What do you mean when you say AI? What kind of words does the phrase conjure up in your mind?
For me, AI is math combined with hardware fast enough to process a large set of data. Stanford[1] credits John McCArthy with defining AI as:
“The science and engineering of making intelligent machines”
I am writing this article to future and current clients, or those simply curious to understand the concept beyond the hype. I note that so called AI implementations have started to show up much more often in consutlant broker ads, posts and media coverage sense 2020 and this worries me as a consultant. It worries me because the phrase AI is fuzzy, and can mean so very many different things – leading to misunderstandings at best, ill-fated projects and wasted money at worst.
But lets get back to the topic of “AI, Artificial Intelligence”.
The tl:dr (too long, didn’t read) version of what AI is can be described as:
- Linear algebra
- Vectors and matrices
- Integrals
- Probability statistics
Sprinkle a bit of coding (mainly script languages such as Python) and you have AI. No seriously, that’s it. I wrote my thesis using “AI”, though we didn’t call it that back in 2012. The thesis used a learning algorithm called Support Vector Machines[2] to classify managed futures funds in the due diligence process. How did basic math such as SVM:s which found its way into my Master’s thesis become “the shit” in all IT/tech transformational projects?
This brings us to the longer description of what “AI” is, and why you should stop talking about “AI”. AI is usually divided into a few sub-fields (probably not an exhaustive list):
- Machine learning
a. Supervised learning (incl. Support Vector Machines)
b. Unsupervised learning
- Neural Networks (though it sounds fancy, a simple implementation of it uses regression analysis which is a basic and easy to understand mathematical concept)
- Deep learning (e.g. Generative pre-trained transformer, more commonly know as GPT)
- Natural language processing
Without having spent too much time researching the etymological background of AI, I am guessing it probably means algorithms that “feel” a bit smarter than the usual ones, algorithms that take inputs from a human and gives an output that is much more refined.
So, if you find yourself tasked with or wanting to implement “AI”: ask yourself, what is you want to do. Automate processes? That probably falls under the realm of robotic process automation (RPA). Do you want to classify emails (e.g. Customer Service process efficiency)? Then you might be looking into classification functions falling under the realm of natural language processing, or machine learning. Do you want to work with risk management in crediting? Machine learning and our good old friend SVMs might help.
If you find yourself wanting to work with “AI”, skip the fancy-schmancy expensive courses and get back to basics; enroll on a program or set of courses on the mathematical topics of linear algebra, statistics, vectors and matrices, Python or similar script orineted coding languages that have libraries and frameworks (e.g. sklearn, TensorFlow) pertaining to the mentioned math for ease of use – no seriously, you can’t imagine how much time it took me to code this from scratch back in 2012 when I didn’t fully comprehend the concept of programming libraries and working mainly with MATLAB.
If you find yourself wanting to implement “AI”, ask yourself what you want to achieve and gather people with proper coding and/or mathematical knowledge to you help you in the buying process if you are not creating your solution in-house; do avoid “black box solutions” where you cannot explain what you are paying for and what the solution is supposed to do / what kind problem it is supposed to solve.
What is new is not AI, its how much faster computers are today compared to just a few years ago, allowing us to do much more powerful computations. Computations we could only dream of, now available as apps that anyone can use without any mathematical or programmatical prerequisites.
The risk of continuously using the phrase AI and oversimplifying things is that we create two groups of people; those “in the know”, and those left to guess and mystify AI. I am much more into inclusion, and I rather take the time to explain something difficult to you than bull*** you (though I most probably would make more money the latter).
[1]“Artificial Intelligence Definitions“, Stanford University Human-Centered Artificial Intelligence, https://hai.stanford.edu/sites/default/files/2020-09/AI-Definitions-HAI.pdf
[2] “Managed Futures Strategies- Support Vector Machines as a Due Diligence Tool”, Roula, FEB 2013, Stockholm School of Econoics (Handelshögskolan i Stockholm)
No responses yet