Managing Unexpected Expenses
The rapid expansion of generative AI capabilities, sparked by the popularity of ChatGPT, has led to significant and unanticipated costs for various Army organizations, as revealed by the service’s chief information officer, Leonel Garciga. In response, the Army is prioritizing control over generative AI expenses, with new best practices and policies expected by April.
Deliberate Approach to Implementation
Garciga emphasized the importance of a measured strategy in adopting generative AI technologies. He cautioned against widespread, indiscriminate use, remarking, “Let’s not have a bazillion flowers blooming in this space.” This proactive approach aims to avoid escalating costs and challenges in data protection.
In November, Garciga merged two separate AI pilot projects, #CalibrateAI and Athena, to streamline efforts in analyzing the outcomes and preparing comprehensive guidance for the Army. This guidance will encompass lessons learned, best practices, and specific mandates, without the introduction of additional funding.
Guidance and Budget Autonomy
The upcoming guidance package will be released alongside the conclusion of the Athena pilot in April. Garciga stated, “We’re hyperfocused on how we put out guidance across the force to ensure we’re buying the right, secure items.” Instead of a centralized funding strategy, organizations within the Army will use their own budgets to select generative AI tools, encouraging sensible expenditure practices.
Promising Applications Beyond Costs
Despite the financial challenges, Garciga and Deputy Assistant Secretary Jennifer Swanson highlighted positive outcomes from the Army’s generative AI experiments. These initiatives have revealed numerous commercially available tools and strong interest from various Army organizations. Potential applications include aiding acquisition officers with vendor proposals, helping locate relevant regulations for counsel, and assisting public affairs officers in managing social media.
While there is potential for generative AI to assist in software coding, Swanson cautioned that it is not yet poised to replace human programmers. The majority of promising applications currently focus on “back office” processes that share similarities with private sector operations.
Understanding Military Contexts
Garciga noted that the applicability of generative AI tools depends on the specific challenges faced by the military. He shared an example where the Army sought aviation safety software for helicopters but encountered irrelevant responses from civilian aviation data. This underscores the necessity of understanding the specific data sets involved in generative AI projects.
Cost Management Strategies
Another critical lesson learned is that different generative AI products can vary in their adherence to cost control practices. While some free versions exist, many generative AI tools require expensive high-performance GPUs. Contracts with vendors can sometimes lead to immediate billing for queries, more effectively managing costs. Garciga noted that cloud service providers are generally proficient in implementing safeguards to prevent budget overruns.
However, traditional cloud computing contracts can complicate expenditure tracking, with bills arriving at the end of the month, potentially leading to budget surprises. He humorously remarked about the need for conversations if a substantial portion of the budget is exhausted early in the fiscal quarter.
Overall, the Army’s focus is directed toward leveraging generative AI while ensuring fiscal responsibility and understanding the unique requirements of military applications.
