I enjoy imperfect things. I like my sweater with its holes at the elbows, that painting of mine that my cat walked over while it was drying, and the source code I use for my doctorate that never seems to execute as I expected. I like it that way. Things are more interesting when they are imperfect.
There is money to be made when you are talking about business. In the business world, a small mistake can result in millions of dollars in losses.
That's frightening. When these losses happen because engineers make mistakes while trying to implement a new and rapidly evolving technology, it's even worse. It proves that business leaders are human if they are hesitant about this potential pitfall.
I'm talking about artificial intelligence. Many people, including many business leaders, remain enthusiastic about artificial intelligence. Artificial intelligence can finish processes that used to take hours in seconds. It is an improvement of several orders of magnitude. Billions of dollars are poured into artificial intelligence every year.
Despite the huge investment, there is still a lot of interest in artificial intelligence. The difficulty comes from the fact that many businesses fail to make a realistic assessment of the types of changes that artificial intelligence can and can't bring.
In the Harvard Business Review, Larry Clark shared an anecdote that perfectly illustrates the problem. He spoke with a consultant who said that his client made 25 percent of their predictions correct. The consultant told them that an artificial intelligence solution could get this number up to 50 percent. The solution that the executive refused to implement was wrong half the time.
The failure rate of 50 percent is enormous. It would have been twice as good as the existing solution.
Many executives are disappointed that artificial intelligence won't change their company quickly. Kevin Kelly, founding editor ofWired, said that the future is very slow and then all at once.
I think this rule applies to a lot of tech. New developments are on the horizon, but you can't expect them to happen tomorrow. Good things need time to grow. In the fast-paced world of tech, patience is a virtue.
When artificial intelligence doesn't transform their business into the next one, leaders shouldn't be upset. If a new solution brings many small improvements over time, that may be more valuable in the long run. A big wave of disruption can be a good thing, but it can also be a bad thing if it impairs other business processes that were standard to this point.
Artificial intelligence isn't always the best solution. The image is by the author.
If you've worked with artificial intelligence before, you know of concepts such as accuracy, precision, recall, F1 score, underfitting, false positives and false negatives. If you come to them with technical jargon, they will look at you like you are an alien. Executives care more about results than technical details.
The founder of a company that builds artificial intelligence solutions for the insurance industry has made this point. What really matters is whether the solution makes things easier for workers, reduces costs or increases margins. It doesn't matter if you get spectacular results on a technical level or not, as long as your solution improves the status quo in your company.
Data scientists will continue to use technical jargon because it is useful for them. In order to translate this jargon into business terms, executives need to work closely with data scientists, involve them in business operations, and never stop asking them how the performance of different technical metrics might impact the business as a whole.
Data scientists are in high demand. Many companies are understaffed. Many data scientists with too many projects on their plate need to prioritize the hard analytics and not find the time to think about the business part of their job.
Hire data scientists before you need them and give in-house training to new team members to avoid this situation. There are two big upsides to adding training inside the company, even though it requires some upfront investment. Data scientists get in-house training to know the specifics of the company. Second, this type of training is attractive to younger job candidates who bring in fresh ideas and don't demand high salaries from their senior peers. It will pay off in the long run if you set up a rigorous in-house training regimen.
It isn't everything.
Machine learning should be as accurate as possible. We don't want our machines to misclassify a cancer as a benign one. Accuracy isn't always the goal. Let me explain.
There is a risk of overtraining. An artificial intelligence model can learn a data set so well that it can discern small details that aren't relevant for the outcome. An artificial intelligence solution can be used to classify a data set with lots of different animal species. Imagine that the data set contains only one type of animals. There are two types of monkeys: black and orange.
If you train this model too well, it will only recognize a monkey for a monkey, but also know whether it is a black or orange one. If you test the model on a picture of a gray monkey, it gets problematic. How will the model classify that animal? A cat? A gray dog?
The risk of misclassifying new data arose because the model became too accurate during training. To avoid this problem, data scientists and business executives need to pay more attention to performance during testing and less to accuracy during training. The goal isn't perfect.
This would mean allowing the program to misclassify tumors while training. Aiming for 90 percent accuracy could be the result of this recalibration. It will be better prepared to classify a tumor that doesn't look like any of the ones it saw in the training stage when the algorithm is deployed in real life. It's important because a data point is unlike any other. Every new data point gets fed back into the system and helps retrain the algorithm, so you're giving it a chance to improve its accuracy in real life.
Take it one step at a time. The image is by the author.
There are other places where executives need to temper their ambitions. Jon Reilly writes for Dataversity that businesses have a tendency to throw artificial intelligence at big problems and expect good results.
That isn't how it works yet. It works best on small, specialized tasks in which a big volume of data needs to be processed. Start with jobs that will get too repetitive for humans and then build it out from there. This is a bottom-up approach. Top-down approaches are hard to nail. We are far from being able to transfer knowledge from one domain to another, and even further from generalized intelligence. Teaching a machine how to do boring and repetitive tasks at warp speed is much easier than making it complete a complex task, even if there is plenty of time to spare. That doesn't mean that this situation won't change in the future.
The classic 80/20 rule states that 20% of your tools and resources should go to 80% of your output. Make sure your solutions have the biggest impact by focusing on the tools and resources.
It's better to start with the easier parts than to redesign the whole company. A big, overall solution that is too complicated to be effectively deployed should be prioritized over some patchwork solutions.
Hesitant companies will lose.
The early adopters of new technology are the ones who will collect the most cash. It's not too late to get into artificial intelligence.
It's not an excuse to perfect your model and live with it for five years. Despite all the obstacles, more and more companies are starting to see the potential benefits of artificial intelligence, however small or buggy things may be at the beginning.
That is the right approach. We haven't tested all niches and edge cases yet because the technology is new. You should test half-baked solutions. You risk missing out on key lessons if you don't push your updates regularly.
During my studies, this problem has happened to me. I was working on a procedure to process a lot of data in a more efficient way. The procedure was part of the project, so I wanted to perfect it myself before sharing it with my team.
I realized from my colleagues feedback that I had been missing out on some key ideas when I finally shared it after three months. I made the code three times more efficient than the old version. The improvement was five-fold after implementing my colleagues' ideas. Although my work was a public research project and not a business, the thought of not speaking to my colleagues earlier still hurts.
Companies that aim for perfection too early will be left behind. If you want to end up ahead of the pack, you need to be able to turn down your ambitions and sit with an imperfect solution.
Don't worry about imperfect code. The image is by the author.
You can't pretend that your job is done when you drive home from work. There is always a bug to find, a feature to add, and a tweaking to make.
If you want to use artificial intelligence for your business, you need to love this reality. This rule is not just about business. Many life situations work out better with rough-and- dirty pragmatism than with perfectly orchestrated processes that fail as soon as the bus is one minute late.
It isn't an excuse to be lazy or to only do the minimum to keep up with the competition. Do the best you can. The best is not always perfect.
This article was first published on Built In.
All Rights Reserved © 2024
