By Dr John Yardley, Founder and CEO, Threads Software

There can be little doubt that more than most other industries, the global banking and finance industry has been in the vanguard of the digital transformation of business. Yet the people – staff and customers – on the receiving end of the transformation do not always perceive the benefit of digitisation. Why is this the case?

One key separation in the digitisation process is how we get the digits in and out of machines versus what we do with the digits once inside the machines. The latter is the growth area and as we run out of simple things to do, such as adding them up, we are increasingly turning to sophisticated processes to extract hidden information. This largely falls under the umbrella of artificial intelligence.

The ultimate winners will be those who can not only infer meaning from the numbers – but can successfully transfer meaning to humans that use them.

The relatively straightforward process of converting hard-copy transactions into electronic digits has mostly been achieved. As any computer scientist knows, the first issue is getting the basic information into the system as 0s an 1s. The next step is moving those 0s and 1s around at least as well as was done when they were on bits of paper. Having achieved these goals, the focus is “can we get these digits in more cheaply and/or quickly and what can we do with the digits that we could never dream of with bits of paper.” But there is no point in getting PhD whizz kids to predict a failing client from their last 10 transactions, if this information is never used, misunderstood or used in a way that is counter-productive. This falls under the largely neglected area of Man/Machine Interaction (MMI), and its importance in global digitisation deserves significantly more focus than it receives because efforts in making systems usable can yield greater returns than simply making them do more things.

Lest there is any doubt, ask yourself what percentage of Microsoft Word features you use, and whether there are things you wanted to do but did not know how to? If you cannot create a subscript, for example, it is not because Word cannot do it, it is because you do not know how to do it, ie, it is a failing of Word’s MMI not you.

To put this issue in context, we need to go back to Alan Turing’s original definition of Artificial Intelligence, or as he described it, Machine Intelligence. To paraphrase, Turing said that once you abstract the human from the machine – for example with a telephone or computer screen and keyboard – if the human cannot tell that the machine is not a human, then the machine can be said to be acting intelligently. More plainly, does the machine react sufficiently like a human to fool a human?

Turing’s definition generally stands up to any environment where we are seeking to replace humans with machines. In most cases, a human expects an intelligent machine to act like another human and makes no allowance for the fact that the machine may compensate by performing some tasks more efficiently than a human.

As an example, suppose it takes a machine 10 attempts to correctly recognise a spoken account number. This may still be quicker and cheaper than waiting for a human operator to answer the phone, yet the caller would be unlikely to see it that way.

Another example might be a caller that is forced to listen to a 45 second legal disclaimer on each and every call before being able to speak to someone that can deal with the enquiry. This may make a subsequent litigation case easier to defend, but it takes no account of the effect it has on the caller. All the clever computation under the sun is of no use if the caller hangs up.

The point about both these examples is that the machines are being used in ways that cannot be described as intelligent, despite the fact that in the case of the speech recognition at least, the technology deployed can be massive – that is, very large computers executing code that has taken hundreds of man-years to write.

Unfortunately, the term artificial intelligence seems to have lost the meaning that its creator originally intended. Nowadays many people regard the processes rather than the behaviour as intelligent. Intelligent processes are now seen as those that mimic the human brain’s operation rather than its perception. A good example of this is a neural network, which seeks to perform a task by learning in the same way as does the human brain. This is good for the programmer because once a neural network has been coded, you can throw almost any arbitrary task at it without needing to understand the underlying science behind it. All you need are a great many examples of doing it right and doing it wrong.

All this would be academic but for the fact that in our quest to provide more sophisticated processes, we seem to have forgotten the human factors that convey the information to and from its ultimate goal, the human brain.

The reason these situations arise is that little or no attention has been given to the Man-Machine Interface. This is because the system has not been designed in a holistic way. Computer programmers generally love systems to have many features – so-called feature bloat – and lawyers like to preempt possible future cases to be defended. Neither see the system as a whole.

