Angel provides driver and customer support. Naomi teaches high school students. But they are not humans. They’re HumAIns, artificial intelligence (AI) agents.
Designed to handle unscripted communication tasks autonomously and proactively, Angel and Naomi use human-like thought processes to lead conversations instead of merely answering questions. And they get smarter with time and experience.
This generative AI wizardry comes from Jerusalem-based Inpris Innovative Products, founded in 2011 by human-machine interaction expert Nissan Yaron and his father, software architect Ben-Etzion Yaron.
Inpris recently won a Project Voice award for conversational AI from Project Voice Capital Partners, cofounded by Siri inventor Adam Cheyer. At last year’s Microsoft Ignite conference, Inpris was featured by Microsoft’s Satya Nadella and OpenAI’s Sam Altman as pioneers in the field.
In fact, these tailored virtual assistants surpass Siri and ChatGPT, the company claims.
“HumAIns must understand their employer, role, task, available information, and potential actions to accomplish their objective,” says Nissan Yaron.
“Using voice, chat and avatars, our agents can express empathy, joy, sadness and other emotions by changing their voices in real time, creating a more human-like interaction experience. They can perform real-world actions simultaneously while conversing with users.”
HumAIns also provide insights from the conversations, such as sentiment analysis and cooperation levels, via a simple API (application programming interface) and dashboard.
Mimic thought and actions
And unlike humans, HumAIns cannot be distracted from their defined task.
For example, Naomi, a one-on-one teaching HumAIn successfully piloted in the AMIT schools network in Israel, steers the conversation back if the student strays off topic.
This is possible because Inpris devised a “cognitive architecture” for HumAIns, facilitating a multistage thought process.
The process is powered by a combination of technologies, including natural language generation (NLG) and task-based natural language understanding (NLU), orchestrated by additional algorithms and AI models.
This unique architecture, says Yaron, enables HumAIns to mimic human thinking, natural conversations and independent task initiation.
“HumAIns can perform tasks previously only performed by humans, resulting in high-quality work at a fraction of the cost.”
HumAIns are capable of expressing emotions as they interact with customers. Photo courtesy of Inpris
The purpose isn’t to render human work obsolete, “but rather to allow human employees to focus on more critical tasks, provide support for complex cases, and address situations that require research,” Yaron said.
“As a result, HumAIns offers immediate relief for many cases with zero waiting times and enables businesses to scale their outreach capabilities where human labor is simply not feasible.”
Angel in the car
Inpris is collaborating with approximately 10 enterprises, including four large automakers.
As an intelligent driving companion, Angel can help drivers understand, operate and troubleshoot the car’s functions and control third-party systems such as phones and infotainment.
Helping drivers perform in-car tasks without taking their eyes off the road has obvious safety advantages, which is why Inpris is being courted by the automotive industry.
“HumAIns can perform tasks previously only performed by humans, resulting in high-quality work at a fraction of the cost.”
It wasn’t always this way.
ISRAEL21c first encountered Inpris at the Google TLV campus in 2013, when it was developing touchscreen technology for the visually impaired.
The company then built an award-winning prototype of a driver assistance system based on human-machine interaction.
Realizing how the nascent generative AI revolution could improve this invention, Inpris was among the first companies to access OpenAI’s GPT-3 (Generative Pre-trained Transformer) released to a small group of beta testers in early 2020.
“We played with GPT-3 and realized its potential in producing humanlike conversations. We trained it to operate our in-car assistance system so that you could have an assistant you could speak to naturally, that could navigate, send messages, call people for you, even draft emails,” says Yaron.
“With Siri, you have to provide the wording for a message. With our assistant, I can say ‘I think I’ll be late to the meeting so please draft a nice apology and send it to Abigail.’ It can do this by itself. It opens my contact list, opens WhatsApp, enters the message and sends it, completely autonomously.”
And then came ChatGPT
Inpris had 1,000 drivers testing the system about two years ago. But it was ahead of its time until OpenAI’s game-changing ChatGPT came along at the end of 2022.
“When ChatGPT exploded, our traction exploded,” says Yaron.
“Before, it was difficult to explain to companies about natural conversation beyond Siri. In the last six months, we see a tremendous shift as people become aware of the capabilities of generative AI but also its limitations – it’s difficult to control, it can provide text that nobody asked it to, it can hallucinate, it is impossible to connect to APIs,” he explained.
“We had already figured it out. We had a product that could speak naturally, connect to APIs, and is controllable and reliable. So now, thank God, the companies we were chasing then are now chasing us.”
At EcoMotion Week in Tel Aviv last May, Inpris presented its in-car assistant alongside Hyundai, with which it has two ongoing projects.
Angel outside the car
Angel’s fame has spread, leading Inpris to other verticals.
The Israel Electric Corporation came looking for a customer support agent that could speak naturally with customers and take real-world actions in strict alignment with IEC protocols – in Hebrew, which was challenging as the first models were trained in English.
“Big enterprises don’t have the personnel to speak with everyone who is not paying their bills. If they can automate an agent that can engage in a meaningful conversation with thousands of people in a very short time, it can try to solve those issues and charge the customer’s credit card an agreed-upon amount,” says Yaron.
Inpris also is developing an autonomous agent to explain insurance policies to customers for an Israeli company following a successful proof-of-concept (POC) project. Other POCs are underway outside of Israel as well.
Naomi in the classroom
Naomi, the virtual enrichment tutor at an AMIT high school, is really the same thing as Angel engaging drivers and instructing them how to operate the functions in their vehicles, Yaron says.
In both cases, the agent understands the task, knows the information and discusses it with the user dynamically.
“We did one class as a POC and in the coming year it will scale throughout the AMIT network. We’re building them a tool so they can feed more lessons into the agent and support more one-on-one enrichment classes.”
Additional customers are interested in autonomous employees to handle a range of customer services.
“We’ve always been looking ahead and now we are finally aligned with the market,” says Yaron.
Inpris is raising a Series A to expand its staff. Until now, it has been financed by angel investors, awards, grants and, more recently, revenues.
“Our vision involves training HumAIns in natural conversation, adapting based on feedback, and collaborating as team players alongside humans. They will possess reinforcement learning abilities, autonomously analyze data, and continuously improve performance. This ensures they remain effective and relevant in a dynamic market.”
Produced in association with ISRAEL21c