I have been thinking lately about how artificial intelligence (AI) is affecting me. That is a big topic with many avenues to be explored. I would like to explore just two of them here. The first is how AI influences my concept of reality. The second is how AI is influencing my actions. Both of these dynamics are profound and already evident in our world.
In this first of two parts, we will explore how AI influences our concept of reality.
AI-driven systems receive data from a variety of sources. Their universe is defined by what their sensors can quantify and deliver to them. As we interact with these AI systems, we are subtly influenced to adopt their view of reality. The real world is only what can be quantified and made available to an AI engine.
If I go to my doctor and he runs a number of tests, the assumption is that those tests contain all the important variables affecting my health. It is likely that there is some AI involved in processing the samples and providing my doctor with the results. There is a strong influence encouraging me and my doctor to believe that everything important has been measured and reported in those tests. However, there are always many more tests that could have been run or the same tests could have used different methods and possibly arrived at different results. That makes us realize that there is the world of data that is quantified and made available but there is a much larger world that also can be quantified but is not currently available to an AI engine, my doctor, or myself. We have at least two worlds, the one defined by the data that is available and the one that is not readily available.
All sensors have limits and imperfections. The numbers they report are bound by those limitations with the result that they distort the perception of reality an AI-controlled system has, and to the degree we rely on these systems, the perception of reality we have.
I recently experienced this with a sleep sensor I wore as part of a sleep study. The sensor reported its data using Bluetooth Low Energy, a very error prone protocol. The sensor sent its data to an app on my phone, which then forwarded it to a server that presented the analyzed results for my doctor. I noticed on the app that whenever the Bluetooth packets were not received the system didn’t log any information for that time. It might have used a protocol that confirmed receipt and resent packets that were not received successfully, but it did not do this. What then happened is that the analysis software assumed that I was not breathing during all these missed packets. The result is that I was diagnosed with severe sleep apnea, in urgent need of expensive treatment. A retest using different equipment gave a more realistic picture. The problem was with the software, not my body. The software assumed that no data meant I had stopped breathing. It could have made other assumptions or, even better, requested that the data be resent. This is just one example of the many ways that data can be distorted. When readings are too high or too low sensors may report their own noise floor or start to behave in a non-linear fashion. The characteristics of every sensor will impact the data it provides. Our perception of the reality it is reporting is distorted by the characteristics and limits of the sensor.
The data we get from sensors is, at best, distorted, but what about the data an AI system does not get? The implicit assumption is that an AI engine gets the data it needs to perform its function. However, there is a lot of other data that is quantifiable and might be sent to an AI engine but isn’t.
I have both a Fitbit and an Apple Watch. When I was doing that sleep study, I might have also worn one of these and that data would have been a check on the sleep monitor I was wearing. However, the system was not designed to take in data from either a Fitbit or an Apple Watch. The system could have used the data from one of these other devices to confirm the data being sent by the sleep monitor. However, the system designers did not design it to receive and include that data in its analysis. The AI engine controlling the analysis assumed that it had a complete data set, but it did not. There was other quantified data that existed but was not available to the system.
Going further, there is unquantifiable data. We can start with the state of relationships. If I have a big fight with my wife just before going to bed, I am unlikely to sleep as well as when we are in a better frame of mind. We all know that conflict in important relationships can affect our sleep. The state of relationships is difficult, if not impossible, to quantify but it has an impact.
Beyond the realm of relationships there exists what I will call the spiritual realm. This is the realm where our worldview operates and gives rise to our values and ethics. How consistent are we to express our understanding of ultimate realities in our habits, actions, and decisions? As AI-driven systems focus us on the quantified world available to them, they draw our attention away from the unquantifiable realms of relationships and spiritual values. Our challenge is to look at the full reality and perhaps most importantly the unquantified aspects of reality and resist the influence that draws us into a fantasy that reality is only a subset of what can be quantified.
AI-controlled systems are already performing many valuable tasks and their increasing role in our lives seems to be certain. We will receive the benefits of these systems but need to also develop the skills to protect ourselves from their imperfections. AI systems influence our perception of reality by simply focusing us on the world as they perceive it. Our task is to continually refocus on the full reality of life and resist the temptation to buy into the AI’s implied message that reality is limited to the quantified data it receives.
The universe is and always will be larger than any AI-driven system can perceive. There will always be quantifiable data that is not available to any particular AI system. Further, there are the realms of relationships, ethics, values, and the worldview that births them. Our challenge is to constantly reorient ourselves away from AI’s understanding of the world, and toward the more important aspects of reality. To live successful lives, we must renew our focus again and again on relationships and living consistently with our values.