NEW YORK - With nine in ten AI startups addressing a specific business function or vertical, the technologies surrounding artificial intelligence are already ripe for use by businesses. However, whether an enterprise is exploring an initial AI proof-of-concept experiment or reached operational deployment, challenges abound.

Appen is a global leader in the development of high-quality, human annotated datasets for machine learning and artificial intelligence. With over 20 years of experience, expertise in more than 180 languages, and access to a worldwide crowd of over 400,000, Appen partners with global companies to enhance their machine learning-based products.

Following this week's news that Appen are acquiring machine learning startup Figure Eight for AU$340 million, we sat down with Appen CEO Mark Brayan to discuss the state of AI today; both the dominant trends around AI for business, as well as the key roadblocks up ahead in 2019. Mark brings over twenty-five years experience in technology and services, overseeing Appen's leadership, strategy, and culture.

Q: What are the key things holding back AI in business?

M: One major thing holding back AI is the data people need in order to build it. What it comes down to, ultimately, is that if you can get the data, the infrastructure, and the right people building the applications in place, I think that the opportunities are fairly broad.

Q: Say you're approached by a company who want to start working with AI - where do they begin?

M: We often recommend a pilot project. You can build usable AI using thousands of data points, but to build a sophisticated AI that matches human levels of performance, you need millions - if not tens of millions - of data points.

Our clients often ask us for small amounts of data so they can conduct a pilot or an experiment. They can see if they reach the outcomes they anticipate, and then they can course-correct and / or add more data to build a product.

Q: How has the conversation around AI and data developed?

M: The company's been around for 20 years, and we focused for a long time on speech and language data - AI that can effectively communicate with humans being one of the core tenets of AI. One of the real things we've seen change is that people no longer just want 5 hours of speech data. They want 50,000 hours, because the data volume and the quality required for effective AI has grown enormously.

Another thing that's changed is the sheer variety of use cases. We see use cases in financial services, insurance, medical technology, and more. I also think that our customers are certainly more savvy around the quality and particularly the volume of the data they need.

C: Some of your clients are working in verticals with low data density. What are the challenges there in terms of working with them to build their capabilities?

M: That's where we do most of our work. The most effective AI is very narrow, very specific AI, and so we find that people can take existing models and through transfer learning add very specific data and get the use cases they want - which is good for our business because that specific use data doesn't necessarily exist, so we have to go and collect it and label it and provide it to our clients.

I think another big challenge people are concerned about is the implications of AI for jobs. Now, while it might not destroy jobs, AI will change jobs, absolutely. However, it can also create jobs. We pay 40,000 people a month to provide data through our crowd-sourcing operation, so that's 40,000 jobs that didn't exist before. So there'll be a change in the nature of work, and I think that's understandably something to be concerned about, but overall, technology through history has shown it improves productivity and creates opportunities.

C: Moving forward to next year, what do you really think the priorities should be for companies looking to engage in this conversation around AI?

M: For businesses, it's about looking at the business problems you're trying to solve and considering whether AI can solve them for you. It all starts with the outcome you want - is it a product, is it a problem, what are you trying to do with it? Then, break it down into its constituent parts. Do you have the data scientists, the data, the infrastructure that you need? From there, conduct a few experiments to see if you can solve that problem or build the product that you want.

Learn more about Appen here