Here’s a quick look at the exclusive lessons shared by 10 experts at the Institute for Governance & Public Policy event on advancing AI in the public sector I hosted.     

Introducing amazing new Artificial Intelligence (AI) will help you improve productivity, increase trust and save money across the public sector or so we are told. But what are the worst risks and big challenges and what is the secret to getting the results you want from AI?

Ten of the best minds that understand AI challenges in the NHS, HMRC, insurance industry and private sector gave their insider’s view at the IGPP seminar on 16th February 2022. I discovered that four powerful forces play a big role in guaranteeing that hot new AI delivers the results you want and protects you from AI hype and bias.                       

Secret 1: Collaborate

The 1st secret is ‘Collaboration’ and it was revealed by Lord Victor Adebowale CBE, chair of the NHS Confederation and CEO of Turning Point, a social enterprise. He described how new AI can do lots of good in health and social care by enhancing clinical treatment through better utilization of specialist staff. This happens when the best people come together and collaborate. However, he stressed, there will be a disturbing and sinister side to AI if we are not careful how we create it. “AI technologies are clinical devices”, he explained and must be subject to the same clinical tests as other devices and managed carefully. This is because AI can be open to bias since it is created by humans. That bias can lead to discrimination such as when it wrongly identifies the colour of a person’s skin.

Secret 2: Test, Test and Test

Lord Victor’s advice to mitigate risk is to recognise, “It’s not the technology that is biased; it is the people behind it”. He told how the work of people and the new technology they create must be evaluated before roll-out. Thus, the 2nd secret is ‘Test, Test and Test’. To ensure new AI technologies are safe and secure and protect people from bias, create collaborations between creators, users and communities. And before you deploy new AI, complete a clinical test to verify it is safe.

Lisa Allen of the Open Data Institute, echoed the concerns of Lord Victor, describing how data fear sometimes prevented data sharing resulting in AI being fed with data that could be biased. Similarly, public trust can be lost by hoarding data which if shared would help improve results for the public. This instantly made me think of the billions of pounds of taxpayer backed money that could have been protected from fraud if data on COVID loans had been shared and matched as fraud fighters recommended at the start of the pandemic.

Secret 3: Be Transparent

87% of people feel it’s important that organisations use their data ethically and knowing how it is being used, according to a recent survey. So, the 3rd secret is ‘Transparency’, being truthful with people about the data you hold and what you do with it.

Collaboration was the big message again, this time from Leanne Summers, of the NHS AI Lab. She told us, “You can’t do digital to a business; you do it with a business”. The NHS has introduced extraordinary digital transformation, especially during the big challenges of the pandemic. Best of all, surveys show we, the public trust the NHS with our data. Those results have happened due to collaboration and bringing people with you on the DX journey. Compare these positive indicators with the low level of trust many people have for big tech companies like Cambridge Analytica the political consulting firm that came to fame through the Facebook data scandal. 

How to increase the usefulness of big data through participatory AI design which is more than traditional UX design was the learning shared by Aleks Berditchevskaia, from Nesta, the innovation foundation. Techniques to scrape unstructured social media data, reduce noise and use this to make crowd predictions and identify misinformation are examples of how participation can help increase fairness in AI models.

Using the best synthetic data and auto detection models to identify and correct bias in healthcare data, specifically for cardio vascular disease was central to the pioneering work led by speaker Dr. Puja Myles Head of Observational Research, Clinical Practice Research Datalink (CPRD) at the Medicines and Healthcare Products Regulatory Agency.

The power of collaboration to create a revolutionary new technology to prevent insurance fraud was the big take away from Dr. Julie Wall’s inspiring talk. Using data analytics to detect suspected fraud has big challenges and profiling is not allowed because it creates bias and thus is unethical. Conversely, ethical, manual human research and analysis is time consuming and not scalable.

Dr. Wall, Reader in Computer Science, Director of Impact and Innovation at UEL described how her collaboration between EUL and Intelligent Voice, backed by Innovate UK had created a powerful new AI fraud detection tool. Cutting-edge research by the partnership discovered a way to preserve privacy using new speech recognition tools. The software which is now patented and commercially available has even more potential uses such as detecting hoax calls to the blue light services.                 

Secret 4: Understand Clean Data for AI

The 4th secret is to understand the big role ‘Data Cleansing’ now plays in digital transformation and the new AI revolution and this was explained by Gareth Hole from NICE and Jorge Ruiz of Hyperscience. True automation is still to happen in many sectors and it was inspiring to learn that HMRC are leaders in UK government with the creation of bots to help staff. DWP have also led the way with automated complex appeals processes to deliver significant improvements and save public money.

Elsewhere in the public and private sector, 80% of data remains hidden beneath the water in a sea of unstructured paper and e-data lakes. Only 20% of this information is machine readable which presents big challenges and more so when you introduce multilingual big data. Thankfully, revolutionary AI is now coming of age and new technology is now able to digitise quickly and easily. In one example given by Gareth and Jorge, thousands of 16 page documents containing 400 handwritten fields were digitised with 99.4% accuracy. AI can deliver amazing results like this and the challenge now is to know where to prioritise. It was inspiring to learn that automation like cleaning and structuring big data can be completed as an exclusive project, thus giving quick improvements.

Transparency in the use of AI digital assistants in education was a central theme presented by Tom Moule, Jisc’s National Centre for AI in Tertiary Education.  Powerful Bots like Ada, created by Bolton College and Taylor, the Open University’s 2-way chat assistant will not replace humans, Tom assured us. However, they have the remarkable potential to improve lives of students and help end the crisis caused by a severe shortage of teaching professionals.

The ongoing challenges caused by silos in data and how to measure the success of an AI project was the focus of Alberto De Lazzari of Larus Business Automation. The value of using Agile Methodologies to research, plan, test and review projects cannot be understated and this is a big lesson that applies to the public, private and third sectors.

How do you get the best results from AI? 

So, to recap, what are the 4 secrets that increase the likelihood that new AI will deliver the results you want plus present little or no risk?

  1. Collaborate with the best people because it creates the likelihood of success.
  2. Test, test, test AI before deployment to mitigate bias and risk. 
  3. Be Transparent because being truthful increases trust.
  4. Understand how Data Cleansing improves AI results.

You will be assured to know that the new AI-powered multilingual tools such as Guildhawk Aided secure translation being introduced by Guildhawk, with help from extraordinary people at Sheffield Hallam University, Innovate UK and our partners are created in this way. If you want to know more, let’s talk now.