Where next for AI? Three routes we are researching

Reid Hoffman, co-founder of LinkedIn and prominent venture capitalist said: “My ideal investing is stuff that looks a little crazy now and in three years is obvious or five years is obvious.”

Venture capital investment in AI start-ups in the US surged to a record $9.3 billion in 2018, up 72 per cent from 2017, according to a PricewaterhouseCoopers report.

Here we take a look at three areas of research that demonstrate the diversity and depth of impact from the development of AI technology.

Smart shopping to boost profits

By Arne Strauss

Shopping online is a fast-growing trend. In November 2018, the percentage of online retail sales in the UK reached 21.5 per cent according to Statista. In the grocery sector, the rise of online sales is particularly significant, with transactions more than doubling between 2010 and 2016, and total value rising to a forecasted £196.9 billion in 2021.

At a time when walk-in supermarket sales are expected to drop by 4.1 per cent, online grocery shopping, conversely, is expected to grow by 3.2 per cent. This indicates British food shoppers are enjoying having their groceries delivered. And if we look at global trends, the UK appears to be ahead of the game, taking a 7.5 per cent share of online grocery sales worldwide.

Somewhat staggeringly, the UK is on course to become the second largest online grocery market after China by 2020. Clearly, the UK businesses peddling their produce online should be raking in some decent returns.

However, profit margins show the opposite. In 2018 Ocado, an online-only supermarket, reported an operating loss of two per cent.

Our AI-based research, however, could provide significant positive industry-wide impact. By focusing on the dilemma of delivery cost versus service expectations — Tesco, Sainsbury’s, Ocado and others face the classic trade-off between fulfilment logistics and customer demand for specific delivery times — our research team investigated innovative ways of using AI to increase efficiencies, and the potential profit on each sale.

At the heart of the project was the development of a dynamic delivery slot pricing policy that not only calculated the approximate logistical cost of a particular delivery, but also anticipated customers’ delivery time slot behaviour — a “foresight policy”.

By combining the traditionally separate disciplines of demand management and vehicle routing, and by applying a discrete choice model to calculate the probability of a customer selecting a given delivery time, we found we could make real-time decisions on which slots and delivery charge combinations to display.

The delivery charges were optimised dynamically based on the estimated costs of that delivery, taking into account delivery area, time slot, other orders already received that day, and orders expected to come.

Using the results of simulation studies that incorporated real, large-scale industry data, the research demonstrated that the inclusion of anticipated revenue from future orders into dynamic, estimated delivery costs could increase profitability by at least two per cent.

By introducing smart, IT-driven decision-making solutions to delivery costing policies, online retailers could substantially improve service efficiencies and help themselves to deliver faster, healthier profits.

Understanding beauty and increasing wellbeing

By Chanuki Seresinhe

The great outdoors is an essential part of a healthy, human life. By interacting with the outside world — and in particular with beautiful places — we can enhance our levels of wellbeing.

To maximise the advantages of outdoor beauty, however, we need to be able to protect certain natural areas. And we also need to be able to create beautiful spaces, in terms of landscaping, and designing attractive recreational zones.

So is there a way to quantify beauty when it comes to outdoor scenes, to guide us in what to protect, and what to create? It sounds tricky; the subjective nature of beauty makes it hard to measure. Beauty is, as they say, in the eye of the beholder.

So far, small-scale surveys have been able to gather a limited amount of quantifiable data on what type of outdoor places people find beautiful, but the scope of these surveys is restricted.

However, AI research at our Data Science Lab suggests that large amounts of data available through crowdsourcing, can be combined with recent advances in computer vision methods to offer new insights, which may help inform how we design spaces to increase human wellbeing.

Analysis of more than 200,000 outdoor images taken from the online game ScenicOrNot has led to new understandings concerning human perceptions of city images. And the introduction of convolutional neural networks (CNNs) to computer vision methods — in this case, training a CNN to evaluate the aesthetics of the environment, rather than that of a photograph itself — has led to dramatic improvements in computer vision tasks, including visual recognition, and extracting perceptions of urban neighbourhoods.

Some of the findings of this research are not surprising; natural features such as coasts and mountains are indeed associated with greater ‘scenicness’, and in urban areas these are usurped by smaller-scale natural features, such as gardens and trees.

However, for the first time, we have been able to demonstrate that buildings also play a role in how we judge outdoor beauty, notably characterful buildings and bridge-like structures. More interesting is what was not considered particularly scenic, for example flat areas of grass, such as sports fields, and no-horizon views that might be claustrophobic.

The good news is that by applying an innovative approach to data analytics and using the latest computer vision technology, we can produce quantitative insights on outdoor beauty that can guide us in day-to-day decision-making, and deliver benefits to our wellbeing, both at work and at play.

Speeding up the courts

By Joe Nandhakumar

In 2019 the UK Government awarded grants totalling more than £6.4 million to AI research projects in the legal sector.

Covering a range of fields, these projects include the acquisition of confidential data, developing voice-detecting software that can interpret emotion and linguistics, and investigating machine-supported “second opinions”, which could be employed during emotional negotiations. Projects covering consumer-related legal advice, land rights and property conveyancing were also beneficiaries of the funding.

One of the larger grants has allowed us to work in partnership with litigation analytics start-up Solomonic. The awarding of the grant is a reflection of how litigation analytics is now seen as critical to the evolution of the litigation sector.

The research will be based on machine-learning (ML) capabilities around litigation data and analytics. Specifically, it will focus on developing an ML algorithm to analyse large amounts of data, and an ML-powered dashboard to display the algorithm’s findings in plain English.

The aim is to allow litigators to track similar past cases, extract individual data points that might impact their case prospects or tactics, and utilise examples of how relevant judges have dealt with particular fact patterns or arguments.

By using the new ML-powered system, legal professionals will be able to find past cases in a fraction of the time it takes now, leading to a significant acceleration in the process of litigation research.

The new system will also enable them to provide structured insights that they otherwise would not be able to deliver, as well as evidence-based predictions, helping litigation lawyers to evaluate and prepare cases much faster.

This development of an AI-based solution is possible because of recent advances in ML and computational capacity. These have allowed ML platforms to become more sophisticated in the ways they track and analyse court data. Significantly, ML models are now able to outperform experienced litigation professionals in recognising and extracting key information.

Given the far-reaching implications of machine-led data sourcing and data analysing, there is potential to see how AI has the power to transform not just the speed in which a legal case is processed — in particular the impact on how long it takes to compile, hear and judge a case — but the litigation sector as a whole.

As to when the implications might trickle through to the lengthy courtroom dramas on our TV screens, however, the jury — for now — remains out.

Arne Strauss is Associate Professor of Operational Research and teaches Analytics in Practice on MSc Business Analytics and Operations Analytics on MSc Business with Operational Management.

Chanuki Seresinhe is a Visiting Data Researcher of the Alan Turing Institute and did her PhD at the WBS Data Science Lab.

Joe Nandhakumar is Professor of Information Systems and teaches Digital Business Strategy on MSc Management of Information Systems & Digital Innovation and Digital Transformation on the Doctor in Business Administration.

For more articles like this download Core magazine here.

Originally published at https://www.wbs.ac.uk on July 1, 2019.