ICAEW wrestles with artificial intelligence

Artificial Intelligence will be looking for what is right and then be completely satisfied when it finds that. I think this might be its biggest weakness.

A typical example would be those who might submit Tax Returns based on what the system wants to see rather than what the facts are. The anticipated 'nudges and hints' widely promoted as a feature of Making Tax Digital are simply making it easy for those unscrupulous people to provide just what is expected. Any artificial intelligence must really be able to think like we do, illogically, and as we are all different, I think that might be difficult to achieve.

Why not ditch the AI and just rely on humans?

"However, if accountants take the time to understand the capabilities of AI systems and use them to supplement rather than replace human intelligence, the profession could enhance its underlying purpose to solve problems and support better business decisions."

Brexit is a classic example of where many very successful business people had completely opposite views on leaving the European Union. This surely proves that there is no 'right' way to make business decisions and those that use AI to make decisions may well be outwitted by those that don't.

An interesting subject, nonetheless and I am confident that AI will not take over the World but I can see that humans with the power that AI could give them, just might.

the idea of that lot 'wrestling' with AI fills me with glee

AI Explainability From the Analytics Perspective

I recently asked a start up software company to explain to me how AI works in the context of their CXM model they were offering, indeed how did the machine learning and NSP programming come up with the predictions they were so excitedly telling me about. In this particular case, it turns out that very little AI is actually performed. I wonder in general if there is more hype than substance.

In reading an article in www.datafloq.com it would seem it’s OK not to be able to explain how answers are derived. The writer goes on to say “From a purely analytical perspective, not being able to explain an AI model doesn’t matter in all cases. The issue of explainability is very similar to the classic problem of multicollinearity within a regression model. I recall having drilled into my head in graduate school the distinction between (1) Prediction and (2) Point Estimation.”

If the main goal of a model is to understand which factors influence an outcome and to what extent, then multicollinearity is devastating. The variables that are inter-correlated will have very unstable individual parameter estimates even when the model’s predictions are consistent and accurate. Conceptually, the correlated variables are almost randomly assigned importance. Running the model on one subset of data can lead to very different parameter estimates from another subset. Obviously, this is not good and we spent a lot of time learning how to handle such data to get an accurate answer and also be able to explain it. The point is that multicollinearity made it very hard to pinpoint the drivers of the models, even if the models were extremely accurate.

The writer goes on to say “You may have an AI process that is performing amazingly well. However, accurately teasing out what factors are driving that performance is difficult. As I’ll discuss later, there is work being done to help address this. But, AI models leave one in a similar spot as multicollinearity did. Namely, a great set of predictions whose root drivers can’t be well explained and specified.

Well, it strikes me that the old phrase “just trust me I’m a doctor” springs to mind. I wonder if anyone has taken a hard look at what some accounting software is doing in this area as it seems that if financial manipulation can’t be explained then for sure one day it might come to pass that the AI may not be quite as smart as we think. An accountant who ran a big financial brokerage out of New York, 9/11 caused him to return to the UK, told me that any trader wishing to introduce a new derivative or option deal has to pass the “bowl of fruit” test, if you can’t explain the deal and the rationale for doing it using apples and oranges then we don’t do it. Seems like AI may have to go through a similar process.

Good morning Dave.

Most of us are not "big business" accountants and hence how AI is going to affect us is totally different to how it will affect the larger firms where there is far better control of what does and does not go through a company's records (on a basic level)

Yesterday as an example a client who has adopted cloud accounting with us (and wanted to get involved himself) sent up his records for us to quickly review and finalise the quarter.

End result - two screens on the go - screen one to find the clients attempted input, screen two to repost correctly and then delete the original client attempt. A much longer exercise.

Bank feed gave a reconciled account balance on the current account but the client had assumed the system knew what it was doing so the analysis was a nonsense.

Moving forward therefore we are now doing everything on the quarterly basis - the client wants to be able to see and use accurate data and guess what, that's where we come in more now than we ever were before.

An accounting system, be it desktop, cloud, a spreadsheet or a simplex book do not suddenly turn someone into an accountant but the new technology do allow us to do complicated things quicker and that is simply because WE know what we are doing with these powerful tools.

Just don't give me a power drill and a box of screws......

One of the biggest problems with getting computers to do a humans job is the interpretation of the rules used to do said job.

I'm going to show my geekiness here as an example...

Dungeons and Dragons has, since the 1970s, been a pen and paper affair. As computers entered our life, people have developed programs to try and simulate the role of the players to provide tools to play online. Every new book release introduced new challenges for the software and ultimately they failed to fulfil the roll they where designed for.

In order to create a computer version of the game that kept the freedom of its intent they had to redesign the rules from the ground up, sacrificing some of its flexibility to achieve a fully automated version of the game.

The Accounting Ruleset has much the same issues. With every new business, new venture and new technology, the rules must be adjudicated by humans to ascertain an acceptable result. Computers must be continuously updated to cope with this ever changing medium.

Now to make it easier we could redesign the accounting system from scratch with AI/Machine Learning in mind. But that would be a task so huge that by the time it was finished it would be out of date.

Youngloch's example is a good one. It highlights the one thing that has attracted me to the cloud and MTD more than anything. The ability to connect with my clients on a much more personal level!

I too have been working with clients to convert them to cloud software and I can say, with hand on heart, that it has been the most rewarding experience of my career.

To have clients openly thank and praise me for everything I have done to make their life easier and make their business more successful, is something I intend to pursue.