Are you worried about Artificial Intelligence? When Wired magazine publishes an article with the title: “Robots Will Soon Do Your Taxes. Bye-Bye, Accounting Jobs.” it’s easy to become unsettled. Is it true? Are the machines coming to take our jobs? Well, yes. Sort of.
Artificial Intelligence is growing by leaps and bounds, to be sure. But what does that mean? Are we all going to be out of work in the next 5 years? 10 years? Without getting technical, let’s look at where we are with computer cognition. Then we can see how far away Skynet might be.
Does Artificial Intelligence Mean Computers Can Actually Think?
There are many things encompassed by the term “Artificial Intelligence”. Most often it means a set of algorithms (instructions) that help computers make better decisions over time. They work pretty well on certain tasks, but calling them intelligent is a bit of an exaggeration. No matter how sophisticated the algorithm, the computer isn’t really thinking. Can this technology do taxes? It can follow instructions, even a set of complex ones, to make sure forms get filled out right. But, explaining the risks and benefits of different financial strategies would be difficult. They are not without limits.
The thing about machine learning is that it can only learn what it’s shown. It takes a lot of training data to get it to do “the right thing.” To be fair, there is an awful lot of financial data available. Doing taxes are something machines could be good at – at least to a point. If you have a unique case the machines will struggle. This is why computers don’t make good consultants. They have a hard time transferring an idea from one place to another situation or field. They cannot intuit. Well, at least until recently.
From Algorithms to Neural Networks
Deep Blue was a chess-playing computer built by IBM in the 1990s. When Deep Blue beat chess master Kasparov in 1996, it did so by simulating every possible move. It used these simulations to find the moves that resulted in the most “best outcomes.” Computers can run simulations faster than humans can, so for those types of tasks they can beat us. That doesn’t work with a game like Go.
Go is an older game with only two pieces – white pebbles and black pebbles played on a 19 x 19 grid of squares. While the game looks simple, it’s actually more complex than chess – by a lot. The number of legal positions on a standard 19 x 19 Go board is 171 digits long. It’s more than chess by a factor of about 7.
Today’s supercomputers would take several months to simulate every position. So when Google built its Go-playing robot (AlphaGo) it took a different approach. Google used what is know as a Deep Neural Network to help AlphaGo “Learn”. So what is a Deep Neural Network? It’s kind of like a virtual model of your brain.
Think of learning as a game of whack-a-mole. As a baby, you don’t know anything. Over time you learn by trying a bunch of things, and when you get the desired result you notice. If that happens over and over again, you learn the pattern it took to get what you wanted. The important thing here is that in the beginning, you took a bunch of guesses. It’s like playing whack-a-mole. You smack at everything until you hit a mole – you get something right. Eventually, you learn to stop hitting at all the spaces and wait for the mole and then smack it. That’s (kind of) how neural networks work.
In a Deep Neural Network, the computer makes a bunch of “neurons” and then feed information to them. The “neurons” randomly connect and spits out an answer. Then we tell it yes or no – it was right or it wasn’t. This process repeats until it starts getting a bunch of yeses. Basically, it remembers what order of “neurons” to send the information through. It appears to be learning, and in some sense it is.
Google fed AlphaGo a bunch of scenarios from games of Go, asked it to decide what to do, then it either won or it lost. Over time, it started to win. Last year it got good enough to beat professional human Go players. This shocked the world as this wasn’t supposed to be possible. The only way to do that was for the computer to have intuition. It could not simulate all the possibilities to find the right thing to do. It relied on its experience of playing lots of Go games to decided what “felt” like the right thing to do. Let that sink in for a minute.
How Much Can Artificial Intelligence Actually Know?
Now before anyone freaks out, it’s important to point out a few things. AlphaGo is a “Narrow” Artificial intelligence. That means it can only play Go. It can’t drive your car, do your taxes or even make a latte. It still suffers from the same problems of older AI’s. It can only learn what it’s exposed too. The way it learns is pretty amazing because at some level it mirrors how the human brain learns. But that is also the problem.
Think about everything a baby has to learn growing up. How to understand what it sees, how to move, talk, interact with objects and other free thinking beings. That is a lot to take in. But unlike computers, humans can deal with all those different inputs. Humans have the advantage of being able to transfer ideas between situations. Even with that advantage, it takes a long time to learn how to do taxes. First, you have to understand what taxes are. That involves a basic idea on what business, the economy, and other social constructs are. Not to mention the mountains of regulations that, let’s face it, don’t always make logical sense. To get a computer to understand all that would take a huge dataset. If the robots are coming, they ain’t coming tomorrow.
Having said that, the rate of improvement in AI and machine learning is growing rapidly. There will be a day, probably in our or at least our children’s lifetime when AI does taxes. H&R Block is already using IBM’s Watson. The Nest thermostat, Siri, Amazon’s Alexa, hands-free car entry key fobs and countless software wizards are all forms of smart technology. In some ways, AI is already here. But that shouldn’t worry you too much. Research shows it’s most efficient and accurate to have machines and humans working together. We each have our strengths. Computers can think fast and consider vast data sets while can think creatively and conceptually. The perfect solution is to use both. Every example of existing smart technology listed above is a tool used by people to make them more efficient, not to replace them. As AI continues to make its way into our world, it should free us up from the doing what a computer does well so we can spend more time doing what humans do well. For service based companies, that would be providing a better service.
Can Artificial Intelligence Really Meet Customer Needs?
I think the best thing to keep in mind is that no one really hires anyone to do their taxes or to track their expenses. We hire people to help us achieve a goal and provide some peace of mind. That’s the real service people pay for. I don’t care what my accountant and bookkeeper uses to keep track of my expenses. It doesn’t matter to me if they use QuickBooks Online, QuickBooks Desktop, a Green Columnar Pad or an abacus. If it gets the job done, who am I to complain? What I care about is that they understand my unique needs and desires. I want them to tell me the best course of action. I care that the work gets done, and done in a way that gets me where I want to go. How they do that, I don’t care as long as its ethical, legal and I’m not the one doing it.
ClockShark is the #1 time tracking and scheduling app for construction and field service companies! To learn more check out the video below. To get a free 14-day trial visit www.clockshark.com