The excitement around artificial intelligence (AI) has created a dynamic where perception and reality are at odds: everyone assumes that everyone else is already using it, yet relatively few people have personal experience with it, and it’s almost certain that no one is using it very well.
This is AI’s third cycle in a long history of hype – the first conference on AI took place 60 years ago this year – but what is better described as “machine learning” is still very young when it comes to how organisations implement it. While we all encounter machine learning whenever we use autocorrect, Siri, Spotify and Google, the vast majority of businesses are yet to grasp its promise, particularly when it comes to practically adding value in supporting internal decision making.
Over the last few months I’ve been asking a wide range of leaders of large and small companies how and why they are using machine learning within their organisations. By exposing the areas of confusion, concerns and different approaches business leaders are taking, these conversations highlight five interesting lessons.
Choose your question carefully
Far more important than the machine learning approach you take is the question you ask. Machine learning is not yet anywhere near “artificial general intelligence” – it remains a set of specialised tools, not a panacea.
For Deep Knowledge Ventures, the Hong Kong-based venture firm that added a machine learning algorithm named VITAL to its board in 2014, it was about adding a tool to analyse market data around investment opportunities. For global professional service firms experimenting in this space, machine learning could allow deeper and faster document analysis. Energy companies want to make better use of production and transport data to make resourcing decisions while one defence contractor is looking for “wiser” analysis of stakeholder networks in conflict zones.
While there is widespread fear that AI will be used to automate in ways that creates mass employment, the vast majority of firms I spoke to are, at least at this stage, experimenting with machine learning to augment rather than replace human decision making.
It’s therefore important to identify which processes and decisions could benefit from augmentation: is it about better contextual awareness or more efficient interrogation of proprietary data? Precise questions lead more easily to useful experimentation.
Manage your data better
Machine learning relies on data – whether big or small. If your decisions revolve around deeper or faster analysis of your own data, it’s likely you’ll need to get that in order before you can do anything else. This could mean not just new databases and better data “hygiene”, but new inputs, new workflows and new information ontologies, all before you start to build the model that can take you towards recommendation or prediction to support decision making. Don’t forget to double down on your cyber security strategy if data is now flowing to and from new places.
Invest in people
Data scientists are not cheap. Glassdoor lists the average salary of a data scientist in Palo Alto, California, as $130,000 (£100,000). And though you may not think you are competing with Silicon Valley salaries for talent, you are if you want great people: a great data scientist can easily be 50 times more valuable than a competent one, which means that both hiring and retaining them can be pricey.
You may opt to outsource many aspects of your machine learning, however, every company I spoke to, regardless of approach, said that machine learning had required a significant investment in their staff in terms of expanding both knowledge and skills.
The ecosystem is evolving rapidly
The latest rage is bots – application programming interfaces (APIs) that use machine learning to do specialised tasks such as process speech, assess text for sentiment or tag concepts. Bots can be seen as a small and, but imperfect, part of “Machine learning as a service”. If the creator of Siri is right, there will be an entire ecosystem of machine learning APIs that write their own code to meet your needs.
Companies like Salesforce have also started to integrate machine learning into their platforms, lowering the cost and friction of getting started. As the machine learning ecosystem evolves, companies will find interesting ways to combine in-house industry experience with a range of off-the-shelf tools and open source algorithms to create highly-customised decision-support tools.
The values of algorithms matter
Technologies are not “values-free” – all the tools we design, including AI systems, have a series of values, biases and assumptions built into them by their creators and reflected by the data they interrogate. Systems that use machine learning to make decisions for us can reflect or reinforce gender, racial and social biases. Compounding this, the perceived complexity of machine learning means that when it fails there is little recognition of harm and no appeal for those affected, thanks to what Cathy O’Neil calls “the authority of the inscrutable”. As we discussed during UCL School of Management debate on AI on Tuesday night, human beings need to be firmly at the centre of all our technological systems.
When our decisions are assisted by machine learning, the reasoning should be as transparent and verifiable as possible. For humans and intelligent machines to have a satisfying partnership, we need to ensure we learn from machines as much as they learn from us.
Nicholas Davis is head of society and innovation and a member of the executive committee of the World Economic Forum. He leads the organisation’s work on innovation, entrepreneurship and “the Fourth Industrial Revolution”.
To get weekly news analysis, job alerts and event notifications direct to your inbox, sign up free for Media & Tech Network membership.
All Guardian Media & Tech Network content is editorially independent except for pieces labelled “Paid for by” – find out more here.
