How Uniphore is expanding control and reducing hallucinations within its AI models

The enterprise AI company is enacting a narrowly focused approach to exercise greater control over its technology.

Nov 25, 2023 - 19:30

0 22

How Uniphore is expanding control and reducing hallucinations within its AI models

The year since ChatGPT's monumental launch has been marked with a series of milestones as artificial intelligence has become more normalized.

Safety concerns — focused on active harms of disinformation, algorithmic discrimination, political instability, environmental impact and hallucination — have featured in the public discourse. Regulation has started to shape up, with President Joe Biden in October announcing an executive order on AI and the European Union nearing the completion of its landmark AI Act.

Despite these positive trends, Biden's executive order seems to lack teeth and the EU's Act may be in the process of being watered down after pushback from the French, German, and Italian governments.

And hype around AI — whether extremely positive or world-endingly negative — has not yet dwindled.

In lieu of firm regulation, several tech startups and corporations have been exploring internal AI responsibility and safety measures.

The bulk of American voters, according to polling by the Artificial Intelligence Policy Institute (AIPI), however, do not trust tech executives to self-regulate when it comes to AI.

And Uniphore — an enterprise AI company that was founded in 2008 — has been focused on responsible AI since the beginning.

Uniphore's narrow-focus models

CEO Umesh Sachdev founded Uniphore 16 years ago as a conversational AI company with the initial purpose of ensuring that everyone, regardless of their native language, could participate in the predominantly English digital revolution.

Uniphore has since become a multi-modal enterprise AI company that services more than 1,500 enterprises around the world. The 750,000 users within those companies, according to Sachdev, use Uniphore's models daily in work ranging from sales to call centers to health and financial services.

As of February 2022, Uniphore had achieved a valuation of $2.5 billion.

Even in the early days of the startup, Sachdev told TheStreet in an interview, Uniphore was intent on building out its technology responsibly.

"We had empowered a group of people who understood the topic to build policy and guardrails for us internally," he said.

One of the prominent results of its work in this area has been an effort to reduce AI bias and discrimination by training its models on diverse data sets.

Reining AI in

The other offshoot of its responsible intention, Sachdev said, is that Uniphore creates hyper-specific models.

When building a model, Uniphore looks first at the kind of company a client is, the use case they're looking for and the language they require. It then offers the most focused solution.

Sachdev said Uniphore doesn't use stolen data to train the models it is building.

"Unlike OpenAI, we're not scraping the internet to build our models," Sachdev said.

Uniphore instead uses exclusive, anonymized first-party enterprise data to develop models for each client.

"When I meet a prospective customer, I'm not saying 'listen, this is a very powerful generative model, let's find applications in your company,'" Sachdev said. "I'm saying 'I have an AI model pre-trained on 15 specific use cases, which hundreds of companies like you have already applied AI for.'"

Because of the specificity of this approach, Sachdev said, "I'm also narrowing the problem statement for that AI model."

"By narrowing the scope of each of those applications in the enterprise, we are able to better control and put more guardrails around that aspect," he said.

This reduces the risk of hallucination and increases user efficiency.

AI and the labor market: 'The more things change, the more they remain the same'

Sam Altman, the co-founder and CEO of OpenAI, said in March that he believes AI systems will "make a lot of jobs just go away."

"Customer service is a category I could see there are just way fewer jobs relatively soon," he said. "I'm not certain about that, but I could believe it."

Sachdev made a similar prediction nearly a decade ago.

"Guess what? Those call center workers are still around, dramatically more efficient because of AI already, but they are still around," he said.

Likewise, he said that AI coding copilots won't replace engineers.

"Engineers love to write code. They're designers. They are architects. But there's also some grunt work in coding much like any other job that they hate to do," Sachdev said. "And AI, fortunately, is coming out and taking away those mundane tasks."

"AI is a great partner and ally," he added.

The proliferation and mass adoption of AI, Sachdev said, will lead to enormous productivity enhancements across the board. He doesn't think mass layoffs will happen, but rather that the pace of headcount expansion will slow down across several industries.

Headcount growth in the AI sector, however, will continue to grow, according to Sachdev.

"How many employees does Tesla have? How many employees does OpenAI have? Those efficiency gains are being delivered by AI companies dramatically increasing their workforces," he said.

He added that the necessary job growth within the industry impacts far more people than machine learning engineers.

The key right now, he said, is that the next wave of the workforce must be educated on how to use AI.

"Existing jobs are getting impacted, make no mistake; there's no reason to fight. It's good for humanity," Sachdev said. "But net new jobs and new opportunities are created."

"This playbook happened in mobile, happened in the internet, happened in the industrial revolution," he added. "It's the same cycle repeating itself, the only difference is it's happening much faster this time."

Dr. Srinivas Mukkamala, however, an AI authority who serves as Ivanti's CPO, told TheStreet in July that the technology will displace workers in an exponentially larger way than these previous revolutions. The result, he said, will likely be an even wider gap between skilled and unskilled workers and countries that could leave "99% of the world's population behind."

"The only thing I'm concerned with is the haves and the have-nots. This is going to create inequality that we've never seen in our lifetimes," he said at the time.

The AIPI found recently that more than 25% of jobs across America have significant exposure to AI automation.

AI expert Gary Marcus said in May that when it comes to jobs and AI, "history is no guarantee of the future."

Contact Ian with tips via email, [email protected], or Signal 732-804-1223.

Get investment guidance from trusted portfolio managers without the management fees. Sign up for Action Alerts PLUS now.