Enterprise Data and AI: A Path to Democratization

"How can we establish a democratization plan for data and artificial intelligence?" - it's a common question among non-technical C-level executives nowadays.?

Data and AI democratization are clearly top of mind among knowledge worker businesses and sectors. As organizations look for ways to successfully democratize data and AI, consider the following five tips for enterprise and enterprise computing?

1. Build an internal marketplace for algorithms and data.?

Clean data, a strong data structure, and well-labeled data come first. Who will be the data custodian is a question that organizations may pose.?

I frequently suggest that businesses consider setting up a private marketplace to internally crowdsource the creation of labeled data products, data structures, and pre-trained algorithms in order to make this happen.?

The "ImageNet moment" has not yet been produced by big enterprises. For instance, no well-known dataset is used by numerous businesses to bootstrap their own AI algorithms in the financial services industry. The development of well-regulated, internal, and private data cataloging, labeling, and annotation "marketplaces" for businesses can currently be accelerated by several off-the-shelf technologies, though.?

The first step to more revolutionary use cases, such sector-specific intelligent recommender systems, is to crowdsource the unglamorous, data housekeeping job.?

2. Think like a designer.?

You might be wondering how the design or the user experience of an app could possibly matter in the field of AI engineering and data science.?

"Good design is as little design as possible," said Dieter Rams, one of the greatest technology designers.?

Without considering the needs of the end user, we cannot expect business data/AI products to be widely adopted. For instance, a consumer at home would prefer an AI device with vocal capabilities, whereas a contact center might feel more at ease with text-based chat software.?

The experience for enterprise developers may even take the shape of a well-designed #AI application programming interface (API) that incorporates contextual intelligence into a bigger enterprise application.?

In a Fortune 100 company, an API-based approach to AI might even be more pertinent than other strategies. It is preferable to use AI APIs to integrate domain- or industry-specific intelligence into enterprise applications rather than creating them from scratch.?

One of the better illustrations of this strategy is the Open AI API DALL·E API., which is constructed on top of the #GPT-3 model, a language model that makes use of deep learning to generate writing that is human-like. It is amazing to see how many innovative use cases have been developed on top of this simple API.?

Imagine intelligent APIs enhancing the apps used by knowledge workers and financial advisors across all industries and sectors.?

3. Accept the "culture" of your community's toolbox.?

The common tools and data assets that data scientists and engineers use to communicate across silos represent the technological culture of a business. Many CIOs were still rejecting the open-source revolution not too long ago. It would appear that the democratization of data and AI must require paying attention to toolkit standards with an enterprise's data scientist and developer community given the proliferation of open ecosystems like code repositories and code messaging boards.?

Any attempt to impose standards on data science toolkits alone is likely to be unsuccessful. Which Python notebook, for instance, is most used within your company? Has a rogue wiki been set up to compile all the data elements? Exists a shared and reusable data model that analysts can use to organize standard flat files within a given industry? Do your analysts use pre-trained R-algorithms for your domain??

4. Audit your algorithms frequently.?

Who reviews a formula that determines whether a particular sector receives a mortgage? What about a system that evaluates your prior performance to determine if you should be allowed a job interview or retained in your current position??

Recent news reports have discussed huge corporations that have experimented with employing algorithms to rank job interview candidates or even assess industrial workers' job performance. The adoption process is challenging when arguing against allegations of intentional or unintentional bias in algorithms due to the "black box" nature of some types of algorithms.?

This demonstrates how crucial it is in the modern world to "verify trust" in data and algorithms. A point-in-time exercise is not what this is. I counsel groups to develop a culture of regularly testing and reviewing algorithmic and data products within an organization. To make "black box" models more transparent, current research is employing algorithm audit approaches such as contrastive explanations.?

5. Support a program for community leaders.?

In the end, democracy is about the people and bottom-up leadership. A new center of excellence should be run by top performers who can also act as community leaders. Look for contributors to open-source code repositories and technical blogs if you don't already know who your stars are or are actively recruiting.?

If your business has a history of poor adoption of data science-specific knowledge management systems, you may need to encourage or gamify contributions to these systems internally.?

Closing Thoughts?

Even though they are merely guidelines, it's important to emphasize how crucial it is to democratize data and AI across the board. By democratizing these departments, employees are given the freedom to come up with their own solutions, which saves them time and enables the AI and data teams to concentrate on strategic projects rather than on ad hoc support.?