Will AI SDKs Democratize AI Development?

Google+ Pinterest LinkedIn Tumblr +

Every new technology follows a similar trajectory of elitism and then democratization. The first computers, for instance, could only be used by highly skilled computer engineers — a lay person would have been flummoxed by the massive machines. Today, however, most people not only use computers, but are dependent on them. Technologies from cars to elevators have all followed this same progression. Have we reached that point of democratization with AI? Are we at the point where AI will move from being an abstract idea only broached by the biggest players in tech, to being easily developed by the average technology enthusiast?

According to the CEO of an AI SDK (software development kit) with $13M in Series A funding, there’s already an arms race for developing the most user friendly, accessible machine learning/AI SDK. The primary use cases for AI SDKs are in industries with a wealth of data to be analyzed that could be hugely impacted by Machine Learning. These industries include banking, insurance, healthcare, customer service, and marketing. Together, they account for a $300B opportunity.

The AI SDK Landscape is Growing — These Are the Players

NewtonX surveyed three executives at different AI SDKs to gain an understanding of the landscape. The major players include Bonsai, Prowler.io, Skymind, Petuum, Kyndi, Primer.ai, H2O.ai, Cycorp, Embodied Intelligence, Ayasdi, Rage AI, Leapmind, Kyndi, and Primer AI. While each company has a different amount of funding, and targets different industries, they all aim to solve the same problem: AI and ML enterprise solutions today are:

  • Expensive
  • Skill-intensive
  • High maintenance

Solutions also tend to have highly specialized applications that are difficult to reproduce or distribute. SDKs provide a potential solution to all three of these problems: they allow programmers to integrate AI into their existing processes.

Each of the companies listed above has provided a possible solution, each targeting a slightly different vertical. Here’s how AI SDKs work for different applications:


The biggest use case currently for AI in banking is fraud detection through deep learning models. Pattern-based anti-money laundering (AML) systems can identify suspicious patterns faster and more accurately than traditional rules-based AML software. Fraudulent activity can also be identified in near real-time so that customers are rapidly alerted and receive the best care possible.


AI can also be used to detect insurance fraud more rapidly and accurately than traditional software can. Additionally, image recognition technology can process claims by automatically classifying photos and keywords.


As we’ve previously written, AI has numerous applications in healthcare — from preventative care to diagnosis through image recognition. It can also be used to personalize care and predict more accurate drug matching. These use cases for AI have already been applied, but they’re not available to the average healthcare practitioner. The goal of an AI SDK would be for researchers, doctors, and hospital administrators to easily leverage insights to provide superior care.


This is one of the biggest potential markets for AI SDKs. Predictive analytics is a hugely powerful tool for increasing engagement, targeting potential buyers on social media and other channels, and personalizing messages for email marketing and other outreach programs. It can also be used for recommender systems (such as Netflix’s “Recommended for you”) to upsell online shoppers or improve engagement on streaming or gaming platforms. The opportunity in this market is huge, not just because of the myriad applications of ML-based analytics systems, but also because marketers tend to gravitate toward user friendly products, as many marketers do not have a data-heavy background. Tools that are easy for marketers to use will be rapidly adopted.

Customer Service:

AI-powered customer service SDKs automatically classify tickets, autoroute tickets to agents, and deflect tickets by suggesting FAQ articles to customers. These platforms typically also include the option for chatbots to automate a significant portion of the customer service workflow. AI can also be instrumental in customer service by offering real-time insights about emerging issues


The last vertical for AI SDKs is in decision-making software, which can be applied to gaming, finance, autonomous vehicles, and drones. This vertical is still very much in the development phase, but as autonomous machines become a part of everyday life, it is sure to expand rapidly.

SDKs will increase adoption of AI solutions, but development is still relegated to specialists

While AI SDKs will bring the power of AI to a wider audience, the average doctor or marketer will not be actually developing AI-powered systems. Rather, it will take on a similar shape to customer service SDKs today — wherein the initial setup and goals for the technology are established by developers, and the end user — customer service agents — largely interacts with it as a question and answer tool (i.e. “I need to see a forecast of our ticket volume for Topic X next Thursday).

According to the CTO at an AI SDK that targets fraud detection and cyber security, for some applications, this level of user friendliness is still a long way off. When it comes to predictive analytics, image recognition, auto classification, and real-time insights, AI is absolutely ready for enterprise applications through a robust SDK. These systems are also perfect for implementation in enterprise settings today, as they continually update their environmental models in light of real-time data they receive — thereby becoming increasingly accurate over time. This gives an advantage to early adopters. But when it comes to decision making in real world environments, as we’ve already seen with self-driving cars, machines are not superhuman yet.

“AI is incredible at processing and analyzing data. That’s what we have today,” explained the CTO. “But in many respects AI development is still far too complex to be taken out of the hands of data scientists and engineers.”



About Author

Comments are closed.