Businesses looking to take advantage of the latest artificial intelligence tools are in search of any number of potential benefits, including automating repetitive tasks, enhanced data analytics, reduced human error and better and faster decision making.
The problem is, barriers to AI adoption abound, keeping companies from deploying the technology as fast as they'd like.
In fact, a survey of 120 U.S. senior AI/machine learning decision makers, conducted in late 2023 by research and media firm Foundry and technology consulting firm Searce, showed that less than 40% of organizations have successfully deployed an AI project.
Among the biggest barriers to adoption: the rise in cybersecurity threats. The Foundry/Searce study showed that 58% of respondents said data security is a leading barrier to AI adoption.
There is a lack of understanding about the security vulnerabilities of AI applications, said Jake Williams, a faculty member at cybersecurity research firm IANS Research.
"AI apps, especially those using large language models, bring in an entirely new set of vulnerabilities that are poorly understood by most application developers and security testers," Williams said. "Until there's better understanding of these issues, and better tools to help with auditing and defense, some CISOs are warning that additional risks might not be warranted" by launching AI projects.
The most productive step companies can take is to get educated about how AI works, Williams said. "In the coming years, I believe we'll see dedicated security training and certifications for AI," he said. "But in the meantime, that has to be brought in-house by data scientists and those trained in security and AI. Tooling in this area is very immature, so organizations will need to primarily focus on processes to properly threat model their applications using AI to look for those unique risks."
AI return on investment
Another barrier is unclear use cases for AI. Many businesses are not thinking about which organizational use cases will bring them the biggest return on investment, said Vrinda Khurjekar, senior director at Searce.
"Lack of prioritization of a well-qualified use case is the number one cause of poor adoption of AI," Khurjekar said. "If you pick a use case that is too impactful, you risk any failures creating doubts across the organization. Alternatively, if you pick a use case that has extremely minimal impact, it fails to get any momentum from the rest of the organization."
Finding the right balance of both complexity and impact is critical in how AI will be adopted across the organization, he said.
One way to address this is to create an AI council. "Having a focused approach on which use cases to handle first will go a long way in accelerating AI adoption," Khurjeker said. "And you cannot achieve this unless you thoroughly look at your entire organization, understand where AI can have a maximum impact and then prioritize which needs to tackle first."
To do this more efficiently, companies should consider having representatives from the entire organization be part of the council, Khurjeker said.
Many organizations want to use AI in applications, but don't know where they can get value from it, Williams said. "Today, we're seeing a bit of a gold rush feeling of 'don't be left behind,' without any real thought as to applicability to a specific use case," he said. "This reminds me a lot of the early blockchain days."
A lot of companies are also grappling with the lack of talent in the AI area, which can create another barrier to adoption.
"With changes in technology happening so rapidly, organizations are struggling to attract and retain top talent," Khurjeker said. Without the right talent, teams struggle to launch AI initiatives — or worse have faulty launches — causing doubts in the minds of the broader organization, he said.
Investing ahead of time in hiring talent as well as implementing training programs that allow existing employees to become more proficient in AI will help strengthen the talent pipeline, Khurjeker said.
Yet another barrier is low model maturity, which can cause "hallucinations" or events in which AI models generate false information that isn't based on real data or events.
"AI models, specifically [generative] AI ones, are still early in their lifecycle," Khurjekar said. "Hallucinations in the outputs are real and for industries where accuracy is super important, such as health care and financial services, this is causing early adopters to proceed with a lot of caution."
Until the models become more mature, Khurjekar said, this will be a real challenge for companies wanting to move forward rapidly in the adoption of AI tools.
Finally, AI adoption might be slowed by regulatory policies and compliance efforts. "With AI adoption being in its early days, regulators are still evaluating its implications," Khurjekar said. "Most government and regulatory bodies are still in the early days of formulating the guardrails that will define how AI is more broadly adopted across companies."
Given this uncertainty, "Quite a few companies in highly regulated industries want to wait and see where the regulators end up," Khurjekar said. "This is slowing down adoption, as companies don't want to implement something and then have to unwind it if there are major regulatory changes around AI policies."
Businesses need to stay current with regard to AI developments. "AI adoption is not a one-time event that companies need to plan for," Khurjekar said. "It is more of an ongoing mindset shift where we look at all processes with an AI-first lens. Staying up to date with all the latest advancements is crucial in the success of an AI adoption journey."
