An alternative to the neural network architectures at the guts of AI models like OpenAI's o1 is to take a moment. Called symbolic AI, it uses rules related to specific tasks, equivalent to rewriting lines of text, to resolve larger problems.
Symbolic AI can cleverly solve a number of the problems that neural networks struggle with. And current Research has shown that it’s scalable. (Historically, symbolic architectures haven’t been computationally efficient.)
The breakthroughs in scalability have spawned a fountain of startups applying symbolic AI in various areas, equivalent to Orby and TekTonic (which develop business automation tools), Symbolica and Unlikely AI (founded by Alexa co-founder William Tunstall-Pedoe). One of the latest ventures to emerge from stealth is Advanced Intelligencebacked by $44 million from investors including former IBM President Jim Whitehurst.
Augmented intelligence builds conversational AI that it claims is each more predictable and more “agentic” – the most recent AI buzzword – than a typical neural network-based system. For example, as a substitute of simply answering an issue about flights to Mexico with instructions on the right way to book, augmented intelligence AI can provide a listing of fares and book the flight for you, says CEO Ohad Elhelo.
But wait – ChatGPT doesn’t work do it already? Yes, admits Elhelo. But he claims it requires more setup and manual integration than augmented intelligence technology.
“There is an enormous difference between chatbots like ChatGPT, whose most important goal is to speak with the user, and conversational agents that take motion or work on behalf of firms,” Elhelo told TechCrunch. “As soon as you connect AI to tools – either for retrieving information or for taking motion – the model not relies on its training data and the standard of intelligence drops dramatically.”
Elhelo co-founded Augmented Intelligence with Ori Cohen in 2017. Back then, the corporate had a unique name — Delegate (and Stuff before it) — and had a unique, more polarizing mission: providing an AI-powered app that allow customers delegate tasks to low-paid gig employees.
Elhelo gave the elevator pitch for Delegate in 2019 interview: “If an hour of (a CEO's) time is price, say, $50, why spend it on phone calls with customer support or reading online customer reviews when there may be someone in India, East Jerusalem or the Philippines , who’s willing to do that for much less?”
delegates scathing Glassdoor reviews suggest that it was difficult for gig employees to make much money in any respect. Workers complain about errors and a “waiting system” that made the platform difficult to make use of, little to no support or training, and an incentive structure that promised higher wages but rarely delivered on them.
Under pressure to pivot, Elhelo and Cohen selected the AI route.
Augmented Intelligence AI can power chatbots that answer questions on any variety of topics (e.g., “Are the costs adjusted for this product?”) and integrate with an organization’s existing APIs and workflows. Elhelo claims the AI was trained using conversation data from tens of 1000’s of human customer support agents.
I asked Elhelo if the conversations were from Delegate. He wouldn't say – nor would he say – whether agents were paid for his or her contributions or whether agents were informed that their data can be utilized in this manner.
People's feelings about branded chatbots aside, why would an organization select augmented intelligence over one other AI provider? Well, for one, Elhelo says its AI is trained to make use of tools to include information from external sources to finish tasks. AI from OpenAI, Anthropic and others can leverage tools in the same way, but Elhelo claims that augmented intelligence AI performs higher than neural network-driven solutions.
AI can also be more explainable since it provides a log of how and why it responded to requests, Elhelo said – giving firms a solution to optimize and improve their performance. And it doesn't train on an organization's data, but only uses the resources it's allowed to access in certain contexts, says Elhelo.
“Augmented intelligence requires no training on corporate information,” said Elhelo, “and takes into consideration the rules-based instructions of the deploying company.”
Not training based on customer data will definitely appeal to firms wary of showing secrets to third-party AI. Apple, amongst others allegedly banned its employees from using OpenAI tools last 12 months, citing concerns about disclosing confidential data.
Now Elhelo makes some dubious claims, equivalent to that augmented intelligence AI can “eliminate hallucinations” (which it definitely can’t). Still, the 40-employee company appears to be doing good business and recently entered right into a strategic partnership with Google Cloud to bring its models to the platform.
Elhelo wouldn’t share any income information. But he tells TechCrunch that Augmented Intelligence's latest $10 million funding round, led by New Era Capital Partners, valued the corporate at $350 million – a comparatively high number for an AI vendor that only recently launched its product delivered to market (and was not founded by one). Titan of the AI industry).
“Traditional models excel at pattern recognition and language generation,” said Elhelo. “However, these architectures fall back on situations where the model must perform actions, make decisions, or interact with tools. Apollo’s neurosymbolic architecture and the probabilities it opens up for firms solve these problems.”
