Développez l'apprentissage automatique prédictif avec Flink | Atelier du 18 déc. | S'inscrire
Artificial intelligence (AI) solutions that deliver results that businesses can count on are few and far between. Historically, they’ve been difficult to use, often requiring significant monitoring and oversight to prevent or mitigate potential mistakes. Despite this, we’ve been hearing that “the AI renaissance” is here for years.
Over the last year, generative AI tools and large language learning models (LLMs) have overtaken discussions around the future of productivity. McKinsey estimates that generative AI could add $2.6 trillion to $4.4 trillion of value to the global economy across dozens of use cases. But for companies to actually realize this value, they need to be able to trust these tools to accelerate decision-making in mission-critical use cases. Unfortunately, LLMs are black boxes that are biased and unreliable by design.
Enter Elemental Cognition (EC). Its deep reasoning platform solves complex problems when you can’t afford to be wrong by combining LLMs with other AI techniques. EC accelerates and improves critical decision making in complex, dynamic situations where trust, accuracy, and transparency matter.
In this blog, we’ll cover how this innovative platform uses data streaming, powered by Confluent Cloud, to ensure its reasoning engine is being fed the right data, from the right source, no matter where it lives.
Learn more about how data streaming and AI are better together.
When Dr. David Ferrucci, original inventor of IBM Watson, launched EC, his goal was to build an AI platform that augments and overcomes the limits of human reasoning.
Generative AI might initially seem to work faster when used for decision-making use cases, such as travel planning, but would ultimately lead to results ranging from inconvenient to disastrous. As much excitement as we’ve seen over the potential of LLM like Open AI’s ChatGPT, this technology has undeniable shortcomings. While not unique to generative AI, hallucinations—when generative AI confidently and fluently produces misinformation—pose a significant hurdle for businesses to overcome if they want to take advantage of LLMs to support semantic analysis of large data sets. Retrieval-augmented generation (RAG) is a pattern that helps address this, but it needs a real-time, contextualized, and trustworthy knowledge base.
LLMs alone also struggle to solve complex problems where there are clear right and wrong answers, lots of intertwined rules, and resource tradeoffs, and where human experts are even prone to error. This includes use cases like scheduling, planning, resource optimization, product configuration, and research and discovery. For businesses to make confident decisions, they need decisions grounded in the rules of the business that are provably correct.
EC’s AI platform is designed for exactly this type of mission-critical use case. The EC reasoning engine leverages Confluent Cloud to exchange data across various processing stages, scale when needed to increase compute capacity and recover from failure without losing message order or causality.
To create an AI platform that reliably solves complex problems in real time, the Elemental Cognition team built and combined two key elements:
Custom domain models built on their reasoning engine, which are tailored to each use case to produce provably correct results—whether handling resource scheduling or accelerating reviews of medical research
LLMs designed to support semantic processing of large data sets and natural language interfaces (NLIs) for an interactive dialog or chat that fluently engages customers
Elemental Cognition ensures that your business never relies on the LLM itself being the source of truth. Instead, its reasoning engine provides provably correct outputs based on your custom business logic. This is the only way to accelerate mission-critical use cases, because it ensures the solutions it provides are both accurate and optimal, a task that LLMs alone can’t achieve. And unlike solutions that operate black-box models, the platform exposes the reasoning and data sources behind its reasoning engine and the results it produces.
Elemental Cognition built two products based on this platform: its Cogent application and Cora chatbot.
Cogent enables business process owners and subject matter experts to build domain models capable of automating answers to complex and dynamic situations. These systems feed all of their data into Confluent Cloud so it’s ready for downstream consumers and can be used to build interactive and highly accurate problem-solving applications. These applications offer natural language interfaces powered by LLMs, which are capable of fluently interacting with customers and helping them find optimal solutions in real time. However, the answers and solutions themselves are determined by EC’s proprietary reasoning engine, which gets provably correct results with no hallucinations. EC’s combination of these technologies ensures customers get the best outcome, and that the business can have confidence in the source and correctness of the solution offered.
Users can also take advantage of Cora, Elemental Cognition’s collaborative research and discovery assistant. Cora automates ontology induction, enabling it to adapt to new domains and data quickly, using NLU based on unsupervised learning methods. Users can explore and search content using a rich hierarchy of concepts and relationships automatically extracted from content, providing control over precision and recall.
Cora harnesses causal models that provide a transparent foundation for reasoning. Users can see the logical inferences that shape and validate answers to make evidence-based decisions with confidence.
