'Many' ways to create artificial intelligence. Just ask the UK's AI businesses
What? We have them. There's life outside the hype bubble
Nothing brings a smile to the face of Sabine Toulson – co-founder in 1995 of Intelligent Financial Systems – faster than the notion that AI and its associated technologies are “something new”.
Both Sabine and husband Darren were graduates of UCL’s Artificial Intelligence Lab – alongside other veteran entrepreneurs such as Jason Kingdon, who founded UCL spinout Searchspace, which was famous at the time for the quality of its anti-money laundering software.
Searchspace has been using machine learning techniques for years to combat money laundering, employing tools that compared millions of transactions and distinguished between legitimate and fraudulent transactions between buyers and sellers.
Like Searchspace, Intelligent Financial Systems (IFS) succeeded early in cracking the difficult US financial software market. Back in 2000, the company won a contract to study and analyse the enormous volumes of data emerging daily from the Chicago Board of Trade. It was an exceptional feat, and not just because the board had given the contract to a non-US company. The episode reflects the very strong US interest – both then and now – in the future of the UK’s AI sector.
IFS – the subject of many a takeover offer – continues to produce trading software for the London Stock Exchange, big Japanese banks and Euronext-LIFFE, among others.
That early handful of AI wizards has grown and in the past few years – especially after Google and Twitter bought some very young UK AI companies for huge sums – interest in AI applications among a new generation exploded.
At the same time, big improvements in computing power have accelerated a revolution in AI – with Alphabet, Amazon, Apple, Facebook and Microsoft all invested heavily. Much of the popular, if febrile, debate has concentrated on whether AI – and their Earthly agents, robots – will do us out of jobs and, ultimately, dominate us.
Marking students' papers...
In practice, few realise how ubiquitous AI has already become among SMEs. By 2017 one index of SMEs found that no fewer than 192 UK companies claimed to be adopting some form of what they defined as AI or machine learning into their operations – spanning IT, medicine, biotech, the professions, security, and games.
These firms range from newcomers such as advertising decision-maker Adbrain to smart tracking micro firm Armadale Technologies, developing an Intelligent Video Surveillance (IVS) system aimed at analysing and predicting human behaviour. These companies employ word or visual matching, pattern recognition and cluster mapping techniques of pure machine learning.
In 2010 Assessment21 used AI to “mark exam papers electronically.” The software was originally written to help Manchester University cut the costs of setting, administering and marking traditional paper exams. Assessment21 tests students online and is apparently capable of assessing a variety of question types.
Academic software to auto-mark multiple-choice questionnaires is now standard. But Assess By Computer, Assessment21’s product, can mark complex, open-ended questions that test students’ understanding – not just their memory. The software picks up on key words in students’ answers and allows them to be evaluated against a model answer. It can highlight answers that are similar, and be used as an anti-plagiarism tool.
Dr David Alexander Smith, meanwhile, is the key man at Matchdeck – a rival to Experian that offers an introductory service to 16 million companies, fitting buyers to sellers. The firm crunches records using data models and matching algorithms, employing something it calls an AI web extraction engine and a semantic big-linked data platform.
Cutting through the hype
But what exactly is “AI” in this context? It’s a big topic with lots of related subjects and there’s plenty of hype right now. Ian Page, a former Oxford academic, entrepreneur, and now director of Seven Spires Investments, reckons on “many” approaches to creating AI. This allows many Brit tech and engineering SMEs to coalesce under the broader AI umbrella.
“The one that is the hottest news right now is based on a much-simplified model of how individual brain cells (neurons) might connect together and process information. These Neural Nets have been around for decades but it is only with recent reductions in the cost of powerful computers that researchers have been able to build much more complex neural nets, the so-called Deep Neural Nets, and to find ways of training those DNNs on vast amounts of data,” he notes.
The result is software that is able to “learn”, or update itself through the activity of searching and discovering patterns, connections and linkages in large volumes of data – pinpointing the sort of lateral thinking that we used to believe only the human brain was capable of achieving.
In the 1990s, Page’s research group implemented AI algorithms of different types: neural networks, simulated annealing, genetic/evolutionary algorithms, cellular automata, and even a singing synthesiser.
But, in his view, computers and AI software will still have a hard time competing in real world functions with the human brain. “It can’t be irrelevant that the human/mammalian brain has lots of diverse physical structure,” Page said.
“Whatever the human brain is doing, it definitely is not doing it within a single architectural paradigm. So, if nature and evolution couldn’t do it (general intelligence that is) within a single network of neurons, however big, then it seems odds on favourite that AI researchers won’t be able to crack that problem either within the framework of only DNNs.”
Neural networks today typically have a few thousand to a few million units and millions of connections. Hilariously, their computing power is similar to the brain of a worm – and several orders of magnitude simpler than a human brain.
Perhaps the most interesting fact is the way “ordinary” UK companies – those outside the Silicon Roundabout bubble and beyond the blinkers of those focussed on digital personal assistants like Siri – have forged products, processes and markets across the widest range of applications.
IntelliMon – part of STS Defence – this year introduced a satellite-linked monitoring technology that can monitor the biggest marine diesel engines on the high seas and transmit a simple “health score” to a vessel’s operator thousands of miles away. The system employs a combination of sensors to capture vast amounts of data and machine learning.
Being able to predict when a supertanker, container vessel or cruise ship needs to be brought into port for engine maintenance can avoid breakdowns at sea, saving six-figure sums for shipping owners and management companies.
The innovation lies primarily in the algorithms devised by the Institute of Industrial Research at the University of Portsmouth. They analyse vibration readings by extracting key engine performance indicators that can be translated into basic, byte-sized ‘health score’ information. These can then be sent back to shore via satellite link or, potentially, even using the vessel’s own automatic ID transponder.
David Garrity, STS Defence chief scientist, said: “We began work with 450 tests of different faults created on a purpose-built diesel engine test rig [we] developed which operated at constant speed bands, mimicking engines on ships.” Other potential applications lie in off-road vehicles, whether battle tanks or earth movers, and remote diesel generators in oil and gas installations.
Earlier, in October 2016, it had designed an electronic personal protection system designed to detect and predict the rapid rise in temperature that precedes a “flashover” incident for the emergency services. Thermal sensors use artificial intelligence to analyse the rapidly changing temperatures in a smoke-filled contained-fire environment where firefighters frequently operate. Its warnings give fire fighters vital time to flee.
Rainbird Technologies has won an enviable contract with financial services giant Mastercard. The payments giant will use its smarts to power an automated, virtual sales assistant. Rainbird claims to offer a cognitive reasoning platform, something that uses Machine Learning and lots of relevant data to make recommendations. With Mastercard, Rainbird will use the experience gleaned from the entire sales team and the thousands of customer conversations, to help predict which calls might convert to sales.
The UK AI ventures and projects are as strong as they were more than 25 years ago when Sabine got off that plane from Chicago with a contract in her pocket. ®
Sponsored: Beyond the Data Frontier