The work to make AI accessible to the masses may be well underway, but companies are nowhere near the finishing line.
Microsoft, Google, Meta and several other big tech companies are aggressively pushing AI solutions, embedding them across the product lines and making the technology readily usable for the users. But they keep tripping up on some key obstacles.
Public trust in the technology has been on the decline, with trust dropping 15 percentage points within the US in 2024, according to a report. There are specific sectors within which trust is typically low, but even in industries where the technology is well-established, studies show that it is losing leads in most countries.
This begs the question – can the benefits of AI be truly democratized? “Not as we are doing it today,” says Kurtis Kemple, sr. director of developer relations at Slack.
While delivering a talk on the mass adoption of AI at the Ignite Talks at AI Data Infrastructure Field Day event in October, Kemple noted that the problem is, “AI is overly complex for everyday users.”
Besides having glaring security holes, AI needs deep technical expertise making it a technology for a chosen group of people. Only those that are adept at it have a chance at using it to its full potential.
“The reality is that if I take the whole internet and put that into something and ask it a question, I’m unlikely to get a novel answer. At best I could get something that’s pretty good – but there’s probably going to be garbage included in there as well,” Kemple remarked.
Getting a useful answer out of an AI chatbot largely comes down to how effective the prompts are. Slightly different prompts can produce vastly different responses.
But while users’ prompting capabilities remain an unwrit prerequisite for using any AI chatbot, the art of framing powerful prompts remain elusive to the public.
Kemple, who came to tech a bit later than most, while serving time in prison where he took his first courses on the subject, believes that it will take some radical changes to enable people from all walks of life with AI.
He highlighted that the blank canvas – you type in a question and get a response back – which makes AI chatbots look so ridiculously simple, is what makes them so hard for new users to get started with. The lack of guidance and care and feeding often has a paralyzing effect on first-time users.
“Feels like there’s a very big disconnect between the type of experiences users expect daily and what we’re trying to feed them with AI,” Kemple noted.
To remove this problem, Kemple argued that AI should be made simpler. “When AI is requiring you to have specialized skills to adopt it, you’re immediately cutting off how much of the market you can reach,” he says.
He distils it down to a few core principles, and insists companies dabbling in AI to focus on them starting with improving accessibility and building trust.
“Models and AI agents need to be highly accessible to end users in a user interface that is intuitive and feels good to them, and then on the very other end, creation and fine-tuning of models has to push further and further into the no-code space,” he said.
Kemple proposed that the way around gatekeeping is to have an open and connected AI model marketplace that serves as an inventory for all free and paid models – much like Hugging Face – where users can also find no-code model-tuning tools and educational resources. This will serve as the “collective intelligence” hub for both creators and users.
Providing revenue-sharing opportunities on top of this, Kemple said, will encourage more people to invest in the technology.
Taming the technical perplexity, Kemple says, cannot be done without intelligent orchestration. The most personalized experience comes from context-aware interactions, and so a powerful AI agent must infer and analyze the content and intent of all prompts, and route them to the orchestration layer where the most coherent responses will be developed.
“So now I can pull a phone out of my pocket, ask a question about starting a restaurant, and I might get a response about menu design from a chef, or about how to look for locations, deals and loans from somebody like Mark Cuban, if they decide to make a model.”
Kemple favors open-source foundations for effective trust-building. When topped with data transparency and reliance on industry experts for training, he believes, it can lead to fully trustworthy models.
Cherry on top, having a flexible UI that offers users a way to curate and adjust the interface to best fit their needs, preferences and behavior, while offloading their cognitive load and skill requirements, Kemple said, can make AI much more reachable and usable in the future.
