Alternatives to the neural network architectures at the heart of AI models are emerging, such as OpenAI's o1. This AI, called symbolic AI, uses rules related to specific tasks, such as rewriting lines of text, to solve larger problems.
Symbolic AI can deftly tackle some of the problems that neural networks struggle with. And recent research shows it's scalable. (Historically, symbolic architectures have not been computationally efficient.)
Breakthroughs in scalability have enabled companies like Orby, TekTonic (which builds enterprise automation tools), Symbolica, and Unlikely AI (founded by Alexa co-creator William Tunstall-Pedoe) to apply symbolic AI to a variety of fields. A source of startups is emerging. One of the latest ventures to emerge from stealth is Augmented Intelligence, which received $44 million in backing from investors including former IBM president Jim Whitehurst.
Augmented Intelligence builds conversational AI that claims to be more predictive than typical neural network-based systems and “agent-like,” the latest AI buzzword. For example, instead of simply answering a question about a flight to Mexico with instructions on how to book it, Augmented Intelligence's AI can book a flight by providing a list of fares, CEO Ohad Heruelo said.
But wait a minute. Can't we already do that with ChatGPT? Yes, Elhero admits. But he argues that it requires more setup and manual integration than augmented intelligence technology.
“There is a big difference between chatbots like ChatGPT, whose main purpose is to chat with users, and conversational agents, which perform actions or work on behalf of a company,” Elhelo told TechCrunch told. “When you connect AI to a tool to obtain information or take action, the model no longer relies on training data, and the quality of intelligence drops dramatically.”
Elhelo co-founded Augmented Intelligence with Ori Cohen in 2017. At the time, the company had another name: Delegate (previously Stuff). It also had a different, more polarizing mission: delivering AI-powered apps that customers could delegate. Delegate work to low-wage gig workers.
Elhelo gave his elevator pitch for Delegate in a 2019 interview: [a business exec’s] Time is worth it. For example, $50. Why spend money calling customer service or reading customer reviews online when there is someone in India, East Jerusalem, or the Philippines who will do it for you for less?”
Delegate's scathing Glassdoor review suggests that it's hard for gig workers to make much money at all. Workers say the platform is difficult to use due to bugs and a “standby” system, little to no support or training, and an incentive structure that pays little despite promises of high wages. I'm dissatisfied with things like that.
So, forced to change direction, Elhelo and Cohen took the AI route.
Augmented Intelligence's AI can be integrated with a company's existing APIs and workflows to power chatbots that answer questions on a variety of topics, such as “Does this product's price match?” Elhelo claims this AI was trained based on conversation data from tens of thousands of human customer service agents.
I asked Elhelo if the conversation was from a delegate. He would not say whether the agents received compensation for their contributions or whether they were informed that their data would be used in this way.
Putting aside people's feelings about branded chatbots for a moment, why would companies choose augmented intelligence over other AI vendors? As an example, Elhelo's AI uses external intelligence to complete tasks. They say they are trained to use tools that capture information from sources. OpenAI, Anthropic, and other AIs have tools available as well, but Elhelo claims that augmented intelligence AI will perform better than neural network-driven solutions.
AI also provides logs about how it responded to queries and why, making it more explainable and giving businesses a way to fine-tune and improve AI performance, Elhelo said. he claims. It also doesn't train on corporate data, Elhelo says, and only uses resources that it has been given permission to access in a given context.
“Augmented Intelligence requires no training on enterprise information and takes into account the implementing company's rules-based directives,” said Elhelo.
This part of not training on customer data will certainly appeal to companies wary of exposing their secrets to third-party AI. Apple, in particular, reportedly banned employees from using OpenAI tools last year, citing concerns about sensitive data being leaked.
Now, Elhelo makes some questionable claims, such as that an augmented intelligence AI can “remove hallucinations” (which is definitely impossible). But the 40-employee company still seems to be doing well, recently securing a strategic partnership with Google Cloud to bring its model to the platform.
Elhelo did not share information about its revenue. But he told TechCrunch that Augmented Intelligence's last $10 million funding round, led by New Era Capital Partners, valued it at $350 million. This is a relatively high number for an AI vendor that only recently brought products to market (and was not founded by). AI industry giant).
“Traditional models are good at pattern recognition and language generation,” Elhelo says. “But these architectures fall short in situations where models need to perform actions, make decisions, or interact with tools. Apollo's neurosymbolic architecture and what it brings to the enterprise Possibility to solve these problems.”