Flipping the AI Strategy Paradigm

We are more excited about AI today than ten years ago. Generative AI's ability to learn and create new content has blown us away. Show AI 100 photos of cats, and it can create a unique, imaginary cat picture close to the real thing. How can we encapsulate this ability to 'learn and create' in an organization's AI strategy?

Art of the possible

Last week I?explored?the need for a comprehensive data strategy to construct an effective AI design methodology. This week I want to dig that concept deeper while hypothesizing future capabilities. Most AI strategies covered today start with the business domain - identifying priority use cases, ensuring value generation, locking in organization-wide commitment, and identifying matching technology, skills, and data, finally leading to the outcome - an AI application. I call it an "outside-in" approach. This strategy construct has AI as the outcome.

Could there be another way of approaching it that may augment and possibly replace the current approach?

One of the fundamental differences between AI and previous generations of engineering disciplines is its ability to learn and create. Compared to, say, the software engineering discipline, which can solve many complex problems. However, each requires detailed human instructions to get to the goal. Our current approach to AI is still instruction-heavy.

To position AI as a learner within the organization, we need to make it the central player looking into our business through a data lens. I call it the "inside-out" approach. Imagine a world where AI could look at the complete data flow within our organization to identify opportunities, much like AI monitoring your shopping habits and recommending the product you will likely buy. Could it monitor enterprise-wide data flow and point out gaps and redundancies?

Imagine the possibilities.

• By correlating finance, stocking, transportation, and fuel price data - it could propose a revamped distribution plan that reduces cost.
• It could design more user-friendly product documentation by looking at customer complaints and where the support engineers are looking to find answers.
• Analyzing data access hotspots after a significant report release could pre-process future observations to reduce latency and further optimize IT resources.

I can go on and on, but you get the drift. We are talking about a way for AI to look at organization data (as well as metadata) collectively to make more valuable observations, create new products or monetize existing ones, or even offer new ways of doing business.

Are we there yet?

No, we are not. But it is a growing area of interest. We would need a robust?recommendation engine?like deep recommender systems that have the potential to provide the automation and tech necessary to perform?automated multi-modal recommendations?looking at multiple facets of data (as well as metadata).

AutoML?itself needs to gain better capabilities and improve search/optimization performance to tackle the volume of information and magnitude of coverage we will likely have to deal with. It will require further improvement in computing and storage capabilities - which we know is happening.

Then there is the question of combining our enterprise data as a?singular inspectable model. Approaching this could be a big challenge considering governance. We have learned a lot of lessons in the last couple of decades building data warehouses and lake houses, and it is still an evolving space. In the process, we have already embedded micro-AI capabilities in automating individual tasks. For example, BI tools offer alternate visualizations you may be interested in. Query tools are recommending better ways of capturing insights. Code generators provide clever ways to create new products. These individual recommendations could be interconnected to perform broader inspections. Concepts like Data Fabric and Data Mesh are attempting to address some of these goals.

What are the pitfalls?

Much of the concerns attributed to AI generally would apply. Transparency, drifts, biases, ethical use, etc., all will come to play. But keep in mind, the objective here would be to use it as a recommendation tool with a human operator at the end deciding whether or not to put it to play. We are simply trying to reverse how we approach AI.

We must relook at governance (data and AI) to prepare for this approach. How would we balance the need for data visibility and sensitive data? Here I am not just talking about personally sensitive data but data that corporations may deem commercially sensitive. Could AI transparency create further exposure?

Why consider it now?

Is it too early to start thinking of this possibility? I don't think so. The rapid advancement in AI is shaking things up, and the need to be ready when the solutions drop is paramount. An approach like this would heavily rely on establishing an AI-ready data strategy - something we can start working on today. The building blocks are already there; we can even contemplate a semi-automated process using today's tech. We can start by asking, "Do our data scientists have access to any combination of data features in the enterprise?" We can improve our data architecture and governance processes with the gaps we identify.

Conclusion

The real power of AI is its ability to learn. While we are excited to see AI learning how to write a limerick or paint like Van Gough, it would be far more exciting when AI can understand how we do our business and recommend ways to improve it. So instead of finding problems for AI to solve, let AI see problems it has solutions to.

#future #technology #innovation #transformation #strategy #artificialintelligence

Love to know your thoughts on this. Do you see this as a possibility? Do you want to see this as a possibility? What challenges can stand in our way? Comment below.