Interview Jim Stolze

Jim Stolze demystifies AI

‘It really shouldn’t be called artificial intelligence’ 

Jim Stolze, writer and entrepreneur, states, ‘One of the mistakes we have made as nerds and computer scientists is that we have used the term “artificial intelligence”. Stolze, who provides AI solutions to large businesses through his company Aigency, will be the ringmaster for the event The Bright Future of Finance on 2 November 2021. ‘It’s about statistics. The more often data or a pattern appears, the greater the probability that a computer can do something with it.’ 

With this statement, Stolze wants to directly demystify the term artificial intelligence. It has little to do with visions of the future, such as those depicted in the Terminator films. No, there will be no robots that can act independently and unleash a war. ‘Even though it’s my favourite film, Terminator has given AI a bad image’, said Stolze. ‘The term machine learning is actually a better one. It’s all about the comparisons or deviations which a computer can find in a container of data that was once labelled by humans. This makes it a better fit with statistics. Nowadays we have so much processing power in computers, so you can search in a great deal of data: Big Data. So much data that we as humans can never understand it and discern connections. To do that, you could create a software robot which you would send on its way through that data to perform a task. You do have to tell that software robot what it must do, however, as the robot is actually dumb even though it’s very useful. Compare this to humans who can learn from something they have encountered only once. That’s called one-shot learning.’ 

Quick recognition of spots 
Stolze has just completed filming an AI course, which took him and a film crew to Alkmaar’s hospital. ‘I spoke with two radiologists there who specialise in lungs and brains. All day long, they look at scans on a screen — actually, cross sections of brains and lungs. There are 30 or 40 scan slides per lung or brain. The radiologists receive no less than 30 to 40 patients per day. On the slides they are looking for spots, such as a small blood clot in the brain. It’s important that they do this well, because we’re talking about someone’s life. In the hospital, they have created an auxiliary tool which can perform machine learning. Thousands of scans on which radiologists have circled the spots have been collected and entered into the computer. Now the computer is giving back, as it were, on the basis of the input. This feedback helps the radiologists by quickly recognising the spots and circling them already. Now a radiologist needs only to personally check whether the computer has missed anything and to interpret those spots. So, in that sense, AI is an additional tool — one which is more consistent in circling spots than a human is, but which depends on input. The system learns from the best doctors who have entered their knowledge into the system.’  

Deviations and patterns
In a similar manner, there is also a role for artificial intelligence in finance. ‘Humans are not made to scroll endlessly through spreadsheets. At a certain point, your eyes no longer see what you’re looking at. Artificial intelligence — or rather, machine learning — can save time for financial employees by extracting deviations or patterns from the data. But the responsibility always remains with people — in the above-mentioned cases, the radiologist or the CFO. The computer does not take decisions. At most, it gives signals, on the basis of which a finance person can set to work.’ But there must be data, Stolze considers. ‘If your information management is not up to par or if you are collecting too little data, then you won’t be able to take advantage of machine learning. After all, garbage in is still garbage out. As CFO, you will have to set up a data model. This is actually data governance, and that concerns questions like: How do I collect my data? Which sources do I use? How do we handle our data? How do we combine sources? How do we monitor the data quality? These are aspects which are the responsibility of a CFO, certainly now that organisations are becoming exclusively digital. You must have a good perspective of this.’ 

Clustering and classifying
Those who have their data source or sources in order can then start playing, according to Stolze. With cluster algorithms, among other things. ‘Using a set of various data points, data scientists can use a cluster algorithm to categorise and classify each data point in a certain group. The data points in the same group contain comparable attributes or properties. On the other hand, data points in separate groups have very unique attributes or properties.’ Stolze gives the example of Salesforce, who has just launched its AI module, Einstein. ‘Just as is done with the thousands of scans which radiologists must read, this system looks at historical data about paying customers. On the basis of that data, this system can indicate the point at which customers are inclined to cancel their subscription. It gives suggestions for this: It may be that these customers want to stop using our services in the coming month, because they have the same attributes as customers who have previously cancelled. As a company, you can intervene so that you don’t lose those customers. The list of customers can also be created in order of priority, so that you can approach the largest customer first.’

Earn money via the digital path
It’s the million-dollar question among CFOs: In the digital transformation of the entire organisation, must the CFO be leading or should he or she limit themselves to the finance department? Stolze replies, ‘Many of my friends are CFOs. I might not make any friends with this statement, but they are mainly interested in the control in an organisation. They are often also responsible for ICT. But the provision of a laptop is something different than the question “How will we be earning our money in five years?” The first responsibility is more a process that must run well. The second one is much more about relinquishing control. Giving your people permission to conduct experiments, of which the result is uncertain.’ 

Learn to understand
What’s more, a data model must be developed for the organisation. ‘The CFO is perfectly suited to establish that model and to monitor its quality, even from the viewpoint of management control. Besides, the CFO is also the conscience of the organisation and is responsible for compliance. In that sense, he or she can say what is needed about handling data, make the link to the GDPR, and make sure the algorithm doesn’t work incorrectly and discriminate.’ In doing this, the CFO must want to understand technology like the algorithm and machine learning. ‘Once, a CFO took my course in artificial intelligence but did it in 12 hours, not in the three to four hours allotted. When I saw that, I called him. “Jim”, he explained. “This is so crucial for my organisation that I want to understand it to the last detail.” He had ploughed through every link, every bit of text, every video minute by minute. Because he didn’t want to join the conversation but rather direct the conversation. I think that this is the crux for CFOs and their financial people. Flip the switch. Don’t be afraid of technology. Because... If you don’t know where you’re going, you’d better be flexible.’

Go back
Related news
Interview Michael Wade
Interview Michael Wade 13-07-2021

Michael Wade, Professor of Innovation and Strategy IMD Business School

'Finance needs to understand the digitized business model'

Digital transformation is not about technology or ICT, but is actually a matter of adapting the business model by making use of digitization. 'That's where the real added value comes from,' says Michael Wade, Professor of Innovation and Strategy at the prestigious IMD Business School.