One of the most influential voices in AI globally, Ng, 48, who earlier headed Google Brain and was chief scientist at Chinese tech company Baidu, is also founder of AI learning platform DeepLearning.ai, general partner at AI Fund, and an adjunct professor of computer science at Stanford University.
A lot of the media coverage of GenAI has been on the large language model (LLM) or the foundation model layer, but it is very capital-intensive exercise and the real value will now come from building applications, he said adding that his fund is focussing on such start-ups and are averaging one investment a month. He also spoke about AI democratising coding for all and how AI will not take away jobs but people who do not use AI may be replaced by those who do. Edited excerpts:
What will drive the next phase for Coursera, especially given the AI reset?
The whole world is at an inflection point. With the rise of generative AI (GenAI), the nature of work is transforming. This creates a lot of pressure for organisations to upskill people. Many jobs will have 20-30% of the tasks that AI can play a material role in. I don’t think AI will replace people, but people that use AI will replace people that don’t.
What kind of approach should a country like India take to AI?
Discover the stories of your interest
A lot of the media coverage of GenAI has been on the large language model (LLM) or the foundation model layer, but that turns out to be a very capital-intensive exercise. A cutting-edge foundation model may cost $100 million, maybe more. But the applications layer, for whatever reason, has had much less media coverage, but almost by definition it’s going to be even more valuable in terms of revenue generated, say, than the foundation model layer. India has tons of very skilled people. It would be worthwhile if some work on training foundation models, maybe tuning them to local languages. But I think that a much larger fraction of India ‘s value to be created and captured would be in building applications on top of these wonderful foundation models. India’s economy has a large service sector or large industrial sector, huge agricultural sector, quite a lot of FDI (foreign direct investment) as well. India can figure out how GenAI transforms the IT services, telcos, financial services, hospitality, manufacturing, mining and construction, textiles, chemicals or other major industries.
The biggest opportunity for India is to look at the sectors where the Indian economy is already strong and figure out the applications of AI to those sectors that enable it to maintain its advantages.
People raise questions on the ROI of GenAI investments. Which side of the debate are you on?
For the application layer, it is very clear. I think it’s totally worth it, partly because it’s so capital efficient–it doesn’t cost that much to build valuable applications. And I’m seeing revenues pick up. So at the application layer, I’m not worried. Now, at the foundation model layer, given the capex spends on GPUs, I think from a long-term economics fundamentals point of view, it will be worth it. But I think from a timing point of view, there’s a lot of pressure to prove out the value of that investment in the short time horizon. And so that I think is something we should keep an eye on.
There is an AI race afoot. Do you think others will be able to challenge the dominance of the US?
The field is more open than it used to be. But frankly, the greatest concentration of GenAI talent right now is in Silicon Valley. In all my career, I’ve never seen such high concentration before in one place. I think that’s because a lot of the innovations were from two teams –OpenAI and my former team Google Brain.
While there are pockets of talent in Canada and France and the UK and China, China is clearly rapidly growing. It doesn’t feel like any country has an insurmountable lead, especially when you look at the application opportunities. But Silicon Valley does have more technical knowhow than any other city in the world right now.
What would you say is the right approach to AI regulation?
The hope from regulations is increased investments in science, research, support the growth of the industry, and support open source development. I was really surprised last year at the intensity of the lobbying to try to shut down open source. If somebody attempts to shut down or make open source difficult and succeeds, almost all nations, including specifically India, will be losers, because it will cut off a lot of the world from access to cutting-edge AI.
If governments identify specific use cases that are harmful like deepfake porn and then regulate against the applications, that would be much better than blanket attempts to regulate the technology against hypothetical harms. You’d just stifle innovation.
Regulations which support (open source) will empower a lot of startups and large businesses to go on top, as opposed to a small number of American big tech companies becoming the choke point to access GenAI.
What is the AI Fund’s approach and where do you see the most opportunities? How does India look from a startup investment perspective?
We focus mostly on the applications layer, a little bit on the tools layer. I’m seeing so many opportunities. We end up building on average about one startup per month. We often work with large corporations to try to combine the best ideas in AI with unique insights in the industry sector to address a significant industry problem. One example is we built a startup to use AI to make ships more fuel efficient–that saves 600-700 ships about half a million US dollars a year in fuel costs.
We seem to be fortunate to come across a lot of exciting opportunities to build valuable AI projects. One of them wound up launching in India–Jivi Health. We were fortunate to find Ankur Jain, who’s a fantastic CEO, whose India team built pretty cutting-edge technology that was, maybe still is, topping the Hugging Face leaderboard for healthcare LLMs.
But Indian AI startups are not getting as much funding compared to globally.
High interest rates do make life more difficult for a lot of entrepreneurs seeking funding. At the AI Fund, we budget $55,000 to build a working prototype. After that, if you look promising, the cheque is usually $1 million. That level of investment is usually enough for the startup to build a real product. People think building AI is very expensive and that’s absolutely true if you want to build foundation models, but because of all this investment in foundation models, the capital needed to build an application on top of that is now lower than ever. But you do need to identify the valuable idea and then execute well.
What is the future of programming, with AI opening it up for everyone?
The case for a lot of people to learn just a little bit of coding is stronger than it ever has been. The barrier to learning to code is lower than ever before because you have a chatbot companion.
If everyone learns to code, you can customise, you can get your computer to do a lot more tasks beyond what a website can do for you. And so it’s a significant personal productivity booster. Just as today it is not controversial that we want our children to learn language literacy, I hope in the future we will want our children to learn a little bit of coding.