In our upcoming EcoReporter segment on ‘Pentagon Urges AI Companies to Share More About Their Technology’, we delve into the need for transparency in the development and implementation of artificial intelligence (AI) tools. Craig Martell, the Pentagon’s chief digital and artificial intelligence officer, has called on AI companies to share insights into how their software is built, without forfeiting their intellectual property rights. Martell emphasizes that the Pentagon needs a better understanding of AI models’ structure, data sources, and potential dangers in order to confidently adopt the technology.
AI software relies on large language models (LLMs), which use extensive data sets to power applications like chatbots and image generators. However, these services are often offered as black boxes, with little information provided about their inner workings. This lack of transparency makes it difficult for users, including the Pentagon, to comprehend how the technology makes decisions or improves over time.
Martell expresses concern that companies developing LLMs are not disclosing the biases or limitations of their systems. He likens these models to “found alien technology” for the Defense Department, as they are complex and unfamiliar. Additionally, Martell raises the issue that only a few groups have the financial resources to develop LLMs, potentially creating an imbalance of power.
To address these concerns, Martell plans to convene industry and academic experts in Washington in February. The Pentagon’s symposium on defense data and AI aims to determine suitable applications for LLMs and establish benchmarks for assessing their performance. Martell’s team has already identified 200 potential uses for LLMs within the Defense Department.
While Martell acknowledges the potential benefits of LLMs, he stresses the need to understand their use, benefits, dangers, and how to mitigate risks. He highlights the importance of responsible development and usage, particularly in situations where lives are at stake. Martell hopes the symposium will help establish a “maturity model” that sets standards for addressing hallucination, bias, and danger in AI systems.
The Pentagon’s push for transparency and accountability in AI development and deployment reflects the growing recognition of the need to ensure ethical and responsible use of emerging technologies. By shedding light on the inner workings of AI systems, companies can build trust and enable informed decision-making in adopting these technologies.
