Algorithms and Experts

Using an AI provides an unbiased and thorough analysis of data to help humans be exponentially better at what they do.​​

Why use an AI?

It provides an unbiased and thorough analysis of data to help humans be exponentially better at what they do. The benefit of using an AI algorithm is, “it makes you think, rather than simply react” when making a decision. The previous articles dealt with the four main AI technologies that we use in our algorithms, this article discusses, the benefit of using an AI algorithm to support a thoughtful decision verses a knee jerk reaction.

Why do we need help thinking rather than reacting? The “thinking” portion of the human mind is typically lazy and is comfortable with letting the “reacting” part of the brain do most of the work. We justify the use of the “reacting” part of the brain by labeling it “experience”, “skill”, or “intuition”.

In cases where the decision is highly structured, occurs frequently and has rapid unambiguous feedback, like playing sports or driving a car, we are justified believing in “experience”, “skill”, and “intuition”. When the situation is unstructured, occurs infrequently and has slow and vague feedback, like investing in an idea or pursuing a lawsuit, data shows that random luck is as likely to explain the results as is “experience”, “skill”, or “intuition”.

There are many books, studies and articles written by renowned Nobel winning authors that have proven, if we are not on guard our biased reactionary brain will make the decision, rather than our thoughtful analytical brain. The book ‘Nudge’ by Richard Thaler and Cass Sunstein along with the book ‘Thinking Fast and Slow’ by Daniel Kahneman are just two examples of these books.

There are many challenges that exist when it comes to forcing us to think, rather than react, when making a decision which manifest themselves as bias in our decision making which can reduce our long-term wellbeing. One of the many examples of reacting rather than thinking is the question ‘a bat costs a dollar more than a ball, the total cost is a dollar ten, how much does the ball cost’. If you answered ten cents, you got it wrong, the answer is five cents, most of us get it wrong.

These information biases include; influences from unrelated information, belief that you have all the facts, belief you are an infallible expert, taking a narrow view of the facts, a preconceived point of view, substitution of an easier question, the value of a fact is determined by ease of retrieval, pride of ownership, and the last thing you considered is the most important. That is a lot of biases!

Let’s simplify the discussion by creating three groups. The first group, the gathering bias, includes, the belief that you have all the facts, taking a narrow view of the facts, and truth is determined by ease of retrieval. The second group, the filtering bias, includes, influences from unrelated information, substitution of an easier question and the last thing you considered is the most important. The third group, the vested interest bias, contains the belief you are an infallible expert, a preconceived point of view, and pride of ownership.

The Gathering Biases

The first step in making an informed decision is to search for the facts. When searching for and gathering the facts we use our “experience”, “skill” and “intuition” to form queries, interpret the quality of the results, and then create additional queries from those results all of which uses the “reaction” part of the brain and introduces bias into the decision.

The “belief that you have all the facts” and “taking a narrow view of the facts” are issues of volume. We are all stretched for time and we all have deadlines, thus we are going to limit the amount of data we will gather. Many of us have that reading list, of articles, books, web sites, etc. that we know we should read but we just can’t get to it. To ease the stress of not getting to that reading list, we simply discount it as unnecessary and we go with what we have. An AI algorithm addresses this bias by simply ingesting massive amounts of data. The ability to scale an AI algorithm virtually eliminates the issues of time and volume.

The “truth is determined by ease of retrieval” is an issue of search effort. When we are looking for something on the internet, how many of us use more than one search engine? How many of us click past the first page of search results? Numerous articles on Search Engine Optimization tell us the answer to the first question is 90% of people use one search engine, and the answer to the second question is less than 10%. An AI algorithm addresses this bias by using multiple search engines and drills down hundreds of pages to get at all the relevant data and then applying consistent processes to interpret the search results.

The Filtering Biases

Once we have gathered our data, we organize it before diving into the analysis. In general, this organization consists of eliminating useless data, and then sorting what is left into groups. This elimination and sorting introduce the biases of filtering.

The “influences from unrelated information” is the unnoticed impact of background activity on our opinion also known as priming. A news story that was playing in the background on the commute to work or a casual conversation with a friend can subconsciously bring experiences and attitudes to the front of mind. These subconscious influences will impact the data we select and how we sort it which impacts our decision. An AI algorithm addresses this because it is insulated from all data other than the data it has knowingly ingested.

