• From education to hospitality, artificial intelligence will have the power to guide students and employees through everyday decisions where there is high risk of stress and failure.
• These interactions between humans and generative AI models can endure across years in the case of a degree, even decades in the case of a career, according to executives who spoke at a recent CNBC Disruptor 50 event in San Francisco.
• But current large language models are not up to this task, according to a top Scale AI technology officer, whose firm provides much of the data that feeds gen AI systems, including OpenAI's ChatGPT, the information they need to learn.

Last week, startup founders, executives and VCs gathered in San Francisco for CNBC's Disruptor 50 Connect event to hear directly from D50 companies with a focus on – what else – artificial intelligence.  

San Francisco was an obvious choice: "This is the epicenter, where all the work is happening," said Vijay Karunamurthy, Field CTO at Scale AI, which plays a key role in providing the data to AI companies whose large language models need to advance their exponential learning. "A couple of square blocks around our office in Mission, you have OpenAI, Google, Meta… pretty much everyone is flocking here. You can have a mind-meld with all of these people. You walk into a coffee shop and they all want to have these conversations." 

During a D50 Connect panel discussion led by CNBC Senior Media & Tech Correspondent Julia Boorstin —which also featured BeeHero founder & CEO Omer Davidi and Guild chief experience officer & head of platform Bijal Shah — the three top executives from companies that made the 2023 CNBC Disruptor 50 list delved deep into key AI topics. One of the key takeaways was about the transformation of individuals that they see taking place beneath the level of transformed industries — the workers, students and consumers, from hospitality to education and the grocery store. 

It's a point that's been made many times over already that AI will transform the economy on a level that should be compared to the industrial revolution, inevitably at the cost of some jobs — Karunamurthy said it's the biggest change since that era — but the D50 executives focused to a greater extent on how AI will transform individual experience in the job market, especially at key moments when the risk of making bad decisions is high. 

He pointed to the example of someone entering the workforce as a hotel concierge. 

"They can have an AI assistant who understands not just how to access data and sources relevant to their job, but an AI that can learn the history of the last 10-20 years of people in that role, that knows how to have kind and helpful customer interactions that are precise … booking a hotel room or travel experience," he said. "The ability of AI to help someone brand new to that role to understand the best case, and over their career playing out over 10-15 years, is really transformative." 

At Guild — which provides an education platform for employees at many Fortune 500 companies to earn degrees and certificates that help them to advance professionally — human connection remains core to the business model, said Shah, but it is very focused on how AI can improve the lives of the front-line workers at companies which use its services. 

"We're about to see a major disruption in how everyone works, and I do think that front-line population will be impacted most," she said. 

Karunamurthy cited a statement from Chegg CEO Dan Rosensweig, whose edtech company saw its shares crater when ChatGPT was released to the public and has not recovered since: 40% of students don't complete college, and surveying shows that this doesn't happen because the subjects are too hard for students but because everyday life gets in the way. 

That's one reason why Shah sees AI as a tool to travel alongside the learner in a way that goes well beyond helping students get answers to tests and write term papers. She said that the No. 1 predictor of success in a higher education program is the removal of life barriers. 

Child-care bills, not being able to pay rent or for a flat tire are all examples. "They all contribute to the inability to focus on something that feels like an investment in yourself, and once that happens … you lose confidence and then struggle to figure out the right next step forward." 

AI, in the same way it can assist the hotel concierge, may be in a position to point the struggling student to the information they need to not drop out, including connecting them to the experiences of other individuals who have been able to remove those barriers. 

Karunamurthy, whose firm now has a deal with Chegg, said education is a good example of how an industry and the individuals in it will be transformed, but not in the ways we've thought about so far.  

"Yes, the models can answer questions about biology and chemistry. Out-of-the-box models take information from textbooks and professors … but also from blogs and fifth-grader reports published online," he said. 

"The value is not just precise and accurate answers, it's understanding the student's journey and where they want to get to," he said. "A model that understands where you are in your journey and can understand those challenges and relate." 

An AI that knows you have an organic chemistry exam coming up in two weeks but also knows all the other important stuff going on in your life is a better AI than one that can spit out organic chemistry answers. "That will be really useful," he said. 

Before AI can be a greater part of individual life, the D50 executives said that they will need to solve for the risks of AI and the limitations of current models, especially related to inaccurate results, hate speech and misinformation, and just plain awkward (i.e. too personal) interactions with people. Models today aren't up to the job, at least not yet, at a level where they deserve our trust. 

"The models need to be kind and helpful and useful to human beings using it consistently, and right now those models often don't live up to that metric," Karunamurthy said. 

Davidi, whose BeeHero is working on technology that uses AI to prevent bee colony collapse, said his first challenge in the market is convincing farmers to use the technology to predict, and be early to detect, the signs of a bee die-off. That is no easy task in agriculture, which is old-fashioned and used to doing things in one way for a long time. "This is not an industry really open to innovations coming in that fast and trying to create a lot of waves," he said.  

He has learned in selling services to farmers that the first order of business is making sure potential clients don't see "the models taking over the people." 

"We want to make sure they still see the face of the person communicating with them, even if they use AI in order to make things more accessible that weren't accessible before," he said. 

If they can get the conversation right, the implications extend from hotel workers and college students to every consumer decision. Anyone who has used a beta version of Google's Bard in their Chrome browser might notice that in certain search results for shopping it may include "sustainability" in the top pros and cons (e.g. "not sustainably made") for various items. 

The idea was raised during the discussion that AI might need its version of "Intel Inside" as a marketing brand. To be successful, Davidi said his startup may need the accessibility of AI to extend to the consumer buying apples in a store — bee colony collapse is an issue that impacts 70%-75% of the world's food production

"How can we create an environment in which people want to buy 'bee-friendly' products," Davidi said. "We do know we need bees to pollinate those crops and unless we limit population, which is pretty tricky, we need to keep providing food, so we need to make sure it is done in a sustainable way. And if I can choose 'this' apple in a store, then those farmers will choose bee-friendly practices." 

At Guild, Shah said, the company is focused on "applications that allow us to scale the ability for human connection."  

Building that loop between AI and individuals and enterprises is clearly a priority, but the overriding order of business is to get the models to a stage of learning where they can successfully turn the concepts being discussed today into reality. AI is not there yet.

Arguably the most important comment of all came from the executive whose firm, Scale AI, is playing a major role in feeding the exponential learning of the LLMs and monitoring the imperfections up close.

"We will all benefit when we can trust the models," Karunamurthy said.