The unstructured data-focused artificial intelligence startup Clarifai Inc. said today it’s launching Clarifai Community, a free service aimed at promoting collaboration and sharing in AI.

Clarifai is a well-funded startup that has hit upon the somewhat novel idea of helping companies use massive volumes of unstructured and unlabeled data to train AI models. The company has created several application programming interfaces for natural language processing, audio recognition and other AI use cases.

With Clarifai, developers can access tools that enable them to use unstructured data to train their AI models. For instance, the Scribe Automated Data Labeling tool helps to automatically label data for model training. Meanwhile, Clarifai Spacetime Search uses AI to automatically index unstructured data for deep learning training.

It also provides an Edge AI service that layers AI onto data streams using local hardware such as cameras and drones. Not least, Clarifai’s AI Lake is a data lake that’s designed specifically to store AI training data.

The new Clarifai Community service is intended to encourage AI practitioners to share their resources and collaborate with one another to build new AI use cases. The company said it will serve as a distribution channel for AI, collecting hundreds of useful resources in one place that can be accessed through its APIs. In addition to Clarifai’s own models, users will be able to leverage those built by various hyperscalers, academic institutions and startups, the company said.

Clarifai Chief Executive Matt Zeiler said there are fewer than 100,000 data scientists in the world who possess the skills required to build complex deep learning models, hence the need for more collaboration and sharing. “Clarifai Community empowers the world’s 50 million software developers and even more nontechnical users to solve the world’s hardest problems,” he said. “This is a transformative idea.”

Through the Clarifai Community interface and its drag-and-drop feature, the company explained, users will be able to connect different AI models visually into powerful workflows. In this way, the models and workflows can be thought of as “building blocks” that can be combined to create more powerful and useful AI applications, with Clarifai doing the integration in the background. So not only does Clarifai Community encourage sharing and collaboration, but it enables teams to save time spent on the coding that’s usually required to create more complex AI models.

As an example, Clarifai said users can create a workflow that pipes audio transcription into a text summarization model to create a tool that can accurately summarize business meetings. That could then be taken to another level with the use of audio transcription models for foreign languages, making it possible to create a translated summary of whatever was said in a meeting. The beauty of Clarifai Community is that once a workflow has been created and publicly shared, it will then be made available to anyone else within the community.

Andy Thurai, vice president and principal analyst with Constellation Research Inc., said Clarifai’s main differentiator is that it caters not only to unstructured data, but also specifically to unlabeled data. That’s useful because most companies can already create deep learning models with unstructured data. The problem, Thurai said, is that this unstructured data needs to be annotated, labeled and classified before it can be used, and that is a major problem for most enterprises.

“Clarifai claims to use AI to label the data automatically. While the accuracy rate is yet not known, it sounds compelling,” he said.

The analyst said Clarifai’s ability to label unstructured data, train the models and then deploy them to the edge is also extremely cogent. Further, he believes its ability to index unstructured data to make it searchable can fill another black hole for many organizations.

Regarding today’s announcement, Thurai said it’s an interesting idea to enable the community to build AI models and share them through public or controlled domains. “Other community members can use those models as a base and retrain them, customizing those models for their own needs, taking a lot less time to do so. It’s similar to the language transformer concept used by GTP3 and Jurassic-1, but it sits in the computer vision category,” he explained.

Clarifai Senior Vice President of Platform Alfredo Ramos said the company is redefining the way in which AI is distributed among organizations. “We are making the world’s best AI resources available to everyone and letting people contribute their own for unlimited collaboration using no-code tools accessible to all skill levels,” he said.

Image: Clarifai