Now that our technological infrastructure is approaching the intrinsic limits placed upon it by physics, Artificial intelligence seems the only way to gain the competitive edge against rivals. The broad categories where this can be applied are as follows: –

• Data networks – data transmission, analysis of networks and dataflow – for example, switching, transmission lines, etc.
• Data processing – algorithms, databases, indexing, operational analysis, and so on
• Learning and adapting – for example, neural networks.
• Security and privacy – improving security using statistical methods, through, for example, cryptography, block-chains
• Technology – improving performance and efficiency by mechanical and electronic evolution via for example, semiconductors, quantum computing
• Man-Machine Interaction – pattern recognition, language translation, user interfaces (GUIs and so on), heuristics.

Viewed in the context of things we can potentially apply our resources to, MMI seems to not to account for much. Yet it is the critical path.

Yet the term “intelligent,” as defined by Turing, is irrelevant for all but MMI. It does not matter how complicated the interface between one computer and another becomes because either it works or it does not. Yet this is exactly where we concentrate our resources.

So what should we do?

Well, we need to balance the development resources between the data processing and the communication. Furthermore, the skills necessary to produce a great human interface are rarely the same skills that are required to write highly efficient programs or produce watertight legal agreements. The only person that has these human interface skills is the person putting the data in or getting the data out.

While there is a lot of science in human psychology, we don’t often know what works until we try it. We should therefore consult the user before, during and after developing the human interface. Indeed, we should decide on the human factors before we write a line of code to do the data processing.

Some key points in designing human interfaces are: –

• Identify the points where humans are involved
• Establish the minimum amount of information that needs to be exchanged
• Use as much context as possible to avoid exchanging unnecessary or redundant information (eg how many times do you need to ask for an account number you already have?)
• Always allow the human operator to go back, or to continue after a break without re-inputting data
• Test the interaction before, during and after the system development
• Regularly solicit feedback from users and act upon it

Lastly, as a good example of the importance of human factors, let’s look at cookies.

Whenever we interact with a service using a web browser, the interaction is what is technically described as “stateless.” This means that upon each exchange of information, data cannot be carried across from one part of the process to another. This can have some rather drastic effects on the usability of a service, and programmers overcome this using things called “cookies” to store information on the user’s computer. For example, a username might be stored in a cookie to save the user entering it every time.

At various points in time various countries have decided that cookies were potentially a breach of our privacy rights and users should not be using them without knowing. As a result, most websites will not allow access without first going through some interaction agreeing to their use.

On opening a page for one of the main UK clearing banks, the user is asked whether he or she agrees to the bank’s cookie policy. If the user does not know what a cookie is, let alone what the bank’s cookie policy is, the only option is click the “preferences” button (not that it is obvious what preferences are) and get a 1,000-word description of cookies and what they might be used for. While the designers of this website can legitimately claim that they are forced into the cookie interaction due to regulatory constraints, it becomes clear that this has been turned into a form of absolution. This may not be apparent to the bank, but it is to the customers and there is mounting evidence that cookie consent notices are meaningless and manipulative. If the customer rejects all cookies, the situation can be equally counterproductive.

Of course, the legislators must take responsibility for not thinking through the implications of their legislation, but the situation could have been more palatable to clients if some intelligence was applied to the human factors of dealing with them.

So, to summarise, effective man-machine interaction is just as much a part of the digital transformation as artificial intelligence algorithms. The financial services businesses that realise this will ultimately be the winners in the race to attract and keep customers.

 

About Author:

John  began his career as a researcher in computer science and electronic engineering with the National Physical Laboratory (NPL), where he undertook a PhD in speech recognition.  In  1990 he founded JPY Limited, a state-of-the-art distribution, software development and consultancy company.  Today, JPY  represents manufacturers of over 30 software products, distributed through a channel of 100 specialist resellers.

In early 2019, John founded Threads Software Ltd as a spin off from his company JPY Ltd to commercialise and exploit the Threads Intelligent Message Hub, developed originally by JPY Ltd.

John brings a depth of understanding of a wide range of the technologies that underpin the software industry.   He has a PhD in  Electrical Engineering from the University of Essex and a BSc in Computer Science from City University,  London.