5 common mistakes made about AI

Artificial Intelligence

Artificial Intelligence is a major buzz ‘word’ at the moment.

It’s the new Big Data, but people throw it around without really knowing what it is.  AI is a complex field with a vast array of tools and approaches, and, when applied correctly, it can improve businesses in so many ways.

Everyone gets that AI can help predict a better future outcome.

So why is it that AI is so often misunderstood?  While the name suggests intelligence, perhaps the connection between a dumb computer and intelligence is the source of confusion.  Perhaps the Artificial part leads people to believe that anything complex done by a computer qualifies as AI.

In reality, AI is where computers take inputs from its environment and take actions to improve the chance of achieving their goal WITHOUT the assistance of a human.

Effectively, the machine can use what it’s been programmed with to arrive at a solution but takes in feedback loops on the fly to adjust how it calculates the results.  In effect, it learns how to do things better by doing.

That said, computers can do some really cool things even without AI.

Here 5 commonly mistaken AI tools…

Regression

Regression been around for decades and is heavily used in the finance industry for forecasting.  There are myriad methods of regression, each with strengths and flaws:

  • Linear regression
  • Logistic regression
  • Multinational logistic regression
  • Multivariate adaptive regression splines (Our favourite)

But it’s not AI

Simulations

Monte Carlo Simulations are a powerful way to understand the likelihood of certain events occurring, particularly when they are the result of a complex system involving many moving parts.

Monte Carlo Simulations involve creating many, many versions of a system or problem to come up with the highest probable outcome.  It does this using random sampling.  While it’s very clever, it doesn’t use any new information to inform it’s answer.

So it’s not AI

Optimisation

Optimisation is an amazing approach to solve some of the most challenging business problems.

Route Optimisation, packaging, etc. all benefit from using mathematical optimisation to create the best outcome, without relying on a human.

Have a look at our optimisation project GoLoop (www.goloop.com.au)

No alt text provided for this image

The problem set is all it needs to work with, and no new information is taken into the solution to influence a better answer.

So it’s still not AI

Time Series Models

Time series models aim to predict the future of data that moves over time, such as the stock market.  Some of the most common methods include:

  • ARMA (Autoregressive Moving Average)
  • ARIMA (Autoregressive Integrated Moving Average)
  • GARCH (Generalised Autoregressive conditional heteroskedasticity)

These are good, but none of them takes in feedback loops to automatically adjust their forecast, without human intervention.

You guessed it. No AI here.

Classification models

Classification models take inputs about the subject items and use them to infer output or target information about the subject.  More commonly known as Decision Trees, these methods use a series of rules to model the data to come to a maximal solution.  Other techniques include:

  • Random forest
  • Optimal discriminant analysis
  • Decision trees

While classification models are not directly AI, they are used extensively when implementing AI.

So what is AI?

In summary, AI is where a computer takes inputs from its environment and performs actions without human interaction that attempt to reach a certain objective or maintain a particular state.

In our day to day world, we can find AI in cars, smartphones, chatbots, social media, and so on.  Applying it to your business is not a great leap, but before you decide to embark on your AI journey, think deeply about your objective.

Artificial Intelligence can be amazing, but Human Intelligence shouldn’t be underestimated.

Share this post