Video: What’s on Your Mind? AI Edition​​​

BRAD NEUMAN: Hi, I’m Brad Neuman, Director of Market Strategy at Alger and today I’m answering AI questions submitted on behalf of investment professionals. This is “What’s on Your Mind.” 

The first question we have is “Machine learning has been around for a couple of decades. What makes AI so investment-ready now?​​” Well, I’d say a couple things. First, from a technical standpoint, it’s been a confluence of three different factors.

Obviously, much more powerful chips that are getting more powerful every year; two, a proliferation of data, which is really important to feed into these AI programs; and third, in the past few years, there’s been a new transformer architecture that enables parallel processing that has really allowed to speed up and make AI platforms more powerful.

The other thing that I think makes AI investment-ready is that, unlike some things that people have gotten excited about in capital markets in the past, AI dollar spending is actually happening now. So, if you look at Nvidia’s data center revenue, it’s increased about 500% over the past couple of years, because companies are spending real dollars on this opportunity right now.

The second question we have is “Can you explain enablers and adopters? Are enablers the picks and shovels of AI?” So, we typically divide up AI opportunities into enablers. And those are the companies, yes, they are kind of like picks and shovels, developing the infrastructure to make AI work, and we delineate that from AI adopters and those are the companies that are adopting AI to either make their businesses more efficient, or make their products more client friendly, and provide more value to their clients.

So, a good example of a company doing both might be Pinterest, a social media company. On the revenue side, it’s seeing a 25% lift in relevancy from advertisements using AI in some of its key categories. So, better relevancy means higher return on investment for its customers, means it can monetize those advertisements better. And Pinterest is also seeing a benefit on its cost structure from using AI. More than half its engineers are using AI assistance to help in their coding efforts, for example.

In terms of examples on the enabler side, obviously, we’ve already talked about Nvidia being kind of the poster child for the chips that enable the brain power of artificial intelligence. But maybe something that’s not in the press quite so much might be a database type company, like a MongoDB that operates a database for unstructured data. If you think about all the data that goes into these large language models, they’re not neatly categorized, they’re what we call unstructured, things like video and text from web pages, and so that all needs to be organized for AI to work properly.

The next question we have here is “What’s the short-term and long-term impact of AI on the economy?” So, in the short to medium term, I think we’ll see big increases in technology investment. For example, Google said it will spend over $100 billion on artificial intelligence in the next few years, and Microsoft is reportedly building a $115 billion data center. These are large numbers that will need a lot of semiconductors and related equipment to help power them, so that’ll have a big impact on the economy.

Over the medium to long term, I think the big impact on the economy is from a productivity perspective.

So, there have been some studies that suggest that productivity lift using artificial intelligence could be as high as 25%. We multiply that by the 60 percent or so of knowledge workers in America. That would give you about a 15% uplift. Maybe that happens over a course of several years. But that 15% would translate into about $4 trillion for the U.S. economy and if you extrapolate it to the rest of the world, it would be in excess of $15 trillion, so a big impact from productivity.

The last question that we have is “What are some of the most exciting innovations that AI will enable longer term?” I’ll answer this in two ways. First, I’ll talk about the technical leaps that AI may have over the next few years and then I’ll talk about kind of the specific applications. So, from a technical standpoint right now, AI can only have the context of a couple hundred thousand or so tokens. A token is roughly equivalent to a word. So, you can’t give AI too much data like the complete works of Shakespeare and ask it to analyze it. It wouldn’t be able to store all that, have all that context.

But we’re already seeing leaps up to a million tokens of context and I think in the next few years, we’ll see tens of millions and ultimately maybe billions of tokens, meaning that AI will know you better than your parents know you because it will know all of your emails, all of your conversations, etc. And that brings me to one of the most exciting applications of AI in the future and that’s the personal assistant. I think Bill Gates has said whoever cracks the code of the personal assistant will really win the game in AI and so a personal assistant that helps you do everything from plan travel to write an email back to your boss or client, I think that’s going to be really powerful, once it has all of your work and personal data.

I also think another technical leap for AI will be the ability to think deeply. So right now, we give AI a fraction of a second to basically give us an answer. But as we all know, sometimes we want to spend more time thinking about different questions and reason through it. And there’s new architectures that have recently been developed that allow AI to spend longer and think about questions more deeply. And I think that’ll really help with some applications. 

One exciting longer-term application, I think, for AI is drug development. So right now, when we develop a drug, we have to spend typically several years in an FDA process figuring out how efficacious that drug is and what the downsides are of that potential drug. And the only way to do that is to have actual, well first animals and then humans, take the drug. But in the future, AI may be able to simulate the human body to such an extent that we could do this whole trial virtually rather than actually in humans and that could reduce what used to take years into something that takes weeks, days or even shorter and that could have tremendous upside for the speed of drug development and help society in terms of health span and longevity.

So that’s all the time we have today. If you have any further questions, please get in touch with your Alger representative​. Thanks so much for watching.