An alternative to the neural network architectures at the heart of AI models like OpenAI’s o1 is having a moment. Called symbolic AI, it uses rules pertaining to particular tasks, like rewriting lines of text, to solve larger problems.
Symbolic AI can deftly tackle some problems that neural networks struggle with. And recent research has shown that it can be scalable. (Historically, symbolic architectures haven’t been compute-efficient.)
The scalability breakthroughs have fueled a wellspring of startups applying symbolic AI to various domains, like Orby and TekTonic (which are building enterprise automation tools), Symbolica, and Unlikely AI (founded by Alexa co-creator William Tunstall-Pedoe).
One of the newest ventures to emerge from stealth is Augmented Intelligence, backed by $44 million from investors including former IBM President Jim Whitehurst.
Augmented Intelligence has built a conversational AI model, Apollo, that it claims “combine[s] two seemingly opposite technologies (neural networks and symbolic AI) into a cohesive, actionable, trainable model.” This makes the results both more predictable and “agentic” — the latest AI buzzword du jour — than your typical neural network-based system. For example, instead of simply answering a question about flights with instructions on how to book, a business customer could use Augmented Intelligence’s AI to give a list of fares and book the flight for the customer, according to a company-provided fact sheet.
But wait — can’t ChatGPT already do that? Yes, CEO Ohad Elhelo admits. But he claims that it requires more setup and manual integration than Augmented Intelligence’s tech.
“There’s a big difference between chatbots like ChatGPT, whose primary goal is to chat with the user, and conversational agents that take actions or work on behalf of companies,” Elhelo told TechCrunch. “Once you connect the AI to tools — either to retrieve information or to act — the model is not relying anymore on its training data, and the quality of intelligence drops dramatically.”
Augmented Intelligence’s AI can power chatbots that answer questions about any number of topics (e.g. “Do you price match on this product?”), integrating with a company’s existing APIs and workflows. Elhelo claims the AI was trained on conversation data from tens of thousands of human customer service agents.
Setting aside for a moment peoples’ feelings on brand chatbots, why would a company choose Augmented Intelligence over another AI vendor? Well, for one, Elhelo says that its AI is trained to use tools to bring in info from outside sources to complete tasks. AI from OpenAI, Anthropic, and others can similarly make use of tools, but Elhelo claims that Augmented Intelligence’s AI performs better than neural network-driven solutions.
The AI is also more explainable because it provides a log of how it responded to queries and why, Elhelo asserts — giving companies a way to fine-tune and improve its performance. And it doesn’t train on a company’s data, using only the resources it’s been given permission to access for specific contexts, Elhelo says.
“Apollo does not require training on company information,” the company fact sheet says, “and takes into consideration the deploying company’s rule-based instructions.”
That bit about not training on customer data will surely appeal to businesses wary of exposing secrets to a third-party AI. Apple, among others, reportedly banned staff from using OpenAI tools last year, citing concerns about confidential data leakage.
Now, Elhelo makes some dubious claims, like that Augmented Intelligence’s AI can “eliminate hallucinations.” But the 40-employee company appears to be winning business nonetheless, most recently securing a strategic partnership with Google Cloud to bring its models to the platform.
Elhelo did not share info on revenue. But he did tell TechCrunch that Augmented Intelligence’s last $10 million fundraising round valued it at $350 million — a relatively high figure for an AI vendor that only recently brought its product to market (and wasn’t founded by a titan of the AI industry).
“Traditional language models rely on transformer architectures, which excel at pattern recognition and language generation,” Elhelo said. “However, these architectures fall short in situations where the model needs to perform actions, make decisions, or interact with tools. Apollo’s neuro-symbolic architecture and possibilities it unlocks for companies … solves those problems.”
Correction: This story has been revised from its original version. The original version attributed certain statements to the CEO, when in fact those statements came from a company-provided fact sheet. The original version also misstated the firm that led Augmented Intelligence’s last funding round, according to a company representative. The original version also asserted that the CEO did not respond to a reporter’s questions about an earlier incarnation of the company, Delegate. In fact, due to a communication error, the company was not given a chance to respond to questions about Delegate, therefore references to that company have been removed from the article at this time.