They frequently make headlines, but the differences between them are often overlooked or misunderstood
In the pursuit of simplicity, I wrote this blog post to alleviate confusion between machine learning and artificial intelligence, specifically with regard to their use and application in real estate. If you’re looking for the 10-second version:
Machine learning is the computer learning how to make decisions with human guidance; artificial intelligence is the computer learning how to make decisions on its own.
Before we start, I think it’s important to define a commonly used term that spreads across computer science, statistics, machine learning, and artificial intelligence, and that term is “algorithm.” The term has a variety of different interpretations; however, for this post, let’s use the following definition to be as concise as possible: Simply put, an algorithm is a procedure for solving a problem or accomplishing some goal. Put another way, an algorithm is a set of rules a computer follows to achieve a particular goal.
Machine learning involves training an algorithm. What does that mean? In short, it’s feeding large amounts of data to an algorithm so it can adjust and improve. An easy way to understand machine learning is by thinking about it in the context of computer vision, which is the ability for a computer to “see” or identify specific objects that a human would be able to identify.
Let’s walk through an example. Say you wanted to know whether a photo of a kitchen contained stainless steel appliances. To train your computer vision algorithm, you could gather a large amount of photos, use a group of experts to tag a small sample of the photos based on whether or not they contained stainless steel appliances, then feed those photos to the algorithm. Using those photos, the algorithm would learn to understand what stainless steel appliances look like. From there, the learned algorithm could determine whether or not a new photo — a photo it hasn’t seen before — contains stainless steel appliances.
Notice the key here is that the machine needed some sort of human guidance — in this case, the tagged photos — so that it could learn what pattern to identify. Practitioners of machine learning would refer to this as “supervised learning.”
Artificial intelligence is the ability for a machine to make a decision without human guidance — in a sense, the machine learns on its own through repeat patterns of success or failure in a situation. I like to think of this as the machine learning a decision system or framework so further action can be taken.
Consider the above machine learning example. The machine is essentially learning repeat patterns in order to produce a “label” for stainless steel appliances relative to other examples of stainless steel appliances. It isn’t making a decision on how to further use that information.
This is the distinction between artificial intelligence and machine learning. Artificial intelligence systems can make their own decisions given the input of information like stainless steel appliance labels, condition labels, view descriptions, property characteristics, and more. This complex decision generally has a risk/reward tradeoff associated with the system.
Let’s walk through another example. Say we need to value a property. Should we order an automated valuation model (AVM), a broker price opinion (BPO), a conventional appraisal, or some other valuation product? On the surface, this can seem like a simple question to answer. But in aggregate, across many properties, this can become a challenging question to answer, especially when adding those multiple valuation products into the mix.
Another main differentiator with artificial intelligence systems is that there’s generally an element of attempted inference involved in the decision system that is still typically solved by a human. This is where the bulk of work needs to happen in the field of artificial intelligence. Machines are typically good at solving problems when information is relatively clear and quantifiable — they aren’t so great at solving problems when information has a vast complexity of indirect implications. In other words, machines are not great at reasoning with subjectivity or subjective information.
Artificial intelligence is commonly mis-marketed to give the impression that it has very broad applications that can solve a variety of problems. In my opinion, this is a grandiose representation. The instances where we can be confident artificial intelligence will work a majority of the time — including when risk and consequence are involved — are more narrow than portrayed. That said, there are still many possibilities for change that will positively impact our industry and enable experts to save time in their day-to-day processes, especially with access to new information in value determination and automated rule checks.
So, with regard to real estate valuation, how can we answer the question, “should machine learning or artificial intelligence solve my problem?” Think about the level of complexity and subjectivity in the information that would be required for you to solve the problem yourself. If you don’t think it would take a long time to reason with the information and make an informed decision, the problem might be a good candidate for a machine learning application. Our team would be happy to talk through it with you — we may have already built a solution that can solve your problem.