Davos 2025: Business-friendly shift could boost AI and innovation, says Andrew Ng on Trump inauguration

The AI pioneer wants professionals to learn coding and to harness AI to stay ahead of the curve. He thinks that the US' AI export control policy will be harmful to India and the US, and also believes that President Donald Trump will be more pro-business than the previous Biden administration.  Edited excerpts of the interview:

Reports say that OpenAI has possibly achieved a breakthrough in doing PhD level work with its AI agents or super agents. Meta's Mark Zuckerberg said recently they might not need engineers anymore to write code. Are we moving too fast along the AI curve? 

I think there is no practical way to slow down AI but I think there's a lot of work ahead of us to make sure that everyone is lifted up and we don't leave people behind. I think for a nation like India, what I would say is leapfrog, leapfrog, leapfrog. The nature of work is changing, and in the future, I think almost everyone, not just software engineers, but also marketers, recruiters and journalists, will be able to use computers much more powerfully than it was ever possible before.

So to anyone worried about your job, I say learn to code and take control of it because in the future, the ability to direct computers to do exactly what you want will be one of the most important skills. And, so I don't think AI will replace people but people who know how to use AI will replace people that don't.

Also read: Davos 2025: Andrew Ng advises Indian professionals to learn coding amid AI revolution

Should people really learn to code anymore? There's a view that one can just key in a prompt and you automatically get whatever program you want.  

So it turns out that — to get a little bit technical — writing English or some of the language prompts, it works a lot of the time but not all the time. And so now, I think, people who know how to code will be able to get computers to do much more than people who don't understand at a deeper layer how the computers work.

So for example, I know how to code reasonably well but I code much less because I get an AI to write a lot of my code. But if I didn't know how to code, I would do much less of computers.

Are we reaching the limits of scaling laws in AI or is there significant room for improvement?  

So the lemon is getting harder and harder to squeeze. With every generation, scaling laws get a bit harder. I think there is more juice in that lemon. It is worth, you know, maybe one or two more squeezes. But in terms of AI progress, frankly, I'm not worried because we have multiple fronts by which AI is advancing, not just scaling laws, but also agentic workflows, where you can get an AI to not just tell the response right away, but iterate and think for a long time before giving out a response.

And so I find that the set of tasks that AI could do grew a lot because of scaling laws, just more computers, bigger data. But now we have other paths as well. Very promising to keep on expanding the set of tasks that AI could do.

Click here to watch the full interview

Everybody's talking about agentic AI and AI agents. How soon will they be viable? With the rise of agentic AI systems capable of reasoning and planning, what breakthroughs are needed to make AI agents truly autonomous and widely deployable in business and research?  

I think they are already economical and increasingly widely deployed.  Sequoia Capital had an article on the $600 billion problem or question which talked about with all the capex investments in GPUs and data centre build-outs, when would it pay off? And I think that's a problem or question that the foundation model trainers have to answer. I'm optimistic they'll come up with a good answer.

But it turns out that for building applications on top of these foundation models, if someone else has spent billions of dollars training, the economics works out because with a handful of engineers, you can often build a prototype or build something, and deploy it.

So, I'm seeing a lot of businesses, certainly at AI Fund, I see a lot of businesses get ROI very quickly from a very modest investment.

You said before that India's AI opportunity lies at the application layer. But within India, there's a debate. Should we build LLMs? Should we focus on applications? Are we letting go of the opportunity to build foundational models? Why do you think India should focus on applications? What specific applications should India prioritise to maximize economic impact?  

I think India should do both. But the large, very large majority of the focus should be on applications. India has so many talented people and so many great companies, it has the resources to both train foundation models and build applications.

But one interesting thing about foundation models is the cost of training them falls rapidly over time. So models that were cutting edge, you know, two, three years ago, now many research labs can train and so the moats are not very strong. We'll see how it evolves.

The only way for the businesses training foundation models to kind of work out and do well is the application layer turns out to be even more valuable because we need the applications to generate even more revenue for them to be able to afford to pay the technology companies. So I think in India, I think that the nature of work is transforming.

I'm seeing people from all walks of life, journalists, marketers, recruiters, able to use computers to become much more powerful. And I think India with a very strong STEM education, a lot of tech penetration, lots of software services, IT services is actually in a pretty good position to help many people in many different walks of life.

Also read: Davos 2025: Andrew Ng urges India to focus on AI applications

As AI compute becomes a geopolitical weapon, how should companies and countries navigate the US-China AI decoupling. The US has also placed controls on export of AI chips, which could have an impact on emerging economies like India. What's the solution?

Yeah, I think with the outgoing Biden administration's recent declaration of India Tier 2 in terms of the prioritisation of access to GPUs. Candidly, I don't think that's good for the US or for India. I feel like what the export controls on chips have done for the most part is incentivise the US's competitors and adversaries to invest a ton in building their own independent capability and it's not really slowed others down that much, only a little bit.

Also read: Davos 2025: US AI export controls could harm both India and US, says Andrew Ng

So, I feel like I think it is unfortunate the state of the world. I hope that you know, honestly, I think India and the United States feel like they're natural allies. I really love working with many friends from India. I wish, I really hope for continued improvements in how the US and India collaborate.

It's a big day today for the US as Donald Trump returns to the White House. What impact will this have on AI? Will it take us to the forefront? Will there be more focus on semiconductor manufacturing creating jobs in the US?

It's still too early to say. I know people have very strong feelings about the incoming president. But I think in Silicon Valley, the widespread feeling is that this will be more business-friendly. I feel like the parts of the outgoing administration felt like it was hostile to tech, even hostile to American tech companies to a level that I've never seen before in my career and, so separately from many other issues, which are important issues, I feel like more business-friendly environments will allow more innovation and allow more things to flourish.

What are your go-to AI apps or chatbots apart from Coursera?

I use multiple of them. I use OpenAI Cloud. I use quite a large number of them. I use quite a few agentic workflows and also code up things myself. Recently, my team released an open-source AI suite to help developers use multiple large language models and simultaneously make it more easy. So I feel like it's good to have a suite of AI tools.

Do you think the last year has sort of determined the winners in the AI race? We've seen DeepMind make great proofs, Anthropic do well and OpenAI. Do you think these three have sort of emerged among the multitude of players that we saw at the foundational level?

I would say they definitely have strong positions. Oh, I should say the latest version, Gemini 2.0 was really impressive but I think it's too early to call it. Partly because the cost of training foundation models falls rapidly over time. So it'll be interesting as these businesses continue to grow, to what extent they've built more defensible modes. I know that OpenAI's ChatGPT, the consumer business is a wonderful business but training foundation models, because of, you know, Nvidia and AMD and others driving down the cost of compute, I think it gets easier and easier for many others to train GPT-4 class models. And so I think it'll be interesting to see how the business dynamics evolve.

What would your advise to an 18-year-old who wants to choose a discipline that will not be disrupted by AI? And even if it gets disrupted, his or her job will be secure.

I think the world is changing quickly. I think one of the most important skills in the future will be to be able to tell a computer exactly what you want to do. And this will be a useful skill in pretty much all professions, certainly all white-collar, probably many blue-collar professions as well. So that 18-year-old, I would say, go learn computers, go learn to code, because if you're worried about AI, take control of it and be one of the people who can direct it to do what you want. And that will set you up to be in a much better position for the rest of your career.