What do new developments in artificial intelligence mean for industry, investing and our society?

A panel of experts discussed the potential of artificial intelligence and how we can harness it.

What are the participants’ involvement with artificial intelligence (AI)?

Nathan Benaich is the founder of Air Street Capital, a venture capital firm focussed on building technology companies that accelerate human progress and that specialise in intelligence systems and deep technology products. He is also the founder of the Research and Applied AI Summit (RAAIS), co-runs London.Ai and writes at www.nathan.ai.

Steve Crossan is an executive-in-residence at Atomico, helping portfolio companies with their product and go-to-market strategies, and helping source AI companies. He formerly worked at DeepMind, where he led the product team responsible for putting DeepMind’s technology into Google, from data centre optimisation to Google Assistant.

Justin Lyon is the chief executive of Simudyne, and one of the world’s leading experts in AI and simulation technology. His current focus is using advanced data analytics and AI simulation platforms to model some of society’s most complex problems. He is a serial technology entrepreneur and has presented at Bloomberg, FT Innovate and TEDx

What’s the difference between AI, machine learning and deep learning?

AI is the broad idea of computer systems replicating the actions of a human mind. Within that definition, machine learning refers to systems solving tasks by ‘learning’ how to perform them, using algorithms and statistical models. Deep learning is a more advanced system, where machines automatically learn, for example, how to recognise features from images automatically, through processing a vast amount of data without explicit instructions from the programme developer.

 

What are the practical applications of AI, and what are its limitations?

They are all around us - in Google search, when we talk to Siri or Alexa, in our Facebook newsfeed, make payments on the internet and in voice recognition services. The catalysts driving uptake are the sheer volume of data in the modern world, the fall in the cost of and concomitant increase in computing power, and the influence of key opinion leaders in large technology companies – as well as unexpected champions such as the Bank of England’s Andy Haldane.

Over the last ten years, and particularly since 2012, AI systems have become very good at solving problems involving perception – eg image or voice recognition. They are very good at mathematical problems such as credit prediction and order execution, where there is a large amount of data and it is clear what right/wrong looks like. AI tends to struggle interpreting systems where there are a large number of participants and behaviour adapts over time, e.g. investing.

 

 

Where are the exciting new areas of development?

There is potential in many fields where better processing of information confers big advantages, including healthcare/life sciences, agriculture and industry. AI systems are trying to grow crops in a more efficient way, with fewer pesticides, using automated farming equipment. They are working to find more efficient battery components, and improving industrial automation and inspection, to reduce human casualties. The field of drug discovery is another interesting example - with large amounts of complex patient data and modelling, where traditional methods are expensive, time-consuming and have low success rates. AI could also have exciting applications here, and in genomics, personalised medicine and diagnostics.

In finance, the use of data for better modelling and simulations of complex macroeconomic events (e.g. Brexit) could lead to radically better policy decisions. It could improve private market investing, through more efficient data processing on potential targets, spotting break-out signals for promising companies, and guiding portfolio construction. It could also transform the fight against money laundering, where only a fraction of perpetrators are currently caught. Ultimately, AI technologies will be used to ‘rethink’ most digital products in our economy.

 

How far are we from fulfilling the famous ‘Turing Test’ – articulated by Alan Turing in 1950 - where humans are unable to tell the difference between an interaction with a human and a machine?

The Turing Test is something that captures the public imagination but holds less practical interest for those in the field. In fact, history shows it is relatively easy to fool humans - the ‘Eliza’ computer program in the 1960s was a sort of online counsellor, a simple scripted bot, which some people nevertheless took seriously.

People have been ascribing human qualities to tools/machines for thousands of years. What tends to interest today’s AI practitioners is not emotional interaction with machines, but how to use them as more effective tools - more robust algorithms to improve market execution, better financial modelling of shocks and contagion risks, custom-designing asset allocation etc. The potential for combining simulation with machine learning - these are the exciting areas for us.

 

Who is leading the AI investment landscape globally?

The US remains the clear leader in terms of number of firms and investment volumes, but Europe is catching up. The UK has an unprecedented cluster of academic excellence in computer science and AI, e.g. Imperial College London and University College London, as well as Cambridge, Oxford, Warwick and Bristol Universities. More broadly, over 50% of the world’s most cited computer vision and machine learning researches spent their early careers or were educated in Europe. In China, government policies are helping the field advance greatly, although in some cases raising questions around privacy.

In Europe, ‘unicorns’ (start-ups with a valuation of 1 billion US dollars or more) like Darktrace, Graphcore, OakNorth and BenevolentAI are proving that businesses can be grown to scale. From 2000-2010 there were 10 or so such firms. Since 2010 there have been over 50. European firms have historically suffered from a missing tier of investment after series A and B fundraisings. And since Google acquired DeepMind in 2014, more young people in Europe are recognising the business opportunities in AI, which has driven the proliferation of fast-growing AI-first technology start-ups.

 

What’s important to Big Tech companies in this space - what innovations are coming through there?

There are significant investments from US and Chinese firms, with Alphabet among the leaders. These are huge, data-driven companies. They have unprecedented volumes of data, and can make clear and immediate returns on investment from exploiting it better – e.g. through better advertising placement. One of the drivers behind the proliferation of AI technologies into start-ups is open source. A lot of research and software libraries are published in the open with permissive licenses, which means that companies can literally ‘stand on the shoulders of giants,’ to use Isaac Newton’s phrase. Nonetheless, creating successful AI-driven technology products today requires significant software engineering and product expertise to get right.

 

Where do we stand with ethics and regulation in the AI space? What about fears of machines taking over?

One of the biggest challenges to AI is bias. AI systems are built by humans, who are often unaware of their biases, which get reproduced and amplified in AI systems – so identifying and correcting these is a huge challenge.

Artificial general intelligence (AGI) is machine intelligence that could perform intellectual tasks on a par with humans. ‘The singularity’ is the idea that such intelligence could trigger runaway technological growth, and threaten the human race. But we are a significant way away from this. Indeed, we are so far away from it, that we don’t even know the future problems we would need to solve in order to get there.

 

Should we be concerned about robots taking human jobs?

Recent advances, like driverless trucks, and the suggestion of taxing robots, have brought our attention to a theme that has actually been around for hundreds of years – ‘technological unemployment.’ Labour market disruption by technology is indeed real. There are fewer US auto workers now, for example, than in the past. Theoretically, using machines to perform manual tasks frees up the human workforce to engage in higher-skilled jobs that require more intelligence, empathy and creativity. And the reality is that even in coming decades, the developing world will either need robots, more immigration, or longer working lives to grow the workforce and cope with our ageing populations.