AI tells you what it suspects you want to hear

If we changed Artificial Intelligence making into Automated Decision-Making, we would communicate far more about the aims of the AI movement than we currently do.

At the core of AI is the development of machine learning techniques to enable computers to automatically collect, synthesize, suggest, and sometimes, implement solutions. Call this intelligence if you want but I call it decision-making. It is not necessary for something to be intelligent to make a decision…

Let’s dive deeper into what I mean.

Assume you are at time period 0 in the chart below.

The stylized chart shows that each increment of time, you receive a new bundle of data. In real life, this is near continuous but let’s establish it as a periodic input of new information. Some of this information is repeated information, some of it is new, some of it is important to a current decision, some of it is important to a previous decision, some of it is important (but you don’t know it) to a future decision.

You examine the data in period 0, update your assumptions, consider your options and then make your decisions.

One of the key techniques underpinning AI is the ability to rapidly consider multiple previous data points to understand ‘what has happened in the past?’. Think of this like the analysis of millions of medical records to understand what the link between a particular set of symptoms and a particular disease.

This analysis supports the diagnosis decision in that it manages the issue of human memory bias. People remember things in different subjective ways but the AI dispassionately, at least from its point of view, reviews the history. Add the computing capacity to review millions of records in seconds and you have a powerful decision making support system. Think IBM’s Watson. This enables you to make a a historical data rich decision at time period 0.

Another technique is to build a simulation which takes a set of broad objectives and then feeds in many different future data scenarios. The computer then considers, based on the objectives, how it would respond.

The question that we are asking is ‘how might I react to new data?’ This then enables me to consider my current decisions in that light. It is not that AI is telling me how to react but it is telling me what decisions I might make and some of the implications. There actually few AI simulation applications available based on a Google search but it is this simulation that is the next area of exploration. Programs like AlphaGo run simulations in real time to game out Go.

So where does that leave the ability for humans for making decisions. It might look like we are becoming redundant in decision making and AI should use forward looking and backward looking techniques to arrive at the best decision.

But it is precisely that AI generates this information based on the techniques that we give them which makes the predicted demise of human decision making a lie.

Firstly, if we remove all of the decisions that only require historical data, we are still left with a plethora of decisions that require other inputs. If we remove the scenario planning done by simulation on things that we know, we are only left with the vast number of things that we don’t know we don’t know.

AI is effectively automating the decisions that we are already confident enough to make. It gives us the space to think about the decisions that we are not confident to make.