Nearly every day there is a new headline highlighting the latest advances in artificial intelligence.
Almost every story doesn’t focus on AI.
Although we have made great advances in automation and natural-language processing (which are usually the topics of articles), these two things are not AI.
The two are important steps in the journey to AI. However, a speaker at home that turns on the lights when you ask it isn’t really intelligent.
It’s not because we haven’t tried. Australia’s AI expenditure is expected to hit $3.6 Billion and 44% M$ Government grants will be available for the development of AI and digital capabilities centres. We are determined to get there, but haven’t made as much progress as some would think over the last decade.
The challenge is in how we define AI. It should, in its simplest form be a computer program that can draw from its past experience, think for themselves, and create new solutions, processes, or creations on its behalf.
Can anything less be considered artificial intelligence?
The ‘AI of today’ is a set of algorithms that have a defined and finite purpose. These are known as ‘weak AI’ or ‘narrow AI’. These algorithms are limited by the way they were programmed. They essentially boil down into a series ‘if x then y commands.
Machine learning and automation make our lives easier today, but they are reactive – the machine needs to be told exactly what to do before it will act.
As an example, consider the parking lots at shopping centres.
What’s currently described as “AI” are programs that track how many parking spots are occupied, and how many are available. It may do this by using sensors to monitor each parking space, or it could be done through smart CCTV that recognizes objects. The end result could be that the spots with occupied lights are red, while the ones still free will have a green light. At the entrance of the parking lot, an LED screen would display the number of remaining spots.
This is not AI, but it is an impressive feat in automation.
The program still uses an algorithm which tells it to show a red signal if a vehicle is in the spot. If there is no car, the green signal is displayed. It then counts the number of free spaces and displays that (if x then y).
The car park of a shopping centre would be able to make much more intelligent decisions if it had true AI. It wouldn’t need to be told what they should look for. What would this look like exactly?
It could estimate the average buying power of shoppers by analysing each car that enters the parking lot. It could also discover trends about when different cohorts visit the shopping centre and give advice to each store regarding the optimal staffing levels.
This can be done by tracking the brands of the bags that customers are taking out. Then, using the experience gained over time, you can determine when certain stores will likely be busier.
It could alert stores about a shoplifter who entered the parking lot. It could detect if a stolen vehicle or one with stolen plates enters the facility.
This insight can be shared with the management of the centre to ensure that the fluffy big mascot appears or that children’s entertainers get priority at this time.
The point of a true AI is that you don’t know what it will discover, or how it will connect seemingly disparate data.
Computing and storage are a major bottleneck between where we currently are and where we want to be.
To be able for an AI to create its own thoughts, it must have a massive amount of data that it can analyze instantly.
We’ll keep making better mousetraps until we can overcome this bottleneck. But they’ll tell us nothing new about mice.
