific location in the haystack): This is like a P problem. You have a specific method (someone pointing it out) that lets you quickly verify the answer (the needle).

Searching the haystack blindly: This is like an NP problem. Checking every piece of hay is a way to find the needle eventually, but it could take a very long time (depending on the haystack's size).

The key difference is that verifying the answer (is this the needle?) is easy in both cases. But for NP problems, finding the answer (in a constrained timeframe) itself can be much harder.

AI implementation in companies is like sorting laundry: P problems are the bright, easy-to-match socks, while NP problems are the mismatched pairs that leave you questioning the universe's sense of order.

P problems are the low-hanging fruit of AI. Successful examples include use of LLMs to translate Product documentation in multiple languages. Use of ML to build a catalog of parts using classification algorithms. Chatbots, personalized content for marketing and sales, anomaly detection.

Any issue that involves a routine process, a set template, or minimal deviation from the industry standard will be addressed initially. These problems have a well-defined resolution process.

Consider, for instance, the task of translation. Translating product documentation is straightforward and direct, devoid of emotional undertones or exaggeration. The message needs to be concise and unambiguous. However, it’s not always that simple. Contrast this with translating a work like ‘The Fountainhead’ by Ayn Rand. The translation must pay attention to the context, cultural subtleties, the author’s writing style, symbolic elements, and above all, the author’s fervor in conveying the message.

NP Problems are the business equivalent of a crystal ball. They deal with simulating different features, strategies, and their impact on profits and market share. It's about finding the golden ticket that unlocks explosive growth.

Leaders can read books, hire consultants, seek advice and define a strategy. How to they predict the success of that strategy? The future is laden with Stochastic Process, which basically means our lives are like a wacky episode of Black Mirror directed by a dice-rolling hamster.

The curse of dimensionality reduces your well-defined dataset to the size of a… well, a data-crumb. And then their is the money problem. Huge investments are needed to build, run, maintain and actively retrain LLM or AI/ML models.

NP problems feel like a giant, labyrinthine maze filled with potential solutions for businesses. There are dead ends (inefficient strategies), misleading paths (faulty data), and hidden gems (breakthrough innovations) waiting to be discovered.

The journey might be tough, but the rewards are epic! I don’t have all the answers, but each stride I take brings me nearer to discovering inventive solutions and revealing new prospects.