How to use machine learning
You may not be using machine learning, often referred to as artificial intelligence, for business applications yet, but there is little doubt you have read or heard about how it could or should be used.
The issue is not that there are not legitimate business uses for machine learning (ML) options, the challenge is knowing which types of ML may work best for your business needs and finding the right provider or recruiting the right people to implement it.
Initially understanding machine learning is hard, but with a few big concepts under your belt, it becomes easy. It then gets complicated again, but by then you will be ready to deal with generative adversarial networks!
This is the most basic version about the content and should get you ready to listen to our special A Word On Artificial Intelligence podcast hosted by Primedia Broadcasting Head of Digital Allan Kent.
It will guide you through the technical concepts and how they are used in business. It has a case study looking at how Primedia Broadcasting used sentiment analysis to manage and respond to the volume of feedback the station receives to the 24 hours of programming.
Begin with the business question, not the A.I solution. Let’s assume you have a range of products and you need to decide to expand the range, focus on fewer or keep doing what you are now.
That is a tough but common question and depending on how much data you have, ML may have the answer. It will use a supervised form of ML using a statistical process called a regression. Effectively take historic data and try to forecast the future. Humans will have a certain skill just looking at the data, but machines can access much greater volumes of data and will be less susceptible to human bias. It might confirm your suspicion but it could point to something you might never have considered.
Knowing that there is a potentially powerful way to answer complex business questions has made it a compelling option for business leaders. PWC included an assessment of AI optimism in their 2019 CEO Survey. The five questions reflect a score from 100 for the most optimistic to zero for those that are completely neutral to negative 100 for the least optimistic.
Five questions with multiple choice responses cover what they determined would be the best indicators and in comparing responses across industries and countries. Taking the test for yourself is a good starting point.
The questions are
- Do you agree that AI will become as smart as humans?
- Do you agree AI will have a larger impact on the world than the internet revolution?
- Do you agree AI is good for society?
- Do you agree AI will eliminate human bias such as gender bias?
- Do you agree AI will displace more jobs than it creates in the long run?
The most optimistic A.I. nation perhaps not surprisingly is China with CEO’s having a score of 31. South Africa had a score of 16 while the Russians are the least optimistic at 2.
I am surprised that no nation was lower than two, and was surprised that by sector CEOs were even more optimistic with the lowest score being 7.
Machines, when asked to do tasks with clear rules, will beat humans with some ease. Chess games and any data processing can be processed far faster by machines and so if you equate that with being smart then machines will become as smart as us. But I would argue that while there is a skill in playing chess, a successful machine is not using a skill as much as a brute force in choosing the best option. Google’s AlphaZero took just hours to consider the rules of chess and the best way to play without any human intervention and was able to beat the best chess computer that had been programmed how to best play the game.
It is impressive but it did not show intelligence, it simply used the moves it considered that would most likely lead to isolating the King.
The same approach is being used to train robots to move. Rather than explain movement and physics, the machine is simply given the rules for how it may move and allowed to see what physics does when it tries moving. Because it can process information so quickly you can watch it as it first learns to crawl, walk and then run. When we see a spider running across the table we don’t call it intelligence for co-ordinating its eight legs at such speed, although we are impressed by it.
The takeout is that there is real value to be gained from the correct use of machine learning, but believing it can solve your problems when you ask the wrong question may just be more money wasted on bad technology.
Avoid that by making sure you listen to A Word on Artificial Intelligence.
Get the 10 most-read articles of the week from Bruce Whitfield’s The Money Show, emailed to you every Friday morning: