Businesses' aversion to risk means they miss out on the potential rewards of machine-learning projects – but some still have impractical ideas about what artificial intelligence can really offer them.
That's according to Hilary Mason, founder of analytics and algorithm research biz Fast Forward Labs, which was last month acquired by Hadoop-flinger Cloudera.
Mason started the business back in 2014, having worked as a data scientist at Bitly and VC firm Accel Partners, with the aim of closing the gap between academic machine learning and its practical application in businesses.
"The academic community has theoretical research more than well handled, but they're not usually building a system that can be deployed," Mason told The Register when we caught up with her at the recent Strata Data Conference in New York.
"Cool innovation might happen in startups, but they often lack the resources, or the deep expertise in the problems they want to address."
Enterprises, on the other hand, might have the resources, but they struggle for other reasons, such as a culture that doesn't encourage risk-taking – or because they don't know what data they have, or what questions to ask of it.
Seeing an opportunity, Mason decided to launch a business that took basic machine-learning research and assessed how it could be applied in the enterprise.
Broadly, businesses use Fast Forward Labs for one of three things.
The most straightforward of these – and the revenue stream Cloudera was most interested in – is its quarterly research reports. These look at emerging machine-learning opportunities and prototypes that are likely to be useful in a deployed product in six months to two years.
Mason hopes that linking up with the bigger business will offer her company the chance to expand on the possibilities offered by the prototypes that are developed in this process.
"Sometimes they could be the beginning of the product, but I've struggled to figure out how to – effectively – build enterprise software. We haven't made any commitments, but it's a potential that was being wasted."
In addition to the report series, Fast Forward Labs offers two more options to businesses that fancy using machine learning: a report and recommendations on a specific problem combined with a workshop to help implement the tools; and a full-on feasibility study carried out by the team.
The feasibility study is the only time the team will ever write code for a company, Mason said, as her underlying aim is for businesses to have the data science skills they need embedded in the team.
"The philosophy is that, in order to build excellent data products and data businesses, this must be owned inside the company. You cannot outsource your data science to another company or product," Mason said.
"The way we work with our clients is designed to support them as much as possible in owning their own strategy, their own technology."
As well as helping them implement complex technologies, Mason wants to encourages companies to take risks.
"In the enterprise, often when I see a potential roadmap, every idea is a good idea – and that always scares me a bit," she said. "It's data science – there's science in it. You need to be able to explore and to try things that you don't know if they're going to work.
"If I can look at your list of projects and tell you every single one is going to work, that's engineering. That's fine, but I see my role as coming in with 20 or so out there, sometimes really bad, ideas that can inspire a bit more creativity and risk-taking."
Mason readily acknowledges that part of her role is also to act as a buffer for when things go wrong. "That's the benefit of bringing in an external point of view – I'm not going to lose my job, so I'm the one who's willing to have those ideas."
But she's also aware of the increasing buzz around AI and machine learning, and the pressure it puts on all those involved.
If people don't understand how the technology works, she said, they start to make assumptions about what's possible "that are just not rational", leading to frustrations on all sides.
"That everyone talks about it means a lot of people are talking about it without actually doing it," Mason said.
"I'm a huge optimist about this stuff, and I love to build things that actually work – we're seeing a tonne of value from machine learning being deployed at scale – but it's not what's captured in the majority of the conversation."