There’s not a time in our lives we’re not taking decisions. It’s part of our daily routine: deciding which clothes to wear, which food to eat, which comments to say to ourselves and others, etc. Sometimes, we barely realize we even decided the first place. Interestingly enough, Mark Manson points out that refusing to make a decision altogether is a decision in and of itself.
From first glance, taking decisions may seem harmless. After all, the scale and magnitude of the results are usually limited. For example, what I eat for breakfast doesn’t affect another person.
However, not all of our decisions happen in isolation. Some of them might not only affect our lives but the lives of others. Maybe deciding what to eat for breakfast is harmless, but supporting a particular tax bill could be.
For example, the decisions taken by health institutions greatly impact the lives of those they serve. Unfortunately, this means that a couple of missteps could result in the deaths of thousands of patients.
So, how can we then get the best result from our decisions?
Gathering enough evidence is key in critical decision-making. For health institutions, it guarantees their actions taken during volatile times come from a reliable reflection of a well-understood reality based on experience. After all, the more evidence we incorporate in our decisions, the better they scale to a larger number of people. Let’s look at this in a simple diagram:
Let’s imagine we need to determine whether a particular city presents any significant signs of infection of a particular disease. Should we limit our research to only look at these 3 individuals, we could conclude there’s no risk of infection. However, if we increase our sample size, the result may be different:
Even if this example greatly simplifies the methods used in the decision making of institutions, it illustrates something very important: making decisions with evidence is way better than making decisions with no evidence at all. However, even this is not the complete picture in data gathering.
While gathering more data is a good practice for improving our decisions, it’s far from the only one. If we gather a lot of data, but the data tells us very little about what we want to look at, the data is not useful. This occurs not only in health institutions but also in Artificial Intelligence.
Gathering Relevant Evidence
Robots can perform tasks as long as they have enough sensory information. The scope of this information is limited to technological developments, and the costs involved. For example, an infrared sensor is cheap, but can only detect if an object is close or not. On the other hand, depth cameras, while expensive can tell you what is close, and the distance of the close object. However, device cost is not as relevant as the processing cost of the data.
If your data has many variables, each with their ranges of values, the time it takes to process it is greater than if your data is only limited to one variable, and a limited set of values. This increases the time between the user’s request and the response it gets back. While response time is not a huge issue when loading things such as videos and audio, it is critical with real-time and dangerous environments such as those of autonomous vehicles. In summary, a compromise needs to be taken between how many variables to consider and how much time do you want to delay response to your user.
Juggling with the before mentioned compromise can result in different sources of error, which include:
- Tools and devices used to gather data (sensors, cameras, etc.)
- Level of depth or detail of data (granular or high-level)
- Scope of the data (place and time)
- Format the data is saved (spreadsheets, lists, etc.)
The main limitation of data gathering, in this case, is complexity. Fortunately, the errors from this limitation can be measured using statistical tools like the correlation between an input and its output, or calculating the variance of the results in our models. The errors from this limitation can be corrected by taking more samples, increasing sample size, or changing which variables to consider in our decisions.
Back to the breakfast example, it may not be worth making a research paper on the matter. However, a simple Google search will show you that eating donuts at 10 AM is far from the healthiest option you can pair with your morning coffee. Luckily, this decision only affects us.
