weav.ai | blog

Democratizing AI for the Enterprise

Written by Peeyush Rai | May 24, 2023 8:46:00 AM
Democratizing

verb

/dɪˈmɑː.krə.taɪz/

Make [something] accessible to everyone

A brief history of the democratization of technology

Over the last 50 years, technology has relentlessly democratized what was once accessible to just a few. For example, PCs democratized computing, the internet democratized access to information, smartphones democratized photography (among other things), and YouTube democratized learning.

It’s safe to say that this trend will continue.

However, it’s not just consumers who have seen the benefits. Starting with PCs, email, and word processors, and then progressing to spreadsheets and now complex business workflows, technology has changed the lives of business users for the better.

Or consider business reports, the lifeblood of every business. Twenty years ago, businesses had to hire specialized programmer analysts to analyze data stored in databases and legacy systems and build reports for managers and business leaders. Today, most business users and analysts can perform these tasks using a simple spreadsheet. Since easy access to metrics and reports is crucial for business decision-making, we can appreciate and applaud the humble spreadsheet’s role in democratizing business reporting and decision-making.

Technology democratization lifecycle

Although Gartnerˈs “hype cycle” has become a popular way to describe technology adoption, the “democratization cycle” framework is better because it charts the path of technology from research to democratization and beyond.

The Technology Democratization Lifecycle

While the stages are fairly self-explanatory, for illustrative purposes you could apply them to something that’s highly commoditized today — the cell phone.

Research led to the invention of the first cell phone by Motorola in 1973. Commercialization began with the DynaTAC 8000x, also from Motorola, in 1983. Democratization occurred in the 1990s when Nokia et al took off. Commoditization then brought cheap $25 “feature phones” to everyone.

This framework also helps us think about who has access to this technology at each stage. At the end of the day:

“The purpose and meaning of technology are complete only when they are accessible and usable by a large number of people, irrespective of their educational or professional background.”

So technology in the research phase is available to academia or very large tech firms that have captive in-house research teams. Quantum computers, for example, are at this stage today. In the commercialization phase, businesses, generally large-to-mid-size, can afford products and services powered by this tech. SMBs then gain access to the tech in the democratization phase. Finally, it’s the commoditization phase that makes the technology available to the average consumer on the street.

Finally, as we can expect, as technology becomes democratized, users don’t need to be experts to use it and benefit from it. For example, 25 years ago, in order to take a great photo you had to be an expert at adjusting focal lengths, swapping lenses, and measuring light exposure. Today you open your (smart)phone camera and the rest is magic. So as we progress through the four stages, the need for user expertise drops exponentially, and both usability and accessibility rise exponentially.

Democratizing AI — the good and the bad

With the loud buzz surrounding AI today, it’s natural to wonder about where we are with enterprise AI in this cycle. However, because AI covers such a vast domain, it’s hard to force AI in the aggregate into one of these four categories. Different manifestations of AI correspond to different stages of the cycle.

Great news for consumers

Consider AI and consumer technology. When I take a picture with my smartphone, AI embedded in the camera’s software works quietly to ensure I capture good photos. Users do not need to know the technical details of how it works; they simply use it every day. Today, this kind of AI has been largely commoditized.

Good-ish news for enterprises

In a business context, simple AI technologies like optical character recognition (OCR), speech-to-text, and document scanning have been commoditized. More complex AI technologies, such as self-guiding systems and healthcare imaging systems, have been commercialized.

Different AI technologies — and other non-AI ones — mapped across the democratization lifecycle

The not-so-good news

But amidst all the noise, there’s one aspect of business where AI has been surprisingly lagging: the use of AI to make decision-making easier, faster, and better.

Today, it’s still extremely difficult for the average business user, analyst, or leader to use AI to forecast and predict business outcomes, and make AI-informed decisions to optimize those outcomes.

This couldn’t be happening at a worse time because the volume of user and machine-generated data is at an all-time high, and not being able to use AI to parse this data is a major source of competitive disadvantage.

To be clear, businesses have laid a good foundation when it comes to the data layer. Many businesses have extracted their data from transactional systems and made it accessible to external tools either directly or through a central data warehouse or data lake. This allows business analysts to access their data and build charts and reports using mature and increasingly user-friendly analytics tools or even just spreadsheets.

But data is just the starting point.

