A robot toy that lost its batteries - artificial intelligence business failure concept
PHOTO: Adobe

Last September, Gartner published its Hype Cycle for AI in which it identified two emerging trends (and five new AI solutions) that would have an impact on the workplace. One of those trends was what Gartner described as the democratization of AI. While there are many ways that this can be interpreted, in simple terms what it means for workers is the general distribution and use of AI across the digital workplace to achieve business goals.

Deploying AI

In the enterprise, the target deployment of AI is now likely to include customers, business partners, business executives, salespeople, assembly line workers, application developers and IT operations professionals. As AI reaches a larger set of employees and partners, it requires new enterprise roles to deliver it to a wider audience.

While this was an emerging trend last summer, with COVID-19 and the adoption of many new technologies to enable remote working, the widespread use of AI, while still only anecdotal, now appears to be an established fact in the workplace.

Bill Galusha of senior director of marketing at Calsbad, Calif.-based digital intelligence company ABBYY points out, however, that this is not a new phenomena. “In the past couple of years, we’ve seen AI enabling technology like OCR and machine learning become more accessible to non-technical employees and partners through no code/low code platforms,” he said.

He points out that the technologies designed to help workers understand and extract insights from content have been in high demand as more digital workers increase the number of tasks a knowledge workers have to perform.

In practical terms these new AI platforms enable users to design cognitive skills that are can be easily trained to take unstructured data from type of document like invoices, utility bills, IDs, and contracts, or access trained cognitive skills available through online digital marketplaces. “This new approach to making it easy to train machine learning content models and deliver them as skills in a marketplace are certainly going to fuel the online growth and reusability of AI as businesses look to automate all types of content-centric processes across the enterprise,” he said.

Related Article: The Risks and Rewards of the Citizen Developer Approach

Is AI Adding Business Value?

However, if AI is being used widely across the enterprise, it does not necessarily follow that it is providing business value to every organization, according to Chris Bergh, CEO of Cambridge, Mass.-based DataKitchen, a DataOps consultancy and platform provider.

AI is being deployed everywhere we look, but there is a problem that no one talks about. Machine learning tools are evolving to make it faster and less costly to develop AI systems. But deploying and maintaining these systems over time is getting exponentially more complex and expensive, he told us.

Data science teams are incurring enormous technical debt by deploying systems without the processes and tools to maintain, monitor and update them. Further, poor quality data sources create unplanned work and cause errors that invalidate results.

This is the heart of the problem and one that is likely to impact the bottom line of any business that uses AI. The AI code or model is a small fraction of what it takes to deploy and maintain a model successfully. This means that the delivery of a system that supports an AI model in an application context, is an order of magnitude more complex than the model itself. “You can't manage the lifecycle complexity of AI systems with an army of programmers. The world changes too fast. Data constantly flows and models drift into ineffectiveness. The solution requires workflow automation,” he said.

The Problem With Data

There is another problem for businesses too. Given the explosion in the amount of data that is available to them, at first glance you would think that developing AI was getting easier and, consequently, easier to deploy — democratized — across the enterprise. Not so, according to Chris Nicholson, CEO of San Francisco-based Pathmind, which develops a SaaS platform that enables businesses to apply reinforcement learning to real-world scenarios without data science expertise.

The real problem, he argues is that you cannot decouple algorithms from data, and the data is not being democratized, or made available, across the organization. In many cases, as with GDPR, the data is getting harder to access and because the data is not being democratized, most startups and companies will not be able to train AI models to perform well, because each team is limited to the data it can access.

In a few cases, a general-purpose machine-learning model, can be trained and made available behind an API. In this case, developers can build products on top of it, and that very particular type of AI is slowly percolating into products and impacting customers lives. “But, in most cases, businesses have custom needs that can only be met by training on custom data, and custom data is expensive to collect, store, label and stream,” he said. “At best, AI is a feature. In the best companies, data scientists embed with developers to understand the ecosystem of the data and the code, and then they embed their algorithms in that flow.”

Like the discussion around citizen data scientists (and democratizing data science), business leaders need to know what they want this new “democratized” AI to do. They will not be able to design and build AI models from scratch; that will always require an understanding of what the underlying methods and parameters do, which requires theoretical knowledge.

AI Has Limited Problem Solving

Given some “gray” box AI systems, one can envision such systems learning to solve well-defined classes of problems when they are trained or embedded by non-AI experts, Michael Berthold, Switzerland-based KNIME CEO and co-founder, said.  Examples he cites are object recognition in images, speech recognition, or probably also quality control via noise and image tracking. Note that already here choosing the right data is critical so the resulting AI is not already biased by data selection.

“I think this area will see growth, and if we consider this “democratization of AI,” then yes, it will grow,” he added. But we will also see many instances where the semi-automated system fails to do what it is supposed to do because the task did not quite fit what it was designed to do, or the user fed it misleading information data.”

It is possible to envision a shallower training enabling people to use and train such preconfigured AI systems without understanding all the algorithmic details. Kind of like following boarding instructions to fly on a plane vs. learning how to fly the plane itself.

Enterprise Actions

If organizations take this path to develop AI, there are two ways enterprises can push AI to a broader audience.  Simplify the tools and make them more intuitive, David Tareen, director of AI and analytics at Raleigh, N.C-based SAS told us. 

Simplified Tools - A tool like conversational AI helps because it makes interacting with AI so much simpler. You do not have to build complex models but you can gain insights from your data by talking with your analytics. 

Intuitive Tools - These tools should make AI easier to consume by everyone. This means taking your data and algorithms to the cloud to become cloud native. Becoming cloud native improves accessibility and reduces the cost of AI and analytics for all.

In organizations do this, they will see benefits everywhere. He cites the example of an insurance company that uses AI throughout the organization will reduce the cost of servicing claims, reduce the time to service claims, and improve customer satisfaction compared to the rest of the industry. He adds that some enterprise leaders are also surprised to learn that enabling AI across the enterprise itself involves more than the process itself. Often culture tweaks or an entire cultural change must accompany the process.

Conclusion

Leaders can practice transparency and good communication in their AI initiatives to address concerns, adjust the pace of change, and result in a successful completion of embedding AI and analytics for everyone, everywhere.