As another Nvidia GTC event played out this week, Jensen Huang, the company’s co-founder, president and CEO, reflected on an AI technology space that is becoming increasingly lucrative for Nvidia, but which also continues to face many of the old questions about ethical responsibility and the need for regulatory constraints.
Many of those questions have gained new relevance in light of the rather sudden success of OpenAI’s ChatGPT and now GPT-4, and how these and other Generative AI services are taking over the AI narrative.
As Nvidia has built up its AI technology stack in recent years, Huang has been as vocal a champion of the technology as anyone, and more insistent than most about the future benefits of AI and the massive importance that places on continuing to grow computing power and efficiency. He recognizes that the AI evolution very recently has entered a new phase.
“The first wave of AI was about reinventing the computing architecture altogether,” he said during a media Q&A. “Realizing that machine learning and deep learning were going to change the type of software that we can write and the way that we would write software altogether. Because of that, the nature of a computer and what a computer is would be completely reinvented. We dedicated the last 12, 13, 14 years or so reinventing every single layer of computing, from the chips to the system architectures to the system software to all of the libraries and engines and frameworks for each one of the domains of applications all the way to the applications themselves.”
The second wave of AI revolved around perception, he said. That meant using deep learning, computer vision, speech recognition, large language models, and related technologies to perceive and classify objects and data and use these capabilities to drive breakthroughs in autonomous driving, robotics, and other areas.
“The third wave that we're starting to see now, what we would consider intelligence… reasoning about what's going on in the world. And reasoning, of course, is very complex. And the next step is… action planning.” The combination of reasoning and planning is what we are currently seeing coming to fruition in the form of Generative AI, Huang noted, adding that new Generative AI applications have “triggered an inflection point” in AI technology and inference that increases the demand for AI supercomputers.
Huang described the eventual fourth wave as where “the digital world and the physical world come together” to help a variety of industries through digital twins and related metaverse technology.
Over the last two years, Nvidia has talked up this fourth wave much more than it has the current wave focused on Generative AI, but it is possible that Generative AI inserted itself in between with a force and urgency that no one saw coming, not even Nvidia’s visionary leader.
Count Nvidia among those firms that are still assessing how big of a deal that Generative AI will be for its bottom line. When questioned by one reporter about the revenue potential of Generative AI, Huang responded, “In Generative AI, our current revenue is probably something approximately close to… a single-digit percentage. Tiny, tiny, tiny. If I were to project out 12 months from now–so 12 months ago, tiny, but 12 months from now, quite large. How large exactly is hard to say. But I would say quite large, very large.”
He added that Generative AI solutions have accelerated demand for AI training on Nvidia DGX systems, as well as inference capabilities of other systems. “We're seeing accelerated demand quite strongly for both inference and training,” he commented. The need to train more language models, image models and other types of models to support even more Generative AI applications will continue to grow the overall AI market pot.
Yet, even as Generative AI is creating a new market driver for Nvidia and other companies, it also is raising questions about how these applications will ethically source different types of content and how creators of that content will be fairly compensated as Generative AI applications act on their own to leveraging it. Along with that, many questions continue to be raised about how to ethically manage Generative AI, regulate it, and when necessary, rein it in.
Asked about Nvidia’s own standards for ethical AI, Huang said, “We only sell to customers who do good if we believe that a customer is using our products to do harm, we would surely cut that off.” However, he noted that Nvidia ultimately is the provider of technology to build and train Generative AI solutions, and not the developer of user-facing applications.
“The best that we could do, as employers and as parents and as leaders, is to provide guardrails [for AI] that are partly based on influence and knowledge, partly based on values and partly based on systems.”
Regarding the systems guardrails, which a company like Nvidia has the most control over, Huang advised that Generative AI and almost all AI remains a work in progress and that it requires ongoing testing and refining. “AI as a product will never be done,” he said. “It operates 24/7 and you're refining it 24/7.”
Pressed on the topic of whether or not AI regulation is needed, Huang said Nvidia has not been asked to contribute to regulatory frameworks, but he added, “I've also been clear that regulation is necessary regulation. Anything that affects breaking social norms and the safety of people should be regulated. It would be a very common sense thing to do.”
Still, those who would regulate AI need to understand what it is they are regulating, Huang said.
RELATED: ChatGPT runs on 10K Nvidia GPUs with potential for thousands more