AI is not inherently intelligent, but do we need our computers and machines to think like us?
That's part of what Peter Voss, CEO & Chief Scientist at Aigo.ai, was left wondering after he exited his first software company. The hunt for his next opportunity turned into an exploration into giving AI common sense — AI that can reason and learn.
He’s been on that journey for 20+ years and spent 5 of those years studying all different aspects of intelligence, gaining a well-rounded understanding of how people learn, how our intelligence differs from animals, and what IQ tests measure.
His studies culminated in the theory of how to build a thinking machine — from there, he launched his first AI company and has been developing his theory into practice ever since, pushing the limits of AI into new and exciting realms.
The capabilities and limitations of AI intelligence models
As impressive and useful as AI software can be, there are, of course, inherent limitations. The ChatGPTs of the world can only go so far in delivering a reliable experience.
“It's good at faking intelligence, and, of course, it can do amazing things,” Voss said. “But the problem is, it's not reliable. It's not aware of what it's saying at all. There is no metacognition. So there are a couple of components missing to truly make it human-like and human-level intelligence.”
ChatGPT spells out its limitations in its name: Generative (G) Pre-Trained (PT), making it a read-only model. A powerful one, but still limited by having no ability to learn interactively.
“Whatever you teach, it does not change the model, which is very different from how humans operate,” Voss said. “We can hear one word or two words, and it can completely change the way we perceive things.”
This fact, paired with a lack of metacognition, differentiates GenAI from human intelligence and limits its usefulness in customer service and business environments.
Voss believes the answer to this conundrum is Cognitive AI, as opposed to Statistical or Generative AI.
The necessity of embracing cognitive AI for achieving human-like intelligence
Developers can't brute force their way to human-level intelligence in AI. It will take careful study and trial and error to get there, and not through statistics alone. This thinking led Voss and his team to develop their “Chatbot with a brain,” which continues to evolve with the company.
“What we mean by 'Brain' is basically a cognitive engine that has deep understanding, uses context, has short-term memory, and remembers what was said earlier in the conversation — it has long-term memory, it remembers previous conversations with each individual,” Voss said.
The AI has a comprehensive cognitive engine that manages the whole conversation, not only processing select phrases.
One of Aigo.ai’s customers leverages the system to provide a concierge service to their 20 million customers. Aigo.ai's system remembers who to buy gifts for, what the occasion is, what kind of gift, and more. It's able to perform in this way because of the deep understanding and long-term memory.
A virtual, AI-powered assistant that doesn’t forget everything it learned by the next day.
“This is where the brain allows us to handle very complex conversations,” Voss said. “The same approach can be applied to banking or medical applications.”
Even with the success they’ve found, there is still much to scale and develop — even now, they’re working on leveling up their systems IQ and the level of complexity it can process successfully.
“There's still a fair amount of work involved in actually teaching the system the specifics of each company that uses our system,” he continues. “As our system becomes smarter, it will be able to self-configure to a much larger degree, where it will be able to ask the question, ‘Do you want me to handle it this way, or that way?’” he continues.
AI system's language abilities and self-directed learning
Aigo.ai is not a black box they throw endless data and computer power at — the cognitive architecture is entirely scrutable. Voss and his team can see what it’s doing and why it’s doing it.
“We have a curriculum that we go through to enable the system to handle more and more complex tasks, complex reasoning,” Voss said. “There are little surprises along the way, we say, ‘This worked a lot better than we thought it might,’ or, ‘This is a lot harder than we thought,’ and it takes more time.”
The particular development paths that they’ve mapped out to get from where they are to reach full human-level intelligence are incremental. They follow a hyper-specific roadmap for developing a human-level, commercially fit product.
“We've known for the last 20 years that we've been developing the technology and commercializing it, we've always known that, ultimately, the system has to be able to learn language by itself, with a teacher,” Voss said. “Our current focus is on replacing these human-coded rules with the knowledge that the system acquires.”
The way the system learns echoes the learning pattern of children to a point, meaning laying a stable base of fundamentals can lead to comprehensive understanding and self-directed learning if given enough hands-on attention.
With Cognitive AI, the future of customer service in Fintech and beyond looks more and more advanced in the way of communication and collaboration with technology.