Are We in an AI Revolution? Part Three of Three

Welcome to part three of “Are we in an AI revolution?”. If you haven’t already, be sure to read Part One here and Part Two here.

We left off with a deep dive into the categories of AI capabilities.  Let’s now bring this all back around with some real-world guidance in terms of AI and search capabilities.

Real World Guidance for Search

Obviously, there is a lot to take in around AI capabilities and the varied offerings available within the market. To break this down, I have three core takeaways that should help cut through the noise and refine an AI strategy for your search initiatives.

1. Put Your “Innovate” Hat On

This is not as simple as “turn on AI” and everything will get better.  Yes, AI capabilities can have dramatic effects on quality and user experience, but they require innovative thinking coupled with good planning and execution.

Start from the user perspective, with a little brainstorming on which issues are the most important to your users.  Perhaps you have a lot of knowledge locked away in videos and exposing the content within those videos via search is a great way to start.  Perhaps your users aren’t good at querying the search index, so a project allowing them to use natural language when working with search is a great place to start. Once you understand their issues, look over the AI capabilities and toolkits and come up with an innovative approach to solve their problems.

Once you’ve designed an approach, treat it as a core feature of a deployment with accompanying use cases and user tests that both define the impact and make testing the impact possible.  If you do this, then you’ll be able to easily quantify the impact and investment. Once you find manageable project chunks, focused on specific capabilities, you’ll find that the concepts of AI are easier to manage.

2. An AI Strategy Needs to be Best of Breed

AI capabilities should not be looked at like traditional IT systems.  While it makes sense to standardize on a hardware platform (i.e. use all Dell laptops or HP servers), the same is not true for AI vendors.  You do not need, and should not want, to decide on a single AI vendor. No one should be a “Google” or a “Microsoft” shop when it comes to AI. You need to leverage the vendor that is the best at the capability you need. You need the ability to integrate with multiple providers and then switch providers should capabilities change, or when new features become available.  A best of breed strategy is the way forward. This is the same strategy we have been following in the search market, and the same applies for AI.  The advice we give on search engines is to pick the best one for you today (Elasticsearch, AzureSearch, SharePoint, etc.) and then don’t be afraid to switch if one becomes more compelling than the other.  The same is said for AI capabilities.

I would not be surprised to see a best in class AI-driven search application to leverage Google Cloud Video Intelligence for video analysis, Microsoft Cognitive Services Vision for Image analysis, and Amazon Comprehend for NLP capabilities.  This model takes the best of each vendor and applies them together, forming a complete strategy. I also wouldn’t be surprised to learn that six months later all three of the components have been swapped out for the current market leader. Doing that is the true power of best of breed.

3. Don’t Get Confused with Hype

When evaluating AI capabilities, be inquisitive and thorough in digging into how the system works, how it was trained, what additional training is needed, and what the size of the data set is that is needed to be effective.

As an example, if accurate image recognition requires that you feed millions of images to the solution to properly train it, then will you be able to provide that level of example imagery?  Training and access to data sets of size are one of the key pieces to understand in terms of AI.  In many cases, the training and data sets are baked into the offering (an example of this are the Image and Video recognition services offered by Google, Microsoft, and Amazon).  In these cases, you pay for the API usage, which allows those providers to monetize the training and data sets they have done. In open source solutions, the training often falls on the provider, exposing a hidden cost that may not have been originally understood.

In the end, AI is not magic and be weary of high level “because AI-type” answers when you ask for details.  If it can’t be explained, then it’s hype. If it’s hype, then the result will be overpromising and underdelivering. This hype cycle gives these capabilities a bad name and muddies the waters around the actual results that companies can see.  For every success story, there’s a story of a vendor that was unable to deliver. The foundation of any search solution still requires connectivity to systems and metadata across all content.  AI capabilities sit on top of those foundational blocks. Make sure you build on a solid foundation.

Now What?

This concludes the three part series that I hope has provided valuable information surrounding all things AI.  Please comment with your thoughts and opinions as well.  Sharing experiences and beliefs around AI will only help to further the understanding and growth of this technology. My final recommendation to you is to continue to research and evaluate how AI factors into your overall search strategy, as things are changing fast.  Revolutions wait for no one.