“Ai applications are Deceptive”
With everything in abundance these days - men, money or material – time is of the least supply. If your organization is looking to implement a viable AI solution to solve one or many of the problems – there is no one off the shelf solution. Ai and Data capabilities are to be built into existing organizational structure and cannot be imported as a viable solution.
Even when we have two similar organization looking to solve similar looking problems. An outdoor advertiser in Delhi looking to target clusters of high net worth individuals would be working with a different data set than the one looking to target clusters of high net worth individuals in Ludhiana or Dallas, Vienna and New York.
Same problem, same results, almost similar methodologies – but entirely different data sets. In this case there is no training involved but if there were two auto manufacturers doing defect detection – the training data set required would be totally different. Something as simple as placement of video cameras would make for total disparate data sets.
The Time Cost of AI Applications
The range of possible inputs is huge and handling this huge state space is an ongoing and laborious chore. One possible way is to fine tune model at beginning of each engagement, something we at Hashbrown Systems learnt the hard way. The intent is to remove or identify edge-cases.
All this work entails cost – manpower and financial – till you have a model with acceptable predictions. And there is an inherent risk to that. The results are often unacceptable – hence you devote more time and resources to build and deploy, much more than expected or hitherto anticipated.
Additionally, the traditional prototyping tools, do not work for Ai and Data applications. Prototyping and beta testing are limited to common cases.
Hence, unlike traditional software or software as a service, the financial cost of rigorous development does not disappear or even lower over time, and that’s that.