Bringing AI benefits to business

The meteoric rise of ChatGPT has shown that artificial intelligence tools are ready for mass adoption.

But how should you position your business to take advantage of them? It can be tempting to own the AI systems by building and keeping the code and their parameters a secret.

Doing so is extremely expensive, however. Consider that OpenAI, Meta, Google, Amazon Web Services and a host of smaller startups are spending hundreds of millions of dollars trying to create the best models that exist.

These companies have the resources to hire the top talent in the world and access to data to train their models on. It will be costly to compete against them.

On top of this, companies such as Meta and leading universities will often open source their models. This typically makes them free to use.

Imagine spending hundreds of millions for a machine learning model, only to find that Meta has given away a better one for free on its website. Owning a secret AI is a rapidly depreciating asset that can become worthless overnight.

Don’t own the compute

US software company Nvidia has been one of the biggest winners from the rise of AI. Its GPU chips power huge data centres of AI systems, training and crunching the numbers to create better, more powerful AIs.

The chips are in such high demand they are now very expensive and hard to come by.

Yet owning the compute is also a trap.

Unless you are able to create your own chips to rival Nvidia’s, you will always be needing to buy newer and better chips.

If your business model is to rent out these chips to others, you’ll be competing with big players like Amazon Web Services, which already has extensive data centres and has figured out how to make it profitable.

But the biggest threat to owning the compute is the open source community.

As compute becomes more scarce and expensive, it starts to encourage open source contributions to invent models that don’t need large compute clusters. Researchers with limited budgets are starting to experiment with ways to train models that no longer need huge data centres.

If you invest in the compute you may end up with stranded assets and objects that perform poorly compared with newer models. Worse still, they may not even be necessary to run the next generation of AI systems.

Own the small data

For all of their amazing capabilities, AI systems like ChatGPT can be pretty useless on their own.

A more profitable route is to just apply AI to a particular niche. In this approach, your business maintains a small dataset used to fine tune an AI system, and that AI is applied to your specific problem domain.

Canva provides a good example of this model. Canva is an online social media creation tool that lets its users make pictures or videos for marketing.

Canva has an embedded AI that lets users create images of anything they want by describing it with text. This makes the Canva product even more useful, as sometimes users won’t find the perfect image in their existing catalogue.

Following this model, companies can leverage their existing market channels and expertise to supercharge their offerings with AI.

By aiming to control the market and their small data sets, these companies can avoid the huge costs of AI but still capture its benefits.

No execution? Worthless

Some business owners feel that they should hoard their data because they think it’s valuable.

If your company isn’t a technology company (for example a mining or a consulting company), don’t try to hold onto your data. Instead, use it a bargaining chip to partner with a technology company to help you.

Having a big pile of data is the same as having a big pile of dirt: totally useless unless you know what to do with it.

• John Vial has a PhD in robotics and has spent the past several years leading teams in major Perth businesses focused on AI and robotics