#human work

 Why AI does and why it’s useful

Recent advances in AI are best thought of as a drop in the cost of prediction. Prediction is about using data that you have to generate data that you don’t have. We do that translating large amounts of data into small, manageable amounts. A classic prediction problem is using images divided into parts to detect whether or not the image contains a human face.

Prediction is useful because it helps improve decisions. Other important input is judgment. A good example is a credit card network deciding whether or not to approve each attempted transaction. The idea is to allow legitimate transactions and decline fraud and thy use AI to predict whether each attempted transaction is fraudulent.

However, even the best AIs aren’t 100 % reliable and the people who have run the credit card networks know from experience that there is a concession between detecting every case of fraud and inconveniencing the user. This means that to decide whether to approve a transaction, the credit card network has to know the cost of mistakes. It would be worse to decline a legitimate transaction or to allow a fraudulent transaction? No AI can make that judge, just humans. This decision is what we call judgment.

 What judgment entails

Judgment is the process of determining what the reward to a particular action is in a particular environment. It is how we work out the benefits and costs of different decisions in different situations.

Credit card fraud is an easy decision to explain in this regard. Judgment involves determining how much money is lost in a fraudulent transaction, how furious a legitimate customer will be when a transaction is declined, as well as the reward for doing the right thing and allowing good transactions and declining bad ones. Humans learn by experience, observing their mistakes.

They will specialize in judging the costs and benefits of different decisions, and then that judgment will be combined with machine-generated predictions to make decisions.

 Setting the right rewards

It is true that AIs can also learn from experience and a good technique is reinforcement learning whereby a computer is trained to take actions that maximize a certain reward function. For instance, DeepMind’s AlphaGo was trained this way to maximize its chances of winning the game of Go. Games are often easy to apply this method of learning because the reward can be easily described and programmed – shutting out a human from the loop.

In fact, even if an organization is enabling AI to make certain decisions, getting the payoffs right for the organization as a whole requires an understanding of how the machines make those decisions. There are some types of prediction mistakes more likely and a machine can learn the wrong message.