Overcoming AI’s 5 Biggest Roadblocks
Nobody told you that artificial intelligence (AI) development would be easy. Here’s a look at how to surmount AI's biggest obstacles.
![Businessman drives giant robot cartoon Businessman drives giant robot cartoon](https://eu-images.contentstack.com/v3/assets/blt69509c9116440be8/blt5f1be6856d6286f1/65f1db539a81d4040a027cd6/AI_Silly-nuvolanevicata_-alamy.jpg?width=700&auto=webp&quality=80&disable=upscale)
nuvolanevicata via Alamy Stock
AI is a powerful technology with the potential to transform both industry and customer experiences. Yet there are risks that come with this transformative innovation, says Mia Mayer, a machine learning university instructor and applied scientist with Amazon Web Services.
One such risk is bias, which in AI can lead to unintended consequences or disproportionately harm underrepresented or marginalized groups. “For instance, an AI system that assists with the evaluation of loan applications could deny some people from accessing business loans, widening the economic divide or racial wealth gap,” Mayer says. Meanwhile, AI used to predict health outcomes could generate one-sided results or misdiagnoses. “That’s why it’s an imperative to mitigate bias throughout the entire AI lifecycle.”
To integrate fairness into the AI lifecycle it’s necessary to look at where things can go wrong. Bias can enter AI systems at different stages of the AI development cycle, from data curation to model training to deployment, Mayer explains. “AI systems learn from the data provided by humans, and people aren’t perfect,” she notes. “They may not collect or curate data appropriately or use data and models that are historically biased.”
When it comes to model selection and training, certain trade-offs should be considered, Mayer advises. “For example, some models may perform well across the general population, but may not work well for minorities, either due to algorithmic oversimplifications or misrepresentations in the data.”
Evolving technology and regulatory landscapes are making it difficult for organizations to establish a strong AI governance framework, claims Jim Rowan, AI market activation leader with Deloitte Consulting. He notes that AI solutions and rules regarding their use are constantly shifting. “While regulatory requirements often lag behind the pace of technology innovation, recent developments such as the US presidential AI Executive Order and the European Union’s ambitious AI Act signal a growing recognition among global government leaders of the importance of addressing AI-related risks.”
Deloitte’s recent State of Generative AI in the Enterprise report found that 41% of leaders said their organizations were only slightly or not at all prepared to address governance and risk concerns related to generative AI adoption. Of the surveyed business and technology leaders, the biggest concerns related to governance were a lack of confidence in results (36%), intellectual property issues (35%), misuse of client or customer data (34%), the ability to comply with regulations (33%) and a lack of explainability/transparency (31%).
Cost consolidation and the need to unify workflows and platforms is a top concern for IT leaders, says Mahesh Ram, Zoom’s head of AI. The traditional approach of using multiple vendors for various aspects of AI workflows has led to inefficiencies, including spend duplication, poor points of integration, and high total cost of ownership (TCO), all of which are particularly tricky for enterprises navigating economic headwinds, he notes.
The challenge facing IT leaders, Ram says, is ensuring that AI is accessible to both small- to medium-sized businesses as well as to enterprise buyers. “Cost savings are a crucial part of providing AI equitably to all knowledge workers, especially as it becomes an even more integral part of the modern workforce,” he explains. “The industry’s goal should be to ensure it’s not restricted to only a select few, and instead turn it from an exclusive club into an essential tool available to a wider audience.”
Yet simply lowering costs won’t be enough, Ram warns. Providers must be able to demonstrate AI’s value in quantifiable terms, he states. “This will come from true user engagement, adoption, and proof points on better work output, time saved, and other metrics.”
Ram also believes that overcoming the AI cost roadblock will require enterprises to consolidate and unify their IT investments. “Doing so will not only help companies save money but will also help supercharge adoption and the return on their AI investment.”
The biggest roadblock to both adopting and leveraging AI is a lack of knowledge, skills, and expertise, claims Marinela Profi, AI strategy advisor at AI and analytics platform provider SAS. “The vast majority of people don’t understand AI and have no idea how to put it to work.”
Technology is advancing at a pace that’s outstripping the current educational infrastructure’s ability and capacity to train students, Profi states. “The gap is widening between cutting-edge AI developments and the skills taught in traditional educational settings,” she adds. “That knowledge and talent gap makes it hard for the workforce to keep up with demand.”
There’s often a disconnect between what’s taught in school and the practical AI skills needed in the workforce, Profi says. “This gap means that even graduates in relevant fields may not be ready to tackle real-world AI challenges without additional training.”
The biggest single roadblock to AI adoption isn’t new -- it’s data, meaning access to high-quality data, says Olga Kupriyanova, a principal consultant with technology research and advisory firm ISG. A related hurdle, she notes, is getting data ready for AI to consume on architectures with appropriate compute resources.
“For many organizations, data management is a legacy challenge that’s been hard to solve for a long time,” Kupriyanova says. “Generative AI is offering some new ways of applying AI to actually resolve data challenges, yet there are very few shortcuts that organizations can take,” she adds.
The biggest single roadblock to AI adoption isn’t new -- it’s data, meaning access to high-quality data, says Olga Kupriyanova, a principal consultant with technology research and advisory firm ISG. A related hurdle, she notes, is getting data ready for AI to consume on architectures with appropriate compute resources.
“For many organizations, data management is a legacy challenge that’s been hard to solve for a long time,” Kupriyanova says. “Generative AI is offering some new ways of applying AI to actually resolve data challenges, yet there are very few shortcuts that organizations can take,” she adds.
Artificial intelligence is a powerful tool, yet harnessing its inherent capabilities can be a challenging and, in some cases, overwhelming task. Still, there are several best practices that can help newcomers overcome many of AI’s biggest roadblocks.
Here are five expert views, shared via email interviews, on today’s top AI roadblocks and how to successfully address them.
About the Author(s)
You May Also Like