AI is on everyone’s lips today, but so far it’s mostly about different types of decision support. The next step is to automate transaction management, IBM’s Bill Lobig says.

When Pedab’s partner Apendo invited IBM’s AI expert William “Bill” Lobig to a recent customer event, we at Pedab took the opportunity to have a chat. Bill Lobig, who has a down-to-earth approach to everything regarding AI, automation, data analysis and other things that fall under the concept of digitalization, begins by explaining what AI really is:

"AI makes people’s lives easier", says Bill Lobig, who has the title of Vice President Digital Business Automation at IBM.

Now we know that. But what’s happening on the AI front today?

– A lot is going on, but it’s mostly about different types of decision support. We’re working hard to build solutions for transaction management, using machine learning, Bill Lobig says.

Machine learning is currently the most widely used AI technology. The purpose is that computers learn things from data sets, without specifically having been programmed for how to learn. An application example is that self-driving cars identify objects, and people, in traffic, based on runs made with machine learning.

There are major differences between different types of AI solutions, in terms of use. It’s one thing that an administrator at a bank receives mechanically produced advice regarding the granting of a loan. It’s something completely different that a computer actually grants a loan, without the involvement of a human being.

The latter is a concrete example of automatization, or automation, which is quite possible today. But these types of solutions have not yet begun to be used widely, despite the obvious savings and the efficiency improvements they would entail.

Barriers to automation include cultural aspects, aversion to change and, perhaps above all, a lack of confidence in the solutions. On top of that, some individuals may be afraid to lose their jobs and be replaced by AI solutions. Last, but not least, there may of course also be regulatory barriers.

If we assume that automation is desirable, how do we get past the barriers?

– An important factor is that the AI solutions should be self-explanatory. It’s not enough to say that an AI solution does not grant a loan, an administrator must be able to tell us why the loan is not granted. This is something we work a lot on at IBM.

Another barrier to successful data solutions is the availability and quality of the data required. The boundaries between disciplines such as machine learning, “data science”, data analysis, business process management and others are often blurred. If you want, you can see them as different variants of the same thing, which is using data to get answers to questions.

Data sets may express prejudice

One problem is that data sets may express prejudice. There are examples of AI recruitment solutions that conclude that white, middle-aged men born in the country in question should be employed. Why? Because that type of employee is the majority of the data base.

Is it possible to have completely neutral, unbiased data sets?

– There’s always bias in the data. It’s important to understand the bias and how it can affect the AI solution results.

So you have to live with bias. However, if you understand how the nature of data affects analyses, you must not only understand how the results are affected, but also replace data sets that give far too skewed results.

These are challenges that emerge for companies and organizations that have started with AI and data analysis. But before then, there is probably an even greater challenge: to find, identify, classify, wash, transmit and manage data.

In short, it doesn’t matter much if there is an understanding of AI and data analysis solutions, if you don’t have your data in good order.

"The best way to get started with AI is to identify projects that can deliver quick profits"

Bill Lobig emphasizes that the best way to get started with AI is to identify projects that can deliver quick profits, as so often is the case when it comes to new technology. For IBM, it’s often about helping customers with document management, which is natural as IBM has a wide range of more traditional products in that area.

– It’s about extracting, enriching and classifying information. We add metadata that facilitates the handling of documents.

This is a time-consuming task if performed manually. And in many cases it’s not possible to ignore it, because of legal requirements and other types of rules. So there is a great potential for efficiency here.

But Bill Lobig’s big goal goes beyond that:

– We work to 100 percent on using machine learning for transaction management. The technology exists, but there are major cultural challenges and also challenges in how to implement the solutions.

Transaction management

In IT context, a transaction is a series of activities that all must be performed, or not at all. A typical example is a transfer between two bank accounts. If there’s an interruption after the money has been withdrawn from the first account and before it’s deposited into the other account, the entire transaction will be cancelled. The money is therefore returned to the first account. A transaction doesn’t really have to include money, but in everyday speech the term usually refers to the handling of money, which is also the traditional meaning. The concept of transaction management often refers to technology solutions for handling transactions of various kinds.