Rational discussion of the future of work is threatened by popularization of the term Artificial Intelligence. We need a new, more accurate phrase, that doesn’t conjure up images of sentient robots, but describes the statistical tools we use today to make complex decisions. We haven’t created “Artificial Intelligence”. We have built “decision machines”.
There are two types of Artificial Intelligence (AI): one imaginary, and one real. The imaginary one is “General AI“, a theoretical type of software that could exhibit intelligence indistinguishable from human intelligence. The invention of General AI would either be amazing, or horrifying, depending on who you ask. (Hint: don’t ask Elon Musk.)
General AI has a potent association with the term “Artificial Intelligence”. When we hear the phrase, it feels like General AI is around the corner. In fact, we have no idea whether or when it will happen. Furthermore, the software tools called “AI” today are not the first step towards General AI. They are new tools in and of themselves. “Artificial Intelligence” should only refer to “General AI”, not the real software tools we are building today.
The Real Artificial Intelligence
The real software called “AI” today has nothing to do with sentient humanoid robots. It is confusingly called “Narrow AI“, and is software that:
- Makes a decision, rapidly, and repeatedly
- Tries to make the “best” decision within certain limitations
- Constantly collects data to make the decision
- “Learns” how to make a “better” decision in the future
To illustrate these concepts, let’s examine the Narrow AI software inside a “smart” thermostat for your home. First, the software inside the thermostat makes a decision: whether to turn your heater on or off. The software is programmed to make this decision over and over again, every hour, or every minute of the day.
Next, the Narrow AI tries to make the “best” decision within limits defined by its human creators and users. Let’s say the “best” heating decision uses the least amount of energy, and has no more than two manual adjustments per day. If the software turns off the heat too much, users will turn up the heat manually. However, leaving the heater on too long wastes energy. The Narrow AI makes the “best” decision by adjusting the heat over time to minimize manual adjustments and energy use.
Third, the software collects data about your home’s temperature, energy use, and how often you change the temperature manually. The software uses these variables to determine the probability that turning the heater on is the “best” decision.
Finally, the system “learns” by calculating the probability a decision is “good”, and adjusting over time. For example, imagine the software estimates a 95% probability that you won’t interfere if it turns off the heater at 7PM. However, five days in a row you turn up the temperature at 7PM. On day six, the software will now estimate say, a 5% chance that you won’t interfere. It has “learned” what to do at 7PM and makes a decision to leave the heat on instead. Narrow AI’s “learn” constantly by comparing their estimate of the probability that something will happen, with the actual number of times something happened.
Smart thermostats, like all other Narrow AI tools, are not a new and novel type of intelligence. Narrow AI tools are really just decision machines created by humans to make decisions rapidly and repeatedly.
Humans and Decision Machines
“Decision machines” is a better term than “Artificial Intelligence”, because Narrow AI doesn’t actually think and learn like a human being. Human intelligence has: intent, independence of thought, self-actualization, and will. Decision machines lack all of these characteristics:
- There is no “intent” behind the actions of a decision machine, only a human creator who intends to make their life better by having a tool to make repetitive decisions.
- There is no “independence” of thought, only calculations of probabilities over and over to make a decision with the highest probability of being “good”.
- There is no “self-actualization”, only a human that pushes the start button.
- There is no “will”, only electricity powering the software until a human pulls the plug.
If we use the term “Artificial Intelligence”, we inaccurately imply that a new form of intelligence actually exists, and will eventually become a General AI by becoming “more intelligent” over time. On the contrary, decision machines only get better at one thing…making the decision they were programmed to make! As with all machines, they will get better and faster over time. However, they will not, suddenly, transform into General AI.
Replacing Decisions, Not Humans
Decision machines help humans, they don’t replace them. Imagine the human alternative to the example above: to compete with the thermostat decision machine, you would have check the weather online, the temperature in your house, and stand near the thermostat all day to make manual adjustments. It’s not that it is impossible for you to do this job. It’s that the job is so boring and monotonous that you would never do it!
Decision machines are a powerful invention. They follow on the heels of other human inventions like mechanical machines to move objects, and silicon machines to store information. There are thousands of decisions that people make every day that a decision machine can help with. Today, decision machines are helping doctors to examine medical images to ensure human doctors don’t miss critical information. Tomorrow, decision machines may help us drive cars, by making decisions about the speed and distance our vehicle should travel in relation to other cars.
We should all want a future where we own and control these powerful tools to help us make better decisions. But, we need to stop calling them “Artificial Intelligence”. We are not heading towards a world of sentient robots. We are headed towards a world where machines help us make decisions, rapidly, and repeatedly, using the latest data available.
Let’s call these powerful tools decision machines.