Current barriers to democratizing AI for businesses

DIY — resource constraints, long time to value, and hit-or-miss results

Once the data is available, business leaders must do a number of things to build usable AI apps:

  1. Hire or recruit a data scientist…only to realize that a data scientist alone is not sufficient.
  2. Hire or obtain access to a team of data engineers, ML engineers, and DevOps engineers.
  3. Assemble the underlying infrastructure and “plumbing” necessary for processing the data reliably, at scale.
  4. Build and train machine learning models.
  5. Keep the models up-to-date.
  6. Address AI explainability and overcome the black-box problem, and more…

Even if they’re lucky to have the budget and resources, and their business gives them the luxury of waiting months or years, success and high-quality results are not guaranteed. Worse, by the time usable results are generated, customers and the competition will likely have moved on.

That’s the path for most enterprises that take the “DIY” or do-it-yourself approach.

Existing AI platforms — major assembly required

What about AI platforms that provide all the necessary building blocks “out of the box”? While this may seem like an attractive option at first glance, it is not viable in practice. These platforms often have a steep learning curve and still require expert teams of data engineers, data scientists, and ML engineers to assemble the blocks, build, and deploy production-grade AI applications.

Generative AI — adding more complexity

Introducing the Generative AI stack adds another layer of complexity to the infrastructure, adding ingredients like LLMs and vector databases. It also requires super specialized skills that are still limited to a few hundred people in the world. Additionally, most existing AI platforms do not support the Generative AI stack, which creates another obstacle in successful enterprise adoption.

AI for businesses: abstracting complexity away

Business analysts today create pivot tables in Excel on a daily basis, without worrying about the technical details. Similarly, when using Excel’s many functions, they don’t need to think about how the software performs the associated calculations.

Similarly, for AI to be democratized for businesses, analysts should be able to easily invoke different AI functions without understanding the underlying algorithms, libraries, or infrastructure. Stakeholders and analysts should also be able to interact with data and AI using natural language, just as they do everyday at work.

In the same vein, with Generative AI and foundation models, it is important to be able to use the business context (transactions, customer data, product data, etc.) to provide an intuitive user experience and enhance users’ productivity. It’s not enough to provide a simplistic chatbot because, as we have seen in the past, chatbots that lack business context are essentially useless.

Here’s an example of how a context-rich AI + business user interaction might look like:

Democratizing AI — getting the community involved

The universe of problems that can be solved by AI/ML spans every vertical and company size. It is therefore impossible for any one company to claim that its products or solutions can solve every one of those problems.

The good news is that there are experts in every domain who have deep expertise in understanding the business problems in that domain. In many cases, they have already developed innovative solutions to tackle those problems and may even share their solutions publicly in a GitHub repository, a YouTube video, or a blog post. Within an enterprise, they may go as far as creating and training non-production ML models.

Unfortunately, none of these artifacts or solutions can be readily used by business users inside or outside their companies because, as we discussed previously, a small army of engineers and other experts is required to build a complete production pipeline and develop bespoke applications using the pipeline’s output.

So what’s the best way to harness those experts’ knowledge and make it available to all other businesses experiencing similar problems?

In order to harness this expert knowledge and “set it free” so that business users can benefit from these solutions, there should be an easy way for data scientists and other experts to define a “blueprint” / “recipe” for a particular business problem and contribute it to a platform that can (a) make it easy for other business users to consume it using an intuitive “AI to business context” translation layer and (b) create a production-ready AI app using this recipe and potentially other recipes contributed by other experts.

Similarly, for Generative AI technologies, businesses should be have easy, on-demand access to foundation models that are ready to be fine-tuned with their proprietary data and context. This will help their users use it to solve their unique problems and, in the process, create intellectual property that gives them a competitive edge. Once again, the community has a role to play here with domain-specific open-source models and shared prompt repositories.

weav.ai — mission

weav.ai’s mission to democratize AI by making it as accessible to business users as a spreadsheet. weav.ai is designed to help enterprise business leaders, business analysts, and data scientists easily leverage the latest in AI, including Gen AI, GPT4, and foundational models, to drive tangible business results. weav.ai’s no/low-code approach allows businesses to leverage their existing skillsets, connect to any and all of their data sources, and put the latest AI innovation at their team’s fingertips. Finally, weav.ai’s platform is designed to let users can share their solutions and data science recipes with other users.

Please visit weav.ai to learn more, and follow weav.ai on LinkedIn.

About the Author

Peeyush Rai founded weav.ai in 2021. Before weav.ai, Peeyush co-founded Ciitizen, a patient-centric healthcare data platform that was acquired by Invitae. During his 25-year career in technology, Peeyush has co-founded multiple enterprise and consumer data startups, and he has held senior leadership positions at a range of technology companies. Peeyush earned a BS in Engineering from the Indian Institute of Technology (IIT) Bombay.