Вештачка интелигенција

The Future of Generative AI: Overcoming Ethical and Computational Challenges

Summary

Generative AI has garnered significant attention from both governments and the private sector. Cutting-edge generative AI platforms like ChatGPT-4 have the ability not only to analyze existing information to solve problems or provide analysis but also to generate new information, […]

The Future of Generative AI: Overcoming Ethical and Computational Challenges

Generative AI has garnered significant attention from both governments and the private sector. Cutting-edge generative AI platforms like ChatGPT-4 have the ability not only to analyze existing information to solve problems or provide analysis but also to generate new information, much like the human brain. However, there are potential challenges on this path, particularly when it comes to implementing generative AI in government settings. The two main concerns are the ability to use this new technology ethically and the immense computational power required to support the development and maintenance of generative AI programs.

In terms of ethical considerations, governments worldwide are now attempting to regulate this technology or establish ethical guidelines for its use. Concerns over generative AI have even prompted the White House to issue an executive order outlining how agencies should – and should not – utilize this new technology.

According to statements from federal and private sector officials at a meeting, they are all still extremely excited about what generative AI can do in public service and are confident in their ability to use it ethically while fully complying with the executive order.

On the practical side, there is enormous potential for the use of generative AI, with USAID funding research and programs related to gen AI worldwide. In one example, portable X-ray machines equipped with AI capabilities funded by USAID have been implanted to improve tuberculosis detection.

Further use of generative AI could enable predictions such as disease outbreaks or pandemics, allowing USAID to proactively secure resources for assistance instead of reacting to each new crisis as it emerges.

While there are ethical challenges, computational power also poses a hurdle for implementing generative AI in government services. Even training generative AI on large language models that it requires to influence its decisions can take months using standard computer equipment, even if entire data centers are dedicated to the task. Significant computational power is also necessary for new generative AI models to quickly process user commands and requests.

To meet the need for even more powerful hardware, NVIDIA has just announced a brand-new AI-dedicated chip, the H200. The new H200 chips are designed to reduce generative AI processing time by 50% or more, while also decreasing energy consumption.

The H200 design is similar to the H100 chip but significantly enhances memory. Each H200 chip can support a memory bandwidth of 4.8 terabytes per second and has a total capacity of 141GB.

With the upcoming deployment of H200 chips, the only hardware issue might be accommodating the H200 chips in data centers and laboratories of private organizations and government agencies working on AI. Currently, AI-dedicated chips like the H100 model are in high demand and often sold at prices of $25,000 or more per individual chip. Considering that hundreds of such chips are required to support the development of generative AI, this represents a significant initial investment for most organizations, even if availability of these chips can be secured.

The new H200 chips could address this problem as they could achieve similar results with fewer chips and can be added to existing infrastructure already equipped with H100 chips to enhance AI processing capacity. Therefore, Sam Altman, the CEO of OpenAI – the company behind ChatGPT – believes that the acute supply problem is likely to diminish next year.

Assuming NVIDIA can produce enough H200 chips to meet the massive demand, the new chips can significantly expedite the development of generative AI, reducing training and model adaptation time from months to weeks. With hardware challenges being addressed and ethical guidelines established in governments, it seems that generative AI has a bright future in public service.

FAQ