Navigating AI challenges in the enterprise [Q&A]

As more businesses turn to AI, they face a number of challenges around integrating it effectively and obtaining the best value while still ensuring that their data remains secure. It's also important that they select the right AI provider for their needs.

We spoke to Naren Narendran, chief scientist at database specialist Aerospike, to discuss the strategic considerations and concerns enterprises face as they incorporate AI into their operations.

BN: What's driving enterprises to turn to AI?

NN: Interest in AI has moved beyond being the latest trend to becoming a critical tool for innovation and efficiency. In 2023, AI emerged as the biggest tech topic, and now, in 2024, enterprises are getting serious about leveraging it to its fullest potential. They are increasingly turning to AI to harness its potential for enhancing operational efficiency, improving customer experiences, and driving innovation. The allure of AI lies in its ability to process vast amounts of data, identify patterns, and generate insights in real time that would be difficult to achieve through traditional methods. This transforms industries that rely on quick decision-making, such as finance and fraud detection. For example, Aerospike's database technology powers PayPal’s fraud detection systems, enabling split-second transaction approvals.

BN: How should businesses choose between general-purpose AI and more specialized models?

NN: The choice between general-purpose AI and specialized models depends on a business’s specific needs and objectives. General-purpose AI models, such as large language models (LLMs) with tens or hundreds of billions of parameters, are attractive because they can handle various situations seamlessly. They offer versatility but may lack the depth needed for specific tasks.

On the other hand, specialized AI models are designed to provide deeper insights within specific domains. These models can often be developed and managed in-house, offering tighter control and enhanced security. They consume less power and are more democratically accessible, making them a viable option for businesses with limited resources.

BN: What are the practical, financial, and other implications of this choice?

NN: Choosing between general-purpose AI and specialized models has significant practical and financial implications. General-purpose models require substantial investments in computer infrastructure and ongoing maintenance, making them costly. They are typically only accessible to large cloud providers or companies with deep resources. Additionally, relying on third-party services for these models can introduce security and privacy concerns, especially when handling sensitive data.

In contrast, specialized models offer a more cost-effective solution in the long run. Although there are some upfront training and model development costs, as well as periodic retraining expenses, these costs are lower than those associated with developing, maintaining, and running much larger general-purpose LLMs. These upfront costs can be amortized over time. Specialist models reduce the dependency on large cloud providers, resulting in more manageable operational costs. They allow businesses to maintain greater control over their data, enhancing security and reducing the risk of data breaches. Furthermore, developing in-house models can lead to better alignment with the business's needs, leading to more accurate and relevant insights.

BN: How can enterprises ensure their data remains safe when dealing with external AI providers?

NN: Data security is paramount when dealing with external AI providers. To ensure data remains safe, enterprises should adopt a multifaceted approach. First, it’s crucial to vet AI providers carefully, evaluating their track record, financial stability, and the quality of their offerings. With so many companies jumping on the AI bandwagon, assessing how long these providers have been around, the kind of backing they have, and their likelihood of being around in the future is essential. Properly evaluating these parties is a difficult but necessary task to ensure they meet the required standards and can be trusted with sensitive data.

Additionally, businesses must tread carefully and thoughtfully. Security concerns, such as data breaches or inappropriate use of data, can have significant financial and brand implications. The relatively new nature of this technology means that small mistakes can quickly escalate due to the scale of AI applications.

Another point to consider is whether the AI provider plans to incorporate a business’s private data to train its general-purpose public models. Understanding and controlling how data is used can prevent unintended exposure.

Furthermore, businesses should implement a phased approach to AI adoption, starting with projects that can be carefully monitored for potential problems. By watching for warning signals related to security or other issues, enterprises can incrementally expand their AI use as new programs prove their worth. Regular audits and monitoring of AI systems are essential to promptly detect and respond to potential security breaches.

BN: Which factors are most important when selecting an AI provider?

NN: Selecting the right AI provider is critical to successfully integrating AI into business operations. Several factors should be considered in this decision-making process. First, the provider's expertise and experience in the industry are crucial. Providers with a deep understanding of the industry’s specific challenges and requirements are more likely to deliver effective solutions.

Second, the quality and reliability of the AI models offered by the provider should be assessed. This includes evaluating the models' performance, scalability, and flexibility to ensure they can meet the evolving needs of the business.

Finally, the provider's commitment to innovation and continuous improvement is important to ensure that the AI solutions remain cutting-edge and effective.

Image credit: peshkova/depositphotos.com