In response to our readers’ many questions about AI, showing a mix of excitement and concern, we will soon debut a series of monthly articles on the status of AI in the global HPC community and the path toward AGI, artificial general intelligence. The series’ primary author will be BigDATAwire contributing editor Steve Conway, whose day job is as senior analyst with Intersect360 Research (prior stops at Cray Research and SGI).
Steve has closely tracked AI developments for over a decade, leading HPC and AI studies for government agencies around the world, co-authoring with Johns Hopkins University Advanced Physics Laboratory (JHUAPL) an AI primer for senior U.S. military leaders and speaking frequently on AI and related topics. In this kickoff interview, I ask Steve about plans for the series.
BigDATAwire: Thanks, Steve, for taking the lead on this series. AI is a fraught topic with a wide range of users and opinions. How are you approaching this?
Steve Conway: Humbly. It seems clear that AI winters are behind us and AI is here to stay now, although progress may happen in spurts, including surprising advances like China’s DeepSeek. I like Intel’s approach, dividing the market into a kind of “AI light” for enterprise business operations and the segment I’ll focus on, frontier AI for advanced use in HPC and certain hyperscale applications. The series doesn’t aim to be definitive; the goal instead is to lay out a range of current information and opinions on AI for the HPC-AI community to consider. No one has the final take on AI, so I also want to encourage comments and suggestions along the way at steve@intersect360.com.
Steve Conway is a senior adviser at Intersect360 and a contributing editor at BigDATAwire
BDW: Where will your information come from?
Conway: I’ll rely on three main sources: first, everything I’ve read and done during the past dozen years; second, worldwide market research and analysis from our analyst team at Intersect360 research; and above all, the many candid discussions I’ve had in recent months with leading figures in the global HPC and hyperscale communities.
BDW: With all that in mind, where does AI stand today? Can you summarize the situation here?
Conway: I’ll try, Generative AI is popular and useful. It’s really amped up excitement about AI, but it’s early in the game and artificial general intelligence, AGI, is probably decades away by any robust definition. Definitions are one of the challenges we face. There’s no strong consensus definition of AGI, or intelligence itself, or explainability related to today’s main AI methods. I think it’s still useful to talk about the path toward AGI, even if we don’t know exactly what AGI looks like. The series will talk a fair amount about this path, starting with views from leading thinkers on where things stand today and differing schools of thought on what it will take to approach or achieve AGI.
BDW: In general, where do people think AI stands today? I know you’ll explore this more deeply in the series.
Conway: People generally agree that despite the impressive abilities of generative AI, we’re still mostly confined to path problems, where the human defines the goal and the AI agent pursues the goal. This approach works with many scientific, engineering and business problems. There are far fewer examples yet of insight problems, where the human can’t define the goal and the AI agent has to advance independently to a breakthrough solution. People generally agree that AI today is mostly confined to computable problems that can be approached with logic and isn’t adept at handling everyday problems humans tackle with heuristics, methods that kinda-sorta work most of the time.
BDW: How is AI being used in the worldwide HPC community?
El Capitan is number one on the TOP500 list
Conway: Nearly all HPC sites are exploring AI and many are starting to use it in production. That’s no secret. But beyond that it depends on the applications, as it always has with important new technologies in the HPC community. At one end of the spectrum are legacy applications in domains like automotive and aerospace, where results using new technology may need to be validated and recertified, a process that can take years. At the other end are non-mission-critical jobs that can be assigned to AI quickly. It varies a lot by domain. Oil and gas majors still tilt heavily toward seismic processing using traditional physics-based simulation, with AI helping out in alternative energy research. Aircraft companies tell me AI won’t replace pilots anytime soon, but in light of the serious global pilot shortage, some cockpit tasks can probably be offloaded to AI. It depends on the apps.
BDW: What are some of the challenges you’ll discuss?
Conway: Intersect360 Research figures confirm that the AI market is much larger than the HPC market. There’s concern, given the current shortage of GPUs, that processor vendors might bow to the greater buying power of hyperscalers and pay less attention to HPC needs such as FP64. There’s concern HPC sites might have less ability to negotiate prices for new supercomputers. There’s concern that scientific and engineering users might need to learn to do their work on machines designed for AI. A few HPC sites are already far along in that direction.
BDW: How are HPC leaders feeling about the situation?
Conway: Among those my Intersect360 Research colleagues and I have talked with recently, I’d say attitudes vary from cautious optimism to mild pessimism. The optimists say the HPC community has made major technology shifts before, such as migrating apps from vector processors to x86. Pessimists tend to worry more about things like AI potentially replacing HPC at the top of the IT food chain and the financial and other consequences of that diminished status.
BDW: Can you say a little more about the road ahead for AI? How do people see AGI?
Conway: The major schools of thought on AGI split along the lines of the centuries-old mind/body debate, with some believing that technological progress alone will be enough to get us there and others arguing that a major breakthrough will be needed to endow machines with “mind,” consciousness. The Cartesians seem to be in the minority today.
BDW: Thanks, Steve. Looking forward to more details in the articles that will appear about once a month.
The post Watch for New BigDATAwire Column: AI Today and Tomorrow appeared first on BigDATAwire.
0 Commentaires