Disaster Recovery Journal Summer 2024

amount and quality of data used to train the AI tool. This explains why diagnosing hardware failures and ransomware detection represent AI’s first use cases. Providers often have access to large amounts of quality data in those two areas to train their AI tool. Validate how the provider gathers data to train its AI tool and how it sources that data. Also determine if the provider will gather data from your site, what data it will gather, and how it will use it. If a provider cannot clearly answer these questions, again view the quality of the AI tool with suspicion. represents the next major wave of IT innovation. Once organizations get a taste of it, they will like it, perhaps a lot. This will in turn lead to organizations wanting to consume as much AI as possible. However, providers of AI tools often use a consumption model to charge for their AI services. The more questions organizations ask of AI, the more the provider charges for the answers AI provides. This should prompt every organization to minimally assess how to monitor and control AI’s costs. Organizations may even want to ask the AI tool how much it charges for the answers it provides.

A few providers have also started to leverage this machine data to identify potential ran somware attacks or unauthor ized data access. In this case, analyzing the machine data may indicate unusual or high levels of activity on specific network ports. Flagging this activity may indicate a hacker attempting to change backups or copy them offsite. Chatting with Your Backups The larger the organiza tion, the more backups it typically stores and retains. Organizations often store these backups long term for compli ance reasons, disaster recov ery purposes, or both. While backups satisfy one or both use cases, the backups themselves typically provide nominal near-term value. One provider seeks to change that by mining the data in these archived backups. Its AI technology positions orga nizations to query and chat with their backup stores to obtain needed or desired information. This AI tool gets trained by first getting access to the archived organizational back ups. These backups serve as the large language models (LLM) the AI tool needs to understand the data. The AI tool then reads the backup data and indexes it to document the information contained in it. Once it completes this training, organizations may then “chat” with the AI inter face and ask it questions. The AI tool then formulates its answers based upon the data contained in the backups.

Validating the Promise of AI These existing implementa tions of AI hold great promise for organizations to help them better manage and optimize data protection. However, any AI tool deployed will minimally require significant amounts of computing and storage resources to operate well. Further, organizations may still need to dedicate resources to optimize the AI tool for its purposes. To ensure AI delivers on its promised functionality, orga nizations should minimally verify it possesses the follow ing features: n Cloud architecture . Any AI tool requires large amounts of computing and storage resources to store and process the data it needs to function. These AI demands likely outstrip the resources available on a single server or storage system. Meeting them typically requires a cloud-like architecture that both cloud hyperscalers or hyperconverged infrastructures (HCI) offer. Both architectures position organizations to quickly, and independently, add more computing, storage, or both and then utilize them. If an AI tool does not support one or both architectures, approach the AI tool with caution. n A large pool of quality data used to train the AI tool . Every organization wants an AI tool to provide it with valuable, actionable insights. However, the quality of those AI insights correlates to the

AI’s Promise Coming Closer to Reality The various implementa tions of AI in today’s data pro tection solutions hold a great deal of promise for organiza tions. It can help them in the near term proactively identify and resolve hardware issues in their backup targets. It can con tribute to identifying instances of ransomware and unauthor ized user access before these issues escalate. In some cases, it can even help organizations get better insights from the archived backups they possess. However, organizations should keep firmly in mind that AI remains in its early stages. The specific use cases mentioned here represent the few areas where organizations may confidently use AI in a turnkey fashion. Even then, it will come at a cost compared to competitive solutions. If hoping to use AI to address any other needs, organizations must proceed thoughtfully. They should prepare to make significant investments in money, time, and personnel if they want AI to deliver on any other internal objectives. While off-the-shelf AI solutions may soon solve other data protection issues organizations face, it cannot do so yet today. v

n A way to monitor and control costs . AI likely

Jerome Wendt, an AWS Certified Solutions Architect, is the president and founder of DCIG, LLC., a technology analyst firm. DCIG, LLC.,

focuses on providing competitive intel ligence for the enterprise data protection, data storage, disaster recovery, and cloud technology markets.

12 DISASTER RECOVERY JOURNAL | SUMMER 2024