Minerva is a continually learning AI that enables users to build their own AI with no code and no labeling. The purpose of Minerva is to enable users to boost cognitive throughput and surface game changing predictive insights from information complexity.
Minerva is comprised of hundreds of modular components for processing and contextualizing dynamically changing external and internal unstructured and semi-structured data. Minerva understands context with human level accuracy and explainability. As Minerva learns, it accumulates knowledge that it draws from over time to understand increasingly general contexts.
Today, we use Minerva internally to build all of our products and solutions. In the near future, we’ll be releasing Minerva directly to end users with the aim of unleashing the extraordinary creativity and genius of our customers.
Accrete Data Hooks ingest data from a plethora of messy unstructured and semi-structured external and internal data sources including social media, news, filings, PDFs and foreign language documents.
Knowledge Functions are modular building blocks of cognition that process and contextualize data to extract predictive insight. Knowledge Functions perform cognitive tasks such as automated entity extraction, entity normalization, entity linking, influence modeling, topic classification, semantic search, anomaly detection and semantic translation.
Minerva enables natural user interaction with various dashboards and APIs for consuming insight, providing feedback and securely integrating Accrete’s AI products into existing enterprise workflows.
Users can assemble Minerva APIs such as Data Hooks for TikTok or Discord Listening, and Knowledge Functions for Bot Filtering and Authentic Engagement to create their own AI workforce.
Alternatively, users can configure pre-packaged templates that solve entire classes of information complexity problem such as talent discovery or lead generation.
We are working with legislators to help establish universal standards for performance, explainability, bias, and ethics in AI. AI is already powerful enough to manipulate human psychology through misinformation and deep fakes. We are setting standards for ethical AI-focused around respect for privacy, non-manipulation, inclusivity, and respect for individual rights.
We establish performance benchmarks for each of our products with domain experts. It’s critically important to quantify the true cost of an error in order to set appropriate thresholds for accuracy. Certain jobs require higher accuracy because there could be dire consequences of an error whereas other jobs do not require high accuracy because of low cost of an error. Only when our AI achieves expert-level accuracy do we push our products into production.
Although AI explainability is an extremely complex topic, we take it very seriously. We believe that human level explainability is the key to AI adoption. Explainability and accountability are two sides of the same coin. Humans must understand the reasoning behind AI generated insights in order to hold the machine accountable.
Both humans and machines have bias. The entire purpose of our continuously learning AI is to accumulate knowledge that reduces human bias which in turn leads to reduced machine bias through natural interaction. Success to us is a virtuous cycle of bias reduction between man and machine that leads to better decisions and shorter decision cycles.
Real AI is AI that learns exponentially faster than the effort required to train it and allows for transfer learning. That’s only possible when these three pillars are incorporated together. We’re delivering real AI.
Our architecture enables data to continuously enter the system while simultaneously incorporating it into our models dynamically. Our continual learning increases model accuracy over time and maintains accuracy even when data patterns change.
Accurately representing the meaning of data requires an accurate understanding of the context surrounding the data. Accrete’s continual dynamic learning selectively applies regularizer and progressive network approaches to grow and recalibrate its neural architecture while incorporating on-the-fly human feedback. The result is the ability to rapidly learn contextual meaning and adapt to contextual changes over time.
The AI industry’s dirty secret is that in order for models to function well, they need inordinate amounts of initial and ongoing human training and expert labeling. Accrete’s models learn from sparse training data. Our AI automatically defines rules and extracts the features contained within expert-provided ground truths.
The digital revolution has created unprecedented information complexity. Data is constantly growing in size and complexity while dynamically changing its context as well. The only solution is AI that is hands-off, continually learning, and applying new knowledge to new problems automatically.
The more people interact with AI in the form of explicit rules and manual labeling the more human bias is introduced to the model. Our AI requires minimal training and labeling, it learns implicitly and dynamically while reaching outstanding accuracy and performance levels.
AI that needs constant maintenance from both ends and that heavily relies on manual labeling simply cannot scale. Scalable AI is one that can automatically adjust to contextual changes, new data sources, and unfamiliar data points without human intervention.
Let’s discover what we can build together