Artificial Intelligence (AI) may be finance’s current buzzword, but despite its popularity it is not the answer to all its problems. Much like its bed-fellow, Big Data, there is a tendency for many to identify AI as the solution before they know what the problem is, writes Sam Barry, Chief Executive and Co-Founder of Savernake Capital.

Latest research shows the number of AI companies founded in the UK doubled in 2014–16 compared with 2011–13, with a new AI company launching almost every week. In an information-rich age where data is power, it can be tempting for companies to consider whether they can use AI, over whether they should.

Savernake Capital is a fund management company where trading is done systematically by computers rather than humans. Its sister tech company is dedicated to managing its underlying infrastructure and architecture.

We fully embrace AI and machine-learning concepts in several of the key components of our trading architecture – but only where these techniques are the best way to solve particular problems.

Understanding and interpreting the data

When we started designing trading systems over 10 years ago, our core focus was the adaptability of our systems to ever-changing market conditions. The ability to evolve and advance is critical in today’s markets. Information has become so vast and readily available that understanding the data can have incredibly wide-ranging interpretations and outcomes.

It had become a challenge that was far beyond what existing technology could achieve, requiring more innovative ways of processing that data. This is where AI becomes an essential part of our research and development.

The Savernake tech team has been working with AI concepts for several years, and we break their application down into three core areas:

  • Finding solutions and patterns in vast informational spaces. AI can help identify patterns in what would otherwise be considered ‘noise’.
  • Adapting systems. We use key AI techniques to change our systems dynamically and decide where we focus our time and allocate risk, identifying the greatest potential future return.
  • Helping us build components and functions. We utilise machine-learning techniques to improve our own methods within the system. This helps us make accurate predictions, estimates and calculations.

Asking the right questions

The perpetual challenge with using AI techniques is that the answers they provide are intrinsically linked to How they are used, trained and developed into solutions, and therefore can result in seemingly useful but irrelevant answers.

Understanding, interpreting and applying these techniques becomes the key to obtaining any useful information from them.

Many problems we see in processing and understanding data do not require the use of AI. In many cases, more conventional statistical methods provide better solutions. For the innovation does not lie in simply using AI but in how to direct it, to teach it, to enable it to learn and to provide it with the information it requires. In other words, what is it looking to achieve and how does it know if it is right or wrong?

It is more important to understand the question you need answered and what information will lead you to that answer. If these are well understood then the right algorithm, solution or model is typically available.

At Savernake, we spend most of our research and development (R&D) time focusing on how we break down problems into smaller questions, then identifying the information required to answer them.

As an example, the question ‘what statistical models are likely to perform well over the next six months given market conditions?’ would become:

  • What are the current market conditions?
  • Do the current market conditions have any correlation to market conditions for the next six months?
  • What indicators correlate to how markets will most likely behave over the next six months?
  • What is the correlation between statistical model performance now as compared with six months’ time?
  • What scoring functions offer the best correlation to future returns?

Redefining this problem allows us to apply various AI and conventional techniques to solve the smaller challenges. These allow us to build a far more dynamic and better view of the problem and understand how it changes over time; it also allows us to target specific techniques to the different questions.

Just as important, it allows us to define where existing techniques will prove much simpler and just as successful as AI.

Maintaining a human touch

The human element in this process should not be overlooked. For it is in these tasks that our innovative tech team proves integral in breaking the problem down, defining the questions, deciding whether AI can and should be used, and then interpreting the answers.

The renowned cognitive scientist Marvin Minsky, who co-founded the Massachusetts Institute of Technology’s AI laboratory, believed that ‘the power of intelligence stems from our vast diversity, not from any single, perfect principle’.

Here at Savernake Capital we embrace that diversity, and acknowledge that the use of Artificial Intelligence will only ever succeed when it has the best human minds employing it.

---------------------------------------------------------

What is artificial intelligence

  • Artificial Intelligence (AI) describes the theory and development of computer systems able to perform tasks that normally require human intelligence.
  • AI is not a new concept, though the arrival of ‘robo’ apps has brought it to the fore of public awareness. This means that many people instinctively link AI with robotics.
  • In fact, AI can simulate many human traits, such as knowledge, perception, reasoning, learning, planning and problem solving.
  • At Savernake Capital we use machine-learning technology; a type of AI that provides computers with the ability to learn, solve problems and take actions without being explicitly programmed.
  • This ability to use learnt behaviours on new problems is what separates AI from traditional complex processes and statistics.

An original version of this article was first published in the 2016/17 edition of the Parliamentary Review, September 2017.