For years people have been testing their knowledge and skills against computers and on most of these occasions’ computers have come out victorious. A lot of the tests and challenges revolve around vision and games such as chess. To run software for such tasks requires powerful infrastructure like GPU’s (Graphic Processing Units).
GPU’s are not new but the use of the technology is fairly new in the commercial environment. On paper, a GPU cloud based solution could seem expensive when compared to a CPU solution but a CPU alone can not affectively run image based software like AI computer vision and does not have the same power as a GPU. AI and Computer vision is impacting and revolutionising many companies and industries due to the scalability of the software and processing power. Does all this power come at a cost? How does it compare to the cost of a human employee and team?
Comparing software to humans is tricky stuff. The test has to be specific to the purpose of the software. AlphaZero (AI software) is amazing at the board game GO which was thought to be the only game where computers can not beat a human, this theory has now been proven wrong and is now accepted that AI is better at GO than a human. But if you put AlphaZero in an environment where its not familiar or trained to do the software would be useless. If AlphaZero was asked to identify pictures of animals it wouldn’t work, where as a human can play GO and identify pictures of animals, this is because humans have better general intelligence than a computer and a better processing unit (the brain) than a computer.
Intelligence is all relative to training, experience and the environment. Let’s look at a simple use case where a company needs to sort and analyse tens of thousands of images daily. Traditionally, this would be a mammoth task and most companies would not bother to tackle this problem due to the human power needed to sort and analyse this amount of images on a daily basis.
We have a company who wants to take this on, the company is called Company A.
Company A generates 20,000+ images on a daily basis, they’re operational 24/7 all year round. Currently, Company A does not analyse the images but wants and needs to, to grow and scale the company. Company A knows when they scale their business they will generate more image data so it’s a top priority to know how to tackle this problem before it spirals further out of control and becomes a blocker.
Company A could hire a dedicated human processing team to tackle this problem, let’s look at this.
Volume images per day ~20,000, number of humans needed to look at all images could be around 50 on average that’s 400 images per person per day. Company A is a 24/7 business so they need 3 shifts if the shifts are 8 hours. Each shift will need a supervisor or manager. An employee only works 5 days and requires holidays. Each employee will need to analyse each image carefully and complete a report. This is time consuming work, repetitive and not rewarding work.
When a new employee starts Company A must account for induction timing and training and the new employee will need supervising and wont be expected to process the same volume as an experienced employee. Training a large team is costly and time consuming. Once trained, it’s difficult to update the process and even harder to completely change a system or process due to the training and time needed.
The process of hiring a good member of staff on average takes 4 to 6 weeks this doesn’t include notice periods and gardening leave which drastically change from days to months and from industry to industry and location to location. Retention rates for Company A are low as image processing is not a fulfilling role and humans are more ambitions than processing images and completing a report.
The number of data point analysed at a single time is limited in a human workforce, humans are not good at multi-tasking. Company A will also have to account for biases and inconstancies in their data, human driven processes will be riddled in them.
Company A is wondering what to do, there seems like a lot of challenges in taking on this project but company A really need to do this. What other option does Company A have? A GPU based software solution, let’s explore this solution.
A cloud based GPU system will cost around 12,000 USD a year with some additional storage costs of 1000 to 2000 USD a year maybe less. This cloud based system seems very appealing for Company A and cost effective, the system will be more than capable of processing that volume of data on daily basis, scale is not a problem.
Computers don’t require holidays, sick days and can work 24/7 all year round. Computers have their moments but the cloud company maintains the servers so this is not a problem.
Scaling business is quite simple, Company A has a few options with computers to handle scale, Company A can increase the GPU power and optimise the software to handle more data and to increase productivity. To do this with a Human team it can take weeks and what do to do if Company A suddenly needs to process 30% less data? With a computer it’s easy as flicking a switch but this is not that simple with a Human team.
Updating Company A process or changing the process with computers is easy as the software is tested before pushing to live deployment and once its tested and it works Company A just needs to replace the old software with the new which takes less than a couple hours without businesses disruption.
There are no issues with retention rates as a computer isn’t going to go on strike and doesn’t dream of bigger things than processing images and filling our reports.
With all this considered Company A knows what to do and has made the right decision in the end.
Traditional computer software like websites and mobile applications do not require powerful cloud technology to run the software, this is because these software’s lack intelligence, don’t require processing power and don’t have the same return on investment. A PHP website for example can’t make business decisions and run a whole department.
Buying and investing in new software is as important as hiring the best team and the right people.