The first artificial Intelligence (AI) conference was held in Dartmouth, in 1956 – and the participants thought they’d see working intelligent computers as a commonplace in their working lifetimes.
But the stubborn inability for any of the systems those pioneers developed to pass the Turing Test definition of human-like ability and the relative disappointment of Expert Systems at the end of the 1980s meant the overblown hopes of AI got quietly shelved.
However, perhaps it’s time to reassess the contribution AI can make – as the rise of the highly-functional software suite has been taking place, almost without us looking. These machines may not meet all definitions of ‘intelligent’ – but these ‘smart,’ cognitive platforms are capable of many complex or abstract capabilities that have been considered exclusively human.
After all, these are machines capable of winning against world chess Grand Masters, excelling at TV game shows and accurately recognising faces. And they’re even present on the roads – arguably proving to be much safer drivers than humans will ever be.
All that’s been made possible via incremental leaps in computing power and new ways of thinking about how to encode intelligent behaviour in software. As a result, computers can now ‘see,’ ‘listen,’ ‘read,’ and ‘write’; in a limited way, of course, but there is undeniable some understanding taking place.
In fact a lot of what used to be thought of as AI has left the research lab and is available on various consumer technologies – from the game console to the smartphone and desktop. Skype can translate natural language (NL) conversations in real time, while speech recognition and computer dialogue is commonplace in call centres.
NL services like Siri on the iPhone and the Google Now service that tries to anticipate a user’s information needs are used widely, while sentiment analysis allows agents to determine the emotional state of the person they are interacting with, and to respond accordingly.
In effect, interactions with smart agents are getting slicker, more naturalistic and more pervasive. Computers can now do more than interact with humans in semi-realistic ways, they can also problem solve and offer technical assistance in various businesses and industries. Virtual assistants provide technical assistance to experts in fields as diverse as aeronautics to the oil and gas sector, performing complex constructions and repairs, for example.
IBM has been the acknowledged leader in this first wave of practical AI since its Watson machine won the game show Jeopardy three years ago, and when its predecessor, Deep Blue, beat chess wizard Garry Kasparov. Watson’s now moved to more real-world applications, from cancer research (improving diagnosis accuracy as well as revealing new mechanisms of the disease) to helping chefs explore new flavours and create new dishes.
In addition, a system from a firm called IPsoft, Amelia, is a smart agent with a convincing avatar that can read and understand text, follow processes, solve problems and learn from experience. Notably, ‘she’ understands implied, not just stated, meanings, and improves her performance by hearing humans deal with questions she can’t yet answer.
Amelia can digest an oil-well centrifugal-pump manual in 30 seconds – and give instructions for repairs, do the job of a call-centre operator, a mortgage or insurance agent, even a medical assistant, with virtually no human help. To achieve these feats, the software is not set up to cover every possible situation, as not only is this an impossible task, but also pointless in many cases (how can you write a procedure if you don’t know the solution to the problem – e.g., curing cancer?).
This learning aspect to modern AI is key to making cognitive technologies work in real environments: learning agents don’t need to be reprogrammed each time a new situation arises, they naturally improve and expand their capabilities with time.
Which brings us to a key question: could this happen in industries like financial services?
The application of practical thinking machines to old problems
We think the sector naturally lends itself to the use of cognitive technologies. In fact, the complexity of the financial markets, the vast amount of data, and the need for automation and better customer experience make cognitive technologies a convincing solution in a wide variety of situations.
In risk management and compliance, for example, smart agents can evaluate all cases against approved policies and guidelines and understand the complexities of risk exposure. Financial and market analysis could be made more insightful through the analysis of vast amounts of information. In sharp contrast to traditional analytics, smart agents are more than happy, thanks to the design philosophy mentioned, with open-ended questions, detecting key trends and variables human traders miss.
Today, in wealth management, relationship managers advise their clients by analysing large volumes of complex data such as research reports, product information, and customer profiles – how big a step is it before smart advisors also provide cost-effective, personalised investment advice based on an ever-growing corpus of investment knowledge?
In fact, systems like Watson and Amelia are already used by financial institutions. DBS Bank uses Watson to identify the needs of wealth management customers and determine optimal financial options; one of the biggest US banks uses Amelia to manage trading platforms and call centres.
There is also Kensho’s Warren, which can assess how different securities react after the release of a market-moving piece of information and provides advice to investors. The developers of the system expect it will be able to answer more than 100 million types of financial questions – sooner rather than later.
Lastly, Markit produces a newswire of automated reports written by a robot called Quill, where the language is convincingly human – and the output, at 40 a day, impressive. What’s more, the economics make complete sense. The speed at which some tasks can be achieved by smart machines makes them a necessity rather than a luxury. Given the increased volumes of data businesses have to address, their adoption is becoming an increasing requirement.
The conclusion seems inescapable. We have been too dismissive about the limitations of AI after its public missteps in the 1980s – and we are now at a point where cognitive technology capabilities are powerful and reliable enough to be deployed in complex business environments, including the global financial services vertical.
Humans will remain very much part of the equation, but their roles will become increasingly aligned to what they are good at, using the brain power that the 1956 dreamers discovered couldn’t be modelled. But we’re getting closer and closer, nonetheless.