Stanford’s Renowned AI Experts Agree, Clash on State of AI at CES 2024

Artificial intelligence has been around for decades and experienced cycles of euphoria and AI winters − when development, interest and funding dried up after the tech did not live up to its hyped potential.

Will it happen again today? That was the question at hand during the Great Minds, Bold Visions: What’s Next for AI? panel discussion Tuesday at CES.

Not likely, according to Andrew Ng, founder of Google Brain and adjunct professor at Stanford who once taught OpenAI CEO Sam Altman.

“We’re not in a winter and that is because the business fundamentals of AI are stronger than ever, even before the generative AI wave that took off last year,” Ng said. “AI has been moving probably billions of dollars, maybe trillions of dollars. For a single company like Google to show you more relevant ads (using AI), that drives massive amounts of revenue.”

What makes AI relevant today is that it is a general-purpose technology, Ng argued. “It is kind of like electricity. … If I asked you what is electricity good for it is almost hard to answer that because it is useful for so many different things. And AI is like that too.”

“Even if AI makes no technological progress (from this point), … there are so many use cases all around the world to be identified and built out to business use cases. I am confident it will continue to grow,” he added.

Related:The New York Times sues OpenAI, Microsoft

Fei Fei Li, co-director of Stanford’s Human-Centered AI Institute and former chief scientist of AI/ML at Google Cloud, agrees “more or less.”

“What we have seen is another inflection point of AI and that inflection point was with the large language models or the first rollout of ChatGPT and then the ensuing models,” she said. What she sees is AI becoming a “deepened horizontal technology” that translates to a “true transformative driving force of the next digital or industrial revolution.”

“This technology is here to stay, it is here to be deepening into all vertical businesses and consumer experiences and it is changing the very fabric of our societal, economic and political landscape,” Li said. “That is just a fact.”

On AI breakthroughs for 2024

Just as large language models took off, Ng sees the emergence of large vision models, in which AI models will be analyzing images and not just generating them, with implications for self-driving cars.

Another trend is edge AI, Ng said. “It is actually getting quite feasible to run the large language model on your laptop. Not as big as GPT-4 but big enough to be useful.” As computers get more powerful to run AI at the edge, people will upgrade and “drive a lot of device sales.”

A third trend Ng sees is the rise of autonomous agents, which takes generative AI a step further by doing the task, not just writing out the steps.

“It plans out a sequence of actions like do these web searches and download these web pages and summarize these things and it goes off and does half an hour of work … and comes back to me with an answer,” he said. These autonomous agents “plan on and execute sequences of actions.”

Today, autonomous agents are “barely working” and Ng expects more advances to come this year.

But Li disagreed with the term autonomous agent, instead preferring assistive agent, implying humans in the loop to ensure quality outputs. “What I actually see is … part of the work is collaborative (and) is more likely to happen rather than (being) fully autonomous.”

She gives the example of a nurse working an 8-hour shift. The job consists of hundreds of tasks. Li sees AI agents helping the nurse with some tasks but not taking over all of the tasks. The AI is assistive, not autonomous.

Ng argued that in his experience working with companies, the calculation is often more about the business ROI instead of ethical decision. If AI can replace jobs, not just tasks, the company might make that switch.

And what yields the highest ROI from AI might not be what businesses first think it would be, Ng added. For example, for a radiologist, a reasonable assumption would be that AI would help them interpret X-rays. But actually, it turns out there is a higher ROI by automating another job of the radiologist: gathering patient histories.

“To do this exercise systematically (of breaking down a job into tasks and testing each one’s ROI) has often helped businesses identify valuable opportunities to then go through build vs. buy kind of decision,” Ng said.

Generative AI Is an imprecise term

Both Li and Ng agree that the term “generative AI” is not quite precise. “It is an overloaded word,” Li said. “When Andrew and I started, we had very specific mathematical definitions of generative AI. … We also used the (words) content generated versus discriminative.”

“We used to call it machine learning,” Ng added.

Now the mass media is overusing the term generative AI to mean a “large data-driven” model with a “pre-training phase” that uses the Transformer technique and predictive modeling, even if “I am not totally sure if people always do” include them, Li said.

Which tech company will reign in 2024?

While OpenAI took the headlines in 2023, this year will be more diffused.

“I see this technology deepening and also widening into all sectors, and because of that, it is hard to single out one company,” Li said. “I do believe 2024 will be defined as a year (where) we see the widening of AI applications as well as AI technology, not just focusing on one or two” companies.

Ng added that every time there is a new wave of tech innovation, the media focuses on the big companies that build the foundation layer. “But it turns out that for this technology infrastructure layer to be successful, there is another section that needs to be even more successful and that is the application layer built on top of these tools providers … to generate even more revenue so they can afford to pay the tool builders.”

The New York Times’ lawsuit against OpenAI, Microsoft

Ng said he read the lawsuit and in this case sides with OpenAI and Microsoft. The Times accuses the two companies of copyright infringement because their AI models – ChatGPT and Bing Chat, respectively − can quote its news stories verbatim. (Bing Chat uses OpenAI’s models.)

Ng said the lawsuit muddies the argument and is sensationalist. He takes issue with the prompt used by the Times to find those verbatim text. It was a “very strange prompt that I do not think pretty much any normal user of OpenAI will use.”

Instead, the Times had found a bug that enabled ChatGPT to regurgitate verbatim copyrighted text, which Ng said was not appropriate. But it was a bug that OpenAI can fix and “sadly sometimes there are bugs in software.”

Li noted the “messiness” of generative AI. “With a technology as profound as this, it gets messy with the human world.” The lawsuit is indicative of the tension found between the technology and the creator economy, not just for big players but also for the individual artist.

“That whole ecosystem is being challenged, disrupted as well as augmented by today’s generative AI technology,” she said.

Ng added that there is a “very human element” to this disruption, as seen during the Hollywood actors and writers strike. “If you are a creator, you think it is my job that will go away and all my work can be stolen. … Copyright law needs to be cleaned up for the generative AI era.”