"The best time to start implementing artificial intelligence in the future was yesterday."
~Kevin Benedict
Artificial intelligence (narrow AI) today is beyond its proof-of-concept phase - as it is already proven and delivering tactical value in many well documented areas:
- Reduction in human error
- Available 24x7x365
- Improved quality
- Improved productivity
- Improved efficiencies
- Able to dependably complete mundane, repetitive and routine jobs
- Makes faster decisions and taking quicker actions
Artificial intelligence, although still in its infancy, is already delivering impressive results and competitive advantages for those prepared. The preparation, however, is not insignificant and requires much work including:
- AI is designed for narrowly defined use cases. As a result, users must carefully evaluate and select the appropriate AI platforms to be able to support each particular use case.
- Skilled humans must be available to design, develop, test, install, integrate, test, supervise and make adjustments to AI.
- AI doesn't come out-of-the-box knowing how to perform a task. It's like a toddler with a lot of future potential. It must be trained. Training requires the availability of data. Data is often siloed or inconsistent and of poor quality. For prediction or decision models to be trained properly, they need good data - and lots of it.
- AI needs to be trained on what is a good, and a bad result so it can learn. Judging, filtering and testing criteria needs to be established. For example, to identify cats in photos, an AI system needs to be trained on what real cats in photos look like. The AI system can then be trained to identify photos with and without cats.
- Users will want to filter out and ruthlessly eliminate biased training data so AI systems are not trained to make biased decisions. For example, an AI system might identify the fact that only men have been to the moon, and therefore decide only men are capable of flying to the moon. That would be both untrue and biased.
- Be prepared to adjust your AI algorithms whenever your business processes are changed.
- Identify where AI can be inserted into business processes to add value, and understand the inputs and outputs required for integration.
Even with preparation many AI projects, despite the hype, are struggling. Shervin Khodabandeh, co-head of BCG’s AI business in North America, recently was quoted in Forbes as saying, "Companies are struggling to deliver on AI projects, because they overspend on technology and data scientists, without implementing changes in the business processes that could benefit from AI." The failure to implement changes in business processes was also a criticism often associated with early digital transformation endeavors as well. It seems we are failing to learn from the past, yet again.
Innovative technologies applied without proper thought to improving or replacing rusty old business processes is just silly. Business processes are often implemented in particular ways not because of best practices, but because of a platform or system's limitations. Once those limitations are removed or replaced - a whole new world may exist offering all kinds of new forms of value.
************************************************************************
Kevin Benedict
Partner | Futurist at TCS
View my profile on LinkedIn
Follow me on Twitter @krbenedict
Join the Linkedin Group Digital Intelligence
***Full Disclosure: These are my personal opinions. No company is silly enough to claim them. I work with and have worked with many of the companies mentioned in my articles.
No comments:
Post a Comment