The “substitution of an easier question” is a symptom of our busy brain trying to reduce our mental workload. A complex question like ‘does buying this piece of intellectual property provide a meaningful marketplace barrier of entry to my current and future competitors?’, could be substituted with ‘would I pay for this feature?’. When the substitution happens, the brain is unaware of it. We still think we are answering the original question, but we are not. The substituted question will cause us to organize and eliminate data differently than the original question. An AI algorithm does not change the question it is asked to answer.

The “last thing we considered is the most important” is the result of the effort it takes to keep everything we know about a topic, ‘loaded’ front of mind. To be fair we not only keep the last thing we considered, but we also keep the thing that stood out the most. Think back to a major family event, and you will probably find that your impression of how it went is based on your most vivid memory from the event and how it ended. An AI algorithm has a vast capacity to keep data ‘front of mind’ and it processes all data equally, regardless of when it was processed.

The Vested Interest Biases

Now that the data is gathered and filtered, we can commence to analyze it and make a decision. At this stage we need to use our “experience”, “skill”, or “intuition” to conduct and interpret the analysis of the data. Mitigating the bias of our analysis on our decision is difficult to say the least.

“A preconceived point of view” is the effect of our ‘going-in opinion’ which requires a significant amount of mental energy to change. This preconceived perception, which can come from just about anywhere, anchors our point of view. We have all experienced a time in our lives when we need to change someone else’s or even our own mind on, an acceptable curfew, the time it takes to do something, or the quality of a product. It is difficult for us to determine how, why, when and where this preconception came from and how to protect against it. An AI algorithm does not have a preconceived answer, it simply applies the algorithm and delivers the result.

“Pride of ownership” is the belief that if we own something, or if we created it, we endow it with greater value than anyone else would. If you have ever attended a garage sale, you have experienced this firsthand. Items that are the same as ones we own will seem priced low, while items we don’t own will seem overpriced. You can imagine how strong this bias is when someone has worked tirelessly for months to create something and is then asked to rate its quality. An AI algorithm does not have a feeling of ownership regarding the outcome of its analysis, it is indifferent to the outcome.

“Belief you are an infallible expert” is the rejection of data, analysis and outcomes that do not fit what you think the outcome should be. Accepting something that does not fit with your understanding of a situation takes an extreme amount of mental energy. Thus, your reactionary mind will justify omitting or discounting information that would force you to change your mind. An AI algorithm applies a consistent, repeatable approach to arriving at a result, which is uninfluenced by a preconception of what the answer should be.

Using AI Makes the Experts Better

I know that some of you are saying “Wait a minute, who says that the algorithm won’t miss something a human wouldn’t!”. Maybe you are thinking “hold on, I heard AI algorithms can be biased!”. You might be thinking “if I take my time and carefully think through each step and simply reduce the amount of data I collect; the effect of bias is reduced”. Well these are all legitimate statements, however the laws of probability put AI algorithms ahead of experts.

Yes, AI algorithms can miss something a human would not, however the shear amount of data that an Algorithm can collect, and process makes it more likely that it won’t. Yep an AI can be biased, but this bias is consistent and repeatable, and it can be eliminated through adjustment to the algorithm and training data sets. Finally, yes, you can take your time and reduce the amount of data you gather. Ignoring time constraints is easier said than done and reducing the data set introduces additional filtering bias.

To be clear, no one is saying “get rid of the experts”. What we are advocating is the application of AI powered algorithms to assist the humans making decisions that are unstructured, occur infrequently and have slow and vague feedback. Running an algorithm in parallel to the traditional techniques and then combining the outputs will address the biases introduced by the “reactionary” parts of people’s brains while allowing for the creative insights that the human brain is noted for.

There is value to using an AI algorithm, even in a situation where the human and the algorithm differ greatly. The process of reconciling the differences will require deeper and more careful analysis which will create a more resilient result. The use of an AI algorithm is to augment not replace traditional approaches. The strength of the traditional approach offsets the weaknesses of the AI algorithm and vice versa.