An alternative approach to the neural network frameworks powering AI models like OpenAI’s is emerging with increasing prominence. This approach, known as symbolic AI, utilizes task-specific rules, such as rewriting text, to tackle broader challenges effectively.
Symbolic AI is capable of addressing problems that often confound neural networks, and recent studies reveal its potential for scalability—something that traditional symbolic systems have historically struggled with due to inefficient computation.
The advancements in scalability have spawned a surge of startups deploying symbolic AI across various sectors. Notable examples include Orby and TekTonic, which focus on enterprise automation tools, and Symbolica and Unlikely AI, the latter being founded by Alexa co-creator William Tunstall-Pedoe. Recently launched from stealth mode is Augmented Intelligence, which has secured $44 million in funding from investors like former IBM President Jim Whitehurst.
Augmented Intelligence has developed a conversational AI platform, Apollo, which the company describes as merging two seemingly disparate technologies: neural networks and symbolic AI. This combination results in a model that is both actionable and trainable, delivering more predictable and "agentic" results than conventional neural network systems. For instance, instead of merely providing flight booking instructions, a user could utilize Augmented Intelligence’s AI to access fare information and make reservations directly, as highlighted in the company’s information brief.
"There’s a significant distinction between chatbots like ChatGPT, designed primarily for user interaction, and conversational agents that execute tasks on behalf of businesses," explains Elhelo. "When you connect the AI to tools for retrieving information or taking action, the model moves beyond its training data, which often leads to a decline in intelligence quality."
Augmented Intelligence's AI powers chatbots capable of addressing a wide array of inquiries (such as “Do you price match on this product?”), seamlessly integrating with existing company APIs and workflows. According to Elhelo, the AI has been trained using conversation data from tens of thousands of human customer service representatives.
When considering why businesses might opt for Augmented Intelligence over other AI providers, Elhelo points out that its AI excels at leveraging external sources to fulfill tasks. While AI from OpenAI, Anthropic, and others can also utilize tools, Elhelo argues that Augmented Intelligence's performance surpasses that of neural network-based solutions.
The explainability of the AI is another key advantage. It provides a detailed log of its responses and the reasoning behind them, enabling companies to refine and enhance its effectiveness. Notably, it does not rely on training a company’s sensitive data—only utilizing resources to which it has been granted access for defined contexts.
“Apollo does not necessitate training on proprietary company information,” states the company fact sheet, “and adheres to the deploying company’s rule-based guidelines.”
This aspect of not utilizing customer data will likely resonate well with businesses cautious about sharing confidential information with third-party AIs. Reports have indicated that Apple prohibited its staff from using OpenAI tools last year over concerns of data breaches.
While some of Elhelo's assertions, such as the claim that Augmented Intelligence’s AI can "eliminate hallucinations," may be seen as overstated, the 40-employee company continues to gain traction, including a recent strategic partnership with Google Cloud to incorporate its models into the platform.
Though revenue figures remain undisclosed, Elhelo did share that Augmented Intelligence's latest $10 million funding round assessed the company’s valuation at $350 million—a notable figure for an AI firm that has only recently launched its product and wasn’t established by an established AI industry leader.
"Traditional language models typically depend on transformer architectures, which are excellent at recognizing patterns and generating language," Elhelo states. "However, these architectures can falter in scenarios that require action, decision-making, or interaction with tools. Apollo’s neuro-symbolic architecture effectively addresses these challenges."
Correction: This article has been updated. The initial version incorrectly attributed certain statements to the CEO, which were actually from a company-provided fact sheet. It also mistakenly named the firm that led Augmented Intelligence’s latest funding round, per company clarification. Additionally, references to the company Delegate have been removed due to a communication error that prevented the company from responding to prior inquiries.
Ranking