Cora finds, analyzes, and integrates the most current sources into models without retraining, so users can discover, track and understand signals in data by explaining causal relationships between concepts, entities, and events through time.
Elemental Cognition uses Confluent Cloud to build its streaming data pipelines. As data moves through the AI platform, it progresses and is indexed through roughly a dozen stages. Using an Apache Kafka® service for all data ingestion and processing allows the platform to preserve order and causality for transaction workflows, a critical capability for maintaining data provenance.
The streaming data pipelines need to be resilient such that correctness and progress are maintained in the face of intermittent failures. Confluent minimizes Elemental Cognition’s operational burden, removing maintenance, scaling, reliability, and support concerns from the equation. The company can remain focused on its pipeline functionality to securely ingest, process, and index proprietary data from customer systems.
By relying on stream processing rather than batch processing, Elemental Cognition can incrementally incorporate new information and update predictive models. This allows the company to continuously update data inferences that their customers depend upon.
LLMs process data with each query, rather than in batches of queries. Adopting Confluent Cloud has enabled the team to update data indices on a daily basis, ensuring that new customer data is analyzed and business decisions are made using the most up-to-date context.
Additionally, Elemental Cognition relies on Confluent’s expertise to help minimize message losses and system failures that would negatively impact the businesses that rely on its AI platform.
Elemental Cognition has multiple demo use cases showcasing how its reasoning engine helps accelerate business decision-making and knowledge discovery across multiple industries, all while ensuring accuracy.
One promising opportunity is in life sciences research, an area in which semantic processing is key for accelerating literature reviews, but inexplicable AI hallucinations aren’t an acceptable risk.
That need for total accuracy means it's critical that the platform always preserves message order and causality. By ensuring reliable provenance and platform performance, Confluent Cloud helps Elemental Cognition ensure that research never slows down with real-time capabilities, plus ensures that information from new sources can always be tied back to the source.
Elemental Cognition can help life sciences companies cut early-stage drug research from days and weeks to hours or minutes, providing significant value for companies and the patients waiting for new treatments.
Cora combines NLU and logical reasoning to generate accurate summaries of data found in archives like PubMed, which are updated constantly and require real-time processing of new information for researchers to stay up to date. Alone, LLMs would not be able to deliver results with similar precision since they are unable to reason logically, induce ontologies, and identify causal relationships.
In addition to accelerating GenAI powered intelligent research and business decision-making, Elemental Cognition is working on transforming customer experiences as well.
With its oneworld booking tool, Elemental Cognition’s combination of LLMs and its reasoning engine can yield useful, reliable results while saving customers significant time.
Users can choose up to 16 cities they want to visit for a trip lasting up to a year. In the video below, you can see how the booking tool works when trying to book a six-month trip to visit three of Confluent’s offices across North America, Europe, and Asia.
The interface guides the user through a seemingly straightforward process, but behind the scenes, there are roughly a dozen pages’ worth of business rules. For example, users can only fly on carriers that are part of the Oneworld alliance.
This kind of application has potential for other industries as well—for example, for degree or education pathway planning in higher education. Much like with travel planning, this use case would have similarly complex “itineraries” subject to resource availability (i.e., class capacity), compatibility of itinerary entries (e.g., course prerequisites), and individual user preferences.
The underlying logistics problem is something that neither traditional automation technologies nor generative AI could solve alone. Elemental Cognition’s reasoning engine is purpose-built to solve these kinds of constraint satisfaction problems.
Inability to trust the answers AI generates has been a huge barrier to AI adoption across many mission-critical business functions and use cases. This includes research and discovery, complex product configuration, resource management, staffing, and planning. By overcoming this barrier, Elemental Cognition expects to see much more expansive use cases, particularly in life sciences and the public sector.
As the company continues to validate new use cases and onboard new customers, the team plans to use Confluent for platform telemetry, which will help optimize performance as more data pipelines are exposed to customers.
This blog explores how cloud service providers (CSPs) and managed service providers (MSPs) increasingly recognize the advantages of leveraging Confluent to deliver fully managed Kafka services to their clients. Confluent enables these service providers to deliver higher value offerings to wider...
With Confluent sitting at the core of their data infrastructure, Atomic Tessellator provides a powerful platform for molecular research backed by computational methods, focusing on catalyst discovery. Read on to learn how data streaming plays a central role in their technology.