In September 2021, the UK Government launched a brand new National AI Strategy. Similar to the Technology Innovation Strategy (2019) or the Innovation Strategy from earlier in 2021, the aim is to increase the UK’s advantages in successfully working with AI technologies and data.

This is needed. Research from the Economist Intelligence Unit found that when it comes to successfully using AI in the key financial services sector, Asia Pacific leads the global pack. Almost 61% of APAC respondents reported that half or more of their workload is supported by AI. This far outstretches Europe at 41%.

The Government sees the promise of an AI-led future, and has set out ambitions to ensure the UK is able to compete in a changing data and AI-oriented world. To do so, there are challenges that business leaders, technologists and the industry must work together to overcome.

Pandora’s black box

There are many news stories of real-life data disasters and AI-controlled blunders. Given the scale that governments and businesses want to leverage AI at, any less than optimal processes have an outsized effect on potentially great numbers of people.

Users of a business service risk being less efficient or effective, and at national state level, whole groups of the population may be underserved or even accidentally badly served when AI processes are poorly designed.

In fact, the ‘black box’ approach inherent in most AI designs poses serious risks in scaling unintentional bias. When we don’t know the process, it’s hard to trust the results.

Witness the A-Level furore in summer 2020. Teachers in England had 40% of their students’ assessments downgraded by the exam regulator. There was widespread feeling that the algorithm was unfair, and penalised those from less prestigious schools.

Better design and explanation of the algorithmic and data approaches taken will be key to instilling trust and better outcomes. One way of making the black box more transparent, explainable, and equitable, comes from collaboration between a more diverse group of stakeholders.

Diversity is the foundation for better AI

If an AI strategy is to succeed, diverse teams must be put in place from the start. Recognising biases in data is key to improving models and more successfully planning their real world impact across all groups.

AI needs to be explainable and its training data must be transparent. Identifying and remediating data gaps, and continually improving algorithmic logic will reduce bias at scale – but it is a process of continual adjustment. Leadership must be content with an ongoing process that may never be ‘perfect’.

The UK’s diversity within the AI sector is low. The Government’s Understanding the UK AI Labour Market research shows how 53% of businesses had no female AI employees. 40% had none from ethnic minority employees.

This is a clear line in the sand – next steps for the industry are to bring in the people and update processes to overcome the challenge represented by such a limited pool of people to help shape an AI future. There’s already an example of a strong next step: The UK already requires diverse Boards of Directors.

Of course, the majority of the data used to train AI similarly lacks this diversity. Where undiverse, or biased data is drawn on to train data models it’s easy to see these biases scaling with hard to detect but very real, and challenging to reverse, real-world impacts.

One deceptively simple solution: more diverse teams and stakeholders.

Recommended

• Broadband funding gives over 500k Scots full fibre access
• Opinion | Digital tech can help women deliver transformational change
• Student bags record six figure pay-out from Apple bug bounty scheme

AI systems need to be built, tested, and implemented by less homogenous teams and stakeholders. When wider and more representative groups of people engage with AI, society is better able to view and understand how well models are delivering equitable outcomes for all people.

Problematic data examples [set to ‘visible’]

From the HR department at Amazon, housing and crime prevention, and products that simply don’t work for everyone, there’s no shortage of mistakes made when launching data-trained products and services in the real world, damaging trust in these systems.

What’s worse, these mistakes are avoidable.

In Amazon’s case, women were discriminated against via a program trained on male CVs. Unjust and bad outcomes have been widely reported online. One example of poor product performance was the lavatory soap dispensers that did not sense darker skin tones. All these discriminated against underrepresented groups – entirely preventable with the right data and stakeholder input.

So how can we reap the benefits of the data economy, and ensure these are experienced equally by all?

1. Increase the transparency of data and AI systems so society can understand how they work – and if they are appropriate.
2. Reduce the homogeneity of those who construct data-led solutions.
3. Increase general data literacy so that society in general is able to judge the effectiveness of such solutions.

It’s tragic that many of the missteps by AI have been avoidable with appropriate levels of governance and process. It’s even more tragic considering that the field of data analytics has evolved so far from its early days of expert-only use.

No longer are analysts confined to preparing reports, or general users to static dashboards. The cutting edge technology stack here is the ‘Modern Analytics Cloud’, and it offers a consumer-grade front end to managing enterprise data collaboratively and compliantly.

Data democratisation is right here, right now, with this stack, and anyone that can use a computer knows enough to interrogate data and start achieving moments of insight to turn into action. They can be part of a community of data users working with data under transparent compliance guide rails.

Business leaders must encourage all staff to involve themselves with data. Without engagement and investment, decisions will be made in silos. Governance and support is part of modern analytic solutions, with AI guiding users to better decisions.

There’s no need to hang back and allow old ways of working to perpetuate the same old problems. Everything that went before, the old ways of interacting with data are gone, and openness, collaboration, and curiosity have been enabled.

Get the latest news from DIGIT direct to your inbox

Our newsletter covers the latest technology and IT news from Scotland and beyond, as well as in-depth features and exclusive interviews with leading figures and rising stars.

We will keep you up to date on the pivotal issues impacting the sector and let you know about key upcoming events to ensure that you don’t miss out on what’s going on across the Scottish tech community.

To subscribe, click here.