Simply Practical Learning

want to know exactly what Artificial Intelligence is?

The big reveal, don't be too disappointed....

Artificial Intelligence

AI & Machine Learning are actually far simpler than you may think

Ai is the ability for a computer program to be created from data rather than data being created from coding an application.

Take the humble calculator

The program the calculator runs is a set of stored instructions determining an outcome based on inputs. 1 + 1 = 2.

The program understands that when an input number is followed by a plus symbol then another input number, the result is the addition of those two numbers. Simple.

Ai works differently

Instead of a human written program, Ai uses a process know as Machine Learning.

Machine Learning seeks to understand all the data and all the connections in that data to create a predictive model instead of a program. Tis is something that would take a Human a little bit longer than a machine.

How do we know Ai works?

One really clever part of the process is the Machine Learning system will only use a subset of the data for training and the rest for testing.

Machine Learning models typically use usually 75% of the data for training and hold back the rest of the data to test the validity and credibility of the model being used.

It checks the predicted outcomes of the new model created with the held testing data of real world outcomes to check the accuracy of the model and more importantly the predictions.

Even better, as new real world data is introduced, the model retunes itself giving more and more accurate results over time.

We have models, Now What?

How do we turn historical data into new predictions?

Once a model has been created and deployed in the real world, it works like a traditional program; when new data points are fed into the model an outcome is predicted.

OK but still, How does it work?

This example is called a Binary Classification

So in our calculation example instead of knowing the rule that 1 + 1 = 2, instead, we build a model from the data which might look something like this

Row 1 – Input 1, Input 2, Result, Outcome.

When we have 1000’s of examples of these rows showing what inputs give what result and whether this result is correct or not, then we can, with some accuracy understand that when input 1 is “1” and input 2 is “1” then the result is (nearly) always “2”

Row 0 - 1, 1, 2, true
Row 1 - 1, 1, 2, true
Row 3 - 1, 1, 2, true
Row 4 - 1, 1, 2, true
Row 5 - 1, 1, 2, true
Row 6 - 1, 1, 2, true
Row 7 - 1, 1, 7, false - Data is rarely 100% accurate and doesn't need to be with ML.
Row 8 - 1, 9, 2, false - Inputs as well as outputs may vary.
Row 9 - 1, 1, 2, true


So instead of being programmed to understand that 1 + 1 = 2 we can now infer from the real world outcomes, that we know to be correct, in most cases where there are two inputs, 1 and 1, the expected result is 2. Simple. This example is called a Binary Classification

Asking a Question of Data

Now take the example of understanding which Sales Leads will convert into actual sales so we can predict those opportunities that we need to focus on rather than trying to close everything with our limited resources.

All we need is a spreadsheet of historical sales data showing under what circumstances our customers actually buy – for example:

Getting Predictions

We have information on all our previous sales and we have outcomes – did they buy this month

From here we can create a model and predict if those companies who are not yet customers will buy from us and when our customers are predicted to buy and how much they will spend. So you now have all the information you need to concentrate your efforts and resources in order to maximise your growth!