Wed, Mar 24, 2021
Usually, when I talk to decision-makers and executives in companies starting their AI journey, I hear some questions, misconceptions, and concerns shared by most of them.
Many of these business owners are diving into the subject for the first time and ask us for advice on how to approach and "tame" these technologies that may seem too wild at a first glimpse.
Based on those conversations, I came up with some quick tips to inform business owners and allow them to consider AI as a new, powerful tool in their toolset.
This list of what I called "AI Myths" sums up and compiles some common pitfalls. The idea of this post is to try to debunk them for you, slay these mythological fears, and get you all excited about the possibilities of AI for your industry.
The belief that artificial intelligence takes jobs away makes many business managers reluctant to take the first steps towards automation.
One thing it's true: that the inclusion of AI fundamentally changes many aspects of the work dynamic for a company. But the objective is not to remove jobs; it's actually to make human operators more efficient and scalable.
The most successful vision we see for AI in traditional industries is what's referred to as Human in the Loop. The idea is that AI gives us the ability to build tools that will help people focus on higher-value and high-skill tasks and remove the repeatable and less valuable aspects of some operations.
The end goal is to make a company more efficient by enabling its workforce with the right tools to provide more value.
Don't get me wrong here: we love R&D and the academic community; we even have our R&D area at Tryolabs. We give AI courses at Universities around the world, and we're pushing the state of the art of Machine Learning with it.
The truth is, there's a ton of research in the vast field of AI, and techniques are experiencing breakthroughs very fast. For that reason, we've seen decision-makers turning to academic minds to explore what AI can do to help their companies. But that's not the only way, and it's, in my opinion, not the best approach for companies looking to reach a business result with AI.
The biggest challenge we're seeing for companies trying to adopt AI in their processes is not researching a novel model or technique but discovering a solution. That discovery usually means looking at their data, connecting it with already available models and more importantly, putting those systems into production to work and get business results.
Let's hold off the problem-solving stage until we have more data available. That's a widespread misconception. Data is indeed the essential resource, but working on it is an integral part of the AI journey, and it happens during the creation of a solution.
We often see an iterative improvement and a retro-feed of data and processes, where the processes you put in place change the data you can get. Still, the data you need requires changes in your processes and operations. It's a discovery phase, and you shouldn't think of it as two separate stages.
You can think of the concept of "Building the AI" as working with the data, and analyzing which models and solutions will give