AI is a learning system, even Amazon is still getting it wrong...
AI is a learning system, even Amazon is still getting it wrong - former government AI strategist
AI is a learning system, even Amazon is still getting it wrong - former government AI strategist

AI is a learning system, even Amazon is still getting it wrong – former government AI strategist

At today’s DataIQ Future conference, Scott Gallacher, former AI and machine learning strategist at the Department for Work & Pensions (DWP), spoke about the opportunity for AI in the public sector.

Central to his presentation was the claim that it’s very difficult to make an algorithm that can encompass everybody’s situation today and know that it will still work in the future.

“You need something that is essentially going to learn and understand,” he said. That’s why he sees artificial intelligence (AI) as “essential” for both public and private sector organizations.

Digital at the heart of government

The UK government has placed the intelligent use of data at the centre of its business transformation. It’s an admirable aim, but with growing volumes of data, legacy systems and ever-changing social and political requirements, it’s not easy.

Gallacher was brought in by the DWP to provide what he calls an “adrenaline shot” to the digital work the government was already doing. He’s a man with experience in this field, having helped the likes of Sky and with various transformation projects.

In an interview with Internet of Business, he noted that the government had been “making good progress,” but with more than 80,000 staff it was struggling to get all of the various functions of technology, operations and strategy working in tandem.

“The start-up culture of move fast and break things doesn’t work for governments,” he affirmed. The DWP is the largest government department and is responsible for roughly 20 million citizens’ welfare and pensions. However, systems that might have helped, such as a master database of citizen data have been rejected due to privacy issues.

The AI roadmap

To tackle these problems, Gallacher worked with the DWP to create a vision and roadmap of what could be achieved with AI by 2020. This involved bringing in experts from across the marketplace to create a data authority, which Gallacher describes as a “cross-functional group” that assessed the problems and determined how to solve them. The team also used market “Sherpas” to find the best companies to work with.

With the framework in place, he was able to set up a sandbox to test the systems to determine what could be done with the technology and what the business outcomes would be.

The specifics around the government program are still under wraps, but it’s the model that Gallacher is advocating: a simple plan, an achievable and understandable goal, with a specifically framed question about what you want to do with AI.

Related: Artificial intelligence needed to make sense of IoT data

“The truth for a lot of AI and machine learning stuff is that it’s a similar situation to where Big Data was around 2009, in that I speak to people who are telling me they’re going to do Big Data…on Oracle. That’s like trying to win the grand prix in a Skoda. It’s not going to happen.”

“There’s going to be new chips that are going to allow those things to happen, but they’re not here yet, so you’re going to have to make the best efforts with what you’ve got,” he said. “Lots of people say ‘we’ll get AI and therefore we can plug it into the system and get all the answers we need.’ It’s just not the case.”

As an example, he said that most of us have used Amazon at some point, which means most of us have been given a recommendation for a product we didn’t want.

“That’s one of the most advanced technology companies in the world…and they’re still getting [AI] wrong. Given that, how likely is it you can build a system in a short space of time that’s going to…give the information you want? Pretty unlikely,” he said.

Gallacher was keen to stress that this is going to be a learning process for a lot of companies, but we must remember that not everything can be done by machines. He believes that with both humans and AI “the two are needed together” to determine how this technology will work.

While there’s a long road ahead, his key takeaway was both positive and reassuring for UK businesses. “If you’re considering doing with stuff with AI machine intelligence, London and the UK is probably one of the best places in the world to be able do that. There’s a tremendous amount going on and there’s some incredible companies that are doing it.”

Related: IBM Watson’s artificial intelligence to help solve complex medical cases