Think tank: don’t minimise the challenges around delivering successful social AI

Solving some of the issues facing us around AI are as big a challenge as solving public sector legacy data issues, Reform’s Eleonora Harwich warned delegates at last week’s Think AI for Public Sector

Posted 26 September 2017 by Gary Flood

The questions we have to find answers for in the public sector – and indeed, wider British society – when it comes to Artificial Intelligence (AI) seem simple enough: when will we move to mass use of them, and what is the best ethical framework for using the tech when we do?

The reality, trend watcher Eleonora Harwich of think tank Reform, told an audience of public sector IT decision makers and strategists at last week’s Think AI for Public Sector, is a lot more complicated.

For Harwich, a Senior Researcher at her organisation, which styles itself as an independent, non-party think tank whose mission is to “set out a better way to deliver public services and economic prosperity”, there are clearly goals to be achieved in the sector with targeted use of AI.

These include data-rich ways to give policy makers new powers around the prediction and detection of social issues, better support for service delivery, and a chance to automate repetitive tasks.

The problem getting there, she told her audience: challenges around data and a need for more thinking about the right way to accredit, monitor and verify such future services.

“Data is the fuel of AI, but that means we need to work out better ways to share data in the sector for it to get a chance to work with it – and we have some big problems about the quality of that data to work out first, too,” she cautioned.

“Solving those issues are probably as big a challenge as solving all our legacy data issues – another area that isn’t sexy, doesn’t get talked about, will take a lot of resource and skills, but is still a vital task,” she went on.

Possible paths to explore here could include new data models for the sector going forward, she said, but, “This will take political will and isn’t going to be cured just with band-aids.”

An equally big issue is the ethics of AI delivery, she noted. Issues that need hard thinking and work to solve on this front include hidden biases in data that distort AI findings, the fact that so much of its processing is ‘black box’ by nature (and so hard to interrogate) and the vital need to find ways to assign accountability.

Harwich called for greater engagement with these issues by all stakeholders in the sector to make AI a realistic and practical tool for society, she concluded.