How AI Could Empower Any Business | Andrew Ng | TED

For a long time, artificial intelligence has been something only big tech companies could really use. It was expensive to build and needed a lot of skilled engineers. But what if AI could help a local pizza shop figure out which pizza sells best? Andrew Ng talks about making AI available to everyone, so any business can use it to make more money and be more productive. We can build a better society, just by using a few data points.

The AI Divide: Why Big Tech Dominates

Right now, AI is mostly in the hands of a few experts, kind of like how reading was only for priests a long time ago. These are the highly skilled AI engineers, often working at big tech companies. Most people only get to use the AI that these companies build for them. But what if everyone could help build the future with AI? We could have a much richer society.

So, why is AI mostly with big tech companies? It's because many AI projects have been really expensive to build. They might need dozens of skilled engineers and cost millions of dollars. Big tech companies, especially those with millions or billions of users, are good at making these investments pay off. A single AI system that improves web search or recommends products can be used by a huge number of people, bringing in a lot of money.

But this way of doing AI doesn't work for most other businesses. Outside of tech and the internet, there aren't many projects that apply to 100 million people or bring in that kind of money.

AI for Everyday Businesses

Let's think about a local pizza shop. The owner makes great pizza, but often has cold pizzas sitting around, and sometimes runs out of popular flavors. This shop generates data every day. An AI system could look at this data and see patterns. For example, if Mediterranean pizzas sell really well on Friday nights, the AI could suggest making more of them on Friday afternoons.

You might think, "It's just a small pizza shop, what's the big deal?" But for that owner, improving revenue by a few thousand dollars a year would be a huge deal. There's a lot of talk about AI needing huge amounts of data, and more data does help. But AI can often work fine with smaller amounts of data, like the data from a single pizza shop.

The real problem isn't a lack of data from the pizza shop. It's that a small pizza shop can't serve enough customers to justify hiring an AI team. In the United States, there are about half a million independent restaurants. They serve millions of customers, but each restaurant is different with its own menu, customers, and sales records. A one-size-fits-all AI wouldn't work for all of them.

AI in Other Industries

What if small and local businesses could use AI? Let's look at a T-shirt company:

Key Takeaways

• Demand Forecasting: An accountant could use AI to figure out what funny memes to print on T-shirts by looking at social media trends.

• Product Placement: A store manager could take pictures of the store and have AI recommend where to place products to sell more.

• Supply Chain: AI could advise a buyer whether to pay a certain price for fabric now or wait for a cheaper deal.

• Quality Control: A quality inspector could use AI to scan fabric pictures for tears or discolorations.

Big tech companies use AI for these kinds of problems all the time, and it works well. But a typical T-shirt company, auto mechanic, retailer, school, or local farm uses AI for none of these things today. Every T-shirt maker is different, so there's no single AI that works for all of them. This is true for other industries too, even large companies like pharmaceutical companies, car makers, and hospitals.

This is called the "long-tail problem of AI." If you list all possible AI projects by value, the most valuable ones are things like online ads or web search. But further down the list are projects like T-shirt product placement or pizzeria demand forecasting. Each of these is unique and needs to be custom-built. The total value of these smaller, custom projects is huge, but no one is working on them.

Making AI Accessible

So, how can we help small businesses and individuals build AI systems that matter to them? For a long time, building AI meant writing a lot of code. While learning to code is great, not everyone has the time.

But there's a new way to build AI systems that lets more people participate. Just like how pens and paper made reading widespread, new AI development platforms are shifting the focus from writing code to providing data. This is much easier for many people to do.

Several companies are working on these platforms. Imagine an inspector who wants AI to find defects in fabric. They can take pictures of the fabric and upload them to a platform. Then, they can show the AI what tears look like by drawing rectangles, and what discolorations look like by drawing other rectangles. These pictures, with the drawn rectangles, are data that teaches the AI how to find tears and discolorations.

After the AI looks at this data, it might realize it has seen enough pictures of tears but needs more pictures of discolorations. So, the inspector can take more pictures of discolorations to help the AI learn more. By adjusting the data you give the AI, you can make it smarter.

An inspector using such a platform could, in a few hours to a few days, build a custom AI system to detect defects in fabric throughout their factory. You might say, "It's just one factory, what's the big deal?" But it's a big deal to that inspector, making their life easier. This technology can also help a baker check cake quality, an organic farmer check vegetable quality, or a furniture maker check wood quality.

These platforms might need a few more years before they're easy enough for every pizza owner to use. But many are coming along, and some are already useful for tech-savvy people with a little training. This means that instead of relying on a few experts to build AI for everyone, we can empower every accountant, store manager, buyer, and quality inspector to build their own AI systems.

AI is creating a lot of wealth, and it will keep doing so. By making AI available to everyone, we can make sure this wealth is spread widely across society. A long time ago, hardly anyone understood how much widespread reading would change things. Today, hardly anyone understands how much making AI accessible to everyone will change things. Building AI systems has been out of reach for most people, but that doesn't have to be the case. In the future, we'll empower everyone to build AI systems for themselves, and that will be an exciting future.