G’DAY!
I’m Mathew Grace and today I’d like to talk to you about artificial intelligence and how to decide whether AI is right for you. Everyone is excited about AI, but in many scenarios it may be overkill. As a fairly new technology, it is more costly and technical to implement than other technologies, but as it becomes more widely accepted, prices should become more affordable like all advances in technology. Let’s get to it!
The criteria I use to establish whether artificial intelligence is the best solution are:
- You have a MASSIVE data set, but what are you trying to achieve with it?
- How many variables are being considered?
- Can you accomplish the same thing with simple rules based procedures?
After I discuss the criteria for making the decision to go with AI, I will discuss some examples of when AI is justified being used.
You have a MASSIVE data set, but what are you trying to achieve with it?
The point of any type of automation is to perform tasks that are too cumbersome for a human to perform. There are two main ways technology can perform these tasks, rule-based automation and artificial intelligence. Because Boolean logic has been around for decades, it is far simpler and less costly than AI. It will work in most simple tasks such as “If the customer is emailing about a job, send to HR”
So ask yourself, “what am I trying to achieve?”
If it is something you can map out the possible solutions, then you probably don’t need AI. We recently went through this process with one of our clients who manages billions in deposits and could not find one use case where it was really defensible to use AI over boolean logic or other automation processes.
How many variables are being considered?
What I’m talking about here is the number of data points you are measuring for each instance. Depending on what the variables are and how they relate to each other, artificial intelligence might be the right way to go.
To make this really simple, I’m going to use variables that are yes or no. They only have 2 possibilities.
| Variables(V) | Possibilities for Each Variable(P) | Total Possibilities (P^V) |
|---|---|---|
| 5 | 2 | 32 |
| 8 | 2 | 256 |
| 100 | 2 | 3.245×10^32 |
If you have 5 variables, or 32 possibilities, its probably a waste to use AI as that is pretty easily mapped out, but once you get to 8 variables, the possibilities have increased to 256 which increases the likelihood of human error. Once you get to hundreds of variables, good luck actually saying the number. (It’s 3.245×10^32 in case you are wondering.)
If many of the variables are dependent on other variables this may simplify it where it becomes more manageable. For instance, when doing marketing, age, sex, and income are common variables that create 50 possibilities, but you’ll normally be looking for patterns in a specific group so it simplifies the analysis. This doesn’t mean AI won’t work in marketing as Facebook uses it extensively in ads placement. The content users see is based on machine’s learning what each user like, comments, and clicks on to deliver more relevant content. It just means you should be cognizant of how you are using it. When attempting to deliver personalised marketing that speaks to a specific user AI may be useful.
Generally I’d be looking for 25+ variables with few dependencies. I’d also want variables that may be inversely correlated as changes in them will create much harder to calculate scenarios than if everything is directly correlated. In other scenarios, I’d recommend using less costly solutions.
Can you solve the same thing with simple rules based solutions?
Once you’ve answered the previous questions, this question is the one that ultimately decides whether you should go with AI.
Many companies can use rules to help customers search products. For instance, many sitesinclude filters that limit the options using rule based solutions. Let’s pretend you are looking to buy a house. You can filter by number of rooms, postal code, price, and amenities. The site uses these requirements to search the database and deliver the houses that match your requirements. No need for AI.
On the other hand, lets pretend you are the Reserve Bank of Australia and you were trying to calculate what the potential five-year impact on GDP from COVID 19 will be. In this scenario, you have so many different possibilities that you need AI. Each industry is a different variable and could have impacts on GDP that range from -50% per year(Casino revenue drop since COVID began) to +%50 percent per year(food delivery services have increased revenue more than this is 2020). In addition, government intervention can have dramatic impacts on the GDP because the more stimulus spending that occurs, the less disruption will occur over the short term, but it could come with higher interest spending for years or decades. In this scenario, there would be so many possibilities that anything besides AI would be cumbersome. With a financial model using AI, we could see the best, worst, and most likely scenarios based on a wide variety of conditions that could play out in each industry.
Other scenarios that might benefit from AI include:
- Detecting Fraudulent Financial Transactions: By monitoring for outliers in behavior, this could be really helpful in the financial markets.
- Facial Recognition: This is what Facebook uses for suggested tags in pictures. It can also be used by security companies an police departments to identify criminals. Another use would be to help consumers purchase eye glasses remotely.
- Establishing the intent in text: This is currently used by social media to flag inappropriate content, but could eventually be used for summarizing books and long reports.
- Marketing: Establishing target markets based on behavioral characteristics. this is already used in Google Ads to establish who has greater responses to your ads and recommends creating a target audience based on their characteristics. To give you an example, since developing better target audiences for our ads, we have see our total ad spending decrease and our click-through rate increase by 50-100%(as of 8/18/2020).
I hope this demystifies some of the decision making involving the use of artificial intelligence. While AI is a great technology, rule based logic is often a much more justified solution to common business problems. As AI follows the technology pricing curve, we may find that Ai can be used in more scenarios. Until then I’d recommend considering whether the problem you are trying to solve justifies the use of AI. If it doesn’t from both a use case standpoint and cost stand[point, I’d recommend going with rules-based logic. If you’d like help establishing what the best path to take is, feel free to contact us.
I’m Mathew Grace.
Don’t forget to follow our Blogs to learn more about our approach to software development.
Cheers!
