Meet the professor who thinks generic AI models should become obsolete
David Ayman Shamma, who has researched AI for more than two decades, said all-purpose AI models may soon outgrow their usefulness.
OAKLAND – He worked on and researched artificial intelligence for more than two decades for organizations from Flickr to NASA. Now a professor at Northeastern University, David Ayman Shamma has his sights set on exploring the next era of AI.
Shamma, who recently joined Khoury College of Computer Sciences as its director of computing programs on Northeastern’s Oakland campus, believes his diverse professional experiences have led to a deep understanding of the humans who use this technology. As students plan for their futures in the AI industry, they should focus on meeting users and their needs where they are with specialized applications, rather than overwhelming them with models that try to do everything, he said.
Doing so requires a multi-faceted approach to learning, using their computer science curriculum as just one piece of a larger toolkit, he said. Shamma’s journey is one such roadmap to consider.
Before joining Northeastern, he worked on AI sensors for wearable fashion at Centrum Wiskunde & Informatica, an Amsterdam-based research center, and on hardware-based AI models at FX Palo Alto Laboratory, an advanced research and development subsidiary of Fuji Xerox that focused on AI and human-computer interaction before closing in 2020. He also researched human-computer interaction and AI as a director of research at Yahoo Labs and Flickr, user experience at NASA’s Center for Mars Exploration, and carbon neutrality at the Toyota Research Institute. Shamma holds over 100 patents, and previously taught studio art and journalism.
Shamma spoke with Northeastern Global News about his multi-disciplinary career, his predictions about the future of AI, and the skills he believes will be the most valuable for the next generation of Northeastern computer scientists.
The interview has been edited for length and clarity.
What would you say is the through line that connects your diverse career experiences?
My research has always been about building AI that enhances people and their experiences, versus replacing it. It’s easy to try to build AI that does automation, but if you take the time to understand people and the task, the AI you build can support that and almost make them superhuman.
That kind of approach has allowed me to be free and domain agnostic. The methods we use in building human-centered AI and human-centered computing are portable.
How essential is it for computer scientists to understand users and their unique needs before creating a product or service?
When you’re building a system, sometimes you find answers that you don’t want to find.
At Toyota Research Institute, I had the great privilege of working alongside some really good behavioral scientists. We would run these studies to find out how to improve people’s understanding of carbon dioxide, including one where people could see how much CO2 their car trip would take. But they weren’t happy that this was revealed to them, even though they became more aware and had more cognitive utility.
Without understanding people a priori throughout the whole life cycle of end-to-end, you really won’t grasp their needs. Honestly, once you figure out the people, you’ve probably simplified away the need for complexity. You have to take that on the chin and say, “OK, let’s actually think about what we need to build versus what we want to build.”
As an artist yourself, do you think artificial intelligence has its place in creative pursuits?
Editor’s Picks
My first teaching position was in studio art, and I got to audit a lot of classes because I was on faculty. It was interesting to me as a young computer scientist, seeing how painters could teach creative processes and mechanical ones together.
But when it comes to the digitization or industrialization of these creative acts, there’s a part that often gets lost. We get too enamored with generative AI building something to take the creative agency away from the artist.
We should be building “creativity support tools” that help somebody achieve their vision faster, not just do it for them.
What are the most pressing issues in the world of AI at this current moment?
I think we have a lot of people who feel so disenfranchised – rightfully or wrongfully so – with AI that they want total opt-out. But this is becoming problematic, because all the tools you use have some form of “AI” happening.
Across the industry, we need to start to rethink what we are trying to build, why we are building it, and how we build it in a way that can provide the right answers and explanations for what it’s doing.
I’ll predict two trends, and one is already happening.
More stuff will go on-device, less of it in the cloud. There are good privacy reasons for that, and it’s better than setting up massive data centers if your phone can run the compute by itself and not actually have to make these huge round trips to cloud stores.
The second thing is that we will fine tune these models. Right now, we’re too focused on general AI. There’s no reason why, if I’m trying to find how many days are in a fortnight, that I need to go to a general purpose LLM. The general purpose appliance that is AI that we’re all excited and enamored with should go away, and we should think about the more exact, focused bits of AI.
Generic AI is like a hammer, and if you have a hammer, everything’s a nail. It’s the wrong approach, but it’s what we’re doing because people think it’s impressive or it’s a showing of power, and we need to get over that. That’s what I look forward to.
What can Khoury students get excited about, in terms of their proximity to Silicon Valley?
I’m working to bring up the research side in Oakland at an undergraduate and graduate level. Given my two decades of industry background, and how I’ve worked in industry to connect to academia, we’re also looking at finding deeper connections across the board.
That means funding for research, people to come by and give talks, or having people working with our practicums or capstones and other project classes.
I’m excited to build out the campus and really have it shine. Oakland can be differentiated … from what we can offer across the network of campuses, including Boston. Our proximity to the Bay Area? Let’s use it.
