The effect of artificial intelligence on our lives is quickly growing – and will continue to do so. But how can we use data to build trust in this new technology?
Medical diagnosis, language translation, chatbots, fraud detection and facial recognition are just a few of the ways artificial intelligence (AI) is being put to work. And not far behind are autonomous vehicles – not only on roads, but also in the air.
As AI quickly shifts from speculation to experimentation to widespread adoption, an important question arises: how do we know when an AI system is trustworthy? Increasingly, this concern is prompting governments around the world to develop frameworks for ensuring the promise of AI does not come with threats to social and economic rights.
The Data Spark project surpassed our expectations in researching assurance business examples, market size and competitor analysis
Imperial MBA students recently worked on a Data Spark project with Validate AI (VAI), a community interest company in the process of developing a certification scheme for trustworthy AI applications that will benefit consumers.
To develop a universal certification system that effectively reflects the trustworthiness of a programme, the following questions must be considered:
- Is it fair?
- Ethical?
- Socially appropriate?
- Efficient and effective?
To tackle this broad set of questions, the team conducted focus groups with consumers and system developers to understand their biggest concerns, and what these mean for practical certification schemes.
The team also explored the history of certification schemes in other fields to understand how the fusion of technical analysis, organisational development, and branding all contribute to certification. For example, Underwriters Laboratories have built a franchise for assessing and certifying all kinds of electrical goods. Founded in the late 19th century, it has grown to become an organisation with an annual turnover of $2 billion.
Why branding matters
The Data Spark team advised VAI to focus on branding as they develop their certification, to ensure consumer familiarity with the scheme. This involved a shift for the organisation: VAI had to switch from a think tank model, developing ideas and promoting the benefits of certification for a new technology, to developing a scheme that needs to be universally recognised in order to be effective.
The Imperial team then went on to analyse how VAI can build trust with system developers, regulators and consumers, using a three-pronged approach: hosting public conferences, regularly publishing articles and white papers to establish VAI as a thought leader, and implementing its AI certification process.
Finally, the team advised the company on a roadmap for establishing a third-party audit programme for AI systems. A key feature of the programme will be delivery by organisations with expertise in both assurance practices, such as auditors of risk and financial statements, and machine learning and data science.
“We were delighted to commission Imperial MBA students to refine our existing strategic plans, building on our prior Validate AI success of convening experts to showcase how to develop trusted AI systems," said CEO of Validate AI Shakeel Khan.
"The Data Spark project surpassed our expectations in researching assurance business examples, market size and competitor analysis. Daren and Chen also excelled leading focus group discussions to help better understand student, corporate and government demand for AI certification services. A big thank you to the Data Spark team for helping our strategic focus to progress to the next stage of the Validate AI journey.”
Written by Daren Tan (MBA Candidate 2022) and Evie Burrows-Taylor
Team members: Daren Tan, Chen Zhang (MBA Candidates 2022)
Academic mentor: Dr Mark Kennedy
Extracting value from data: Imperial College Data Spark on the “wicked problem” of nuclear weapons
The importance of understanding the role of nuclear weapons in international security cannot be understated – and to achieve this, data is key
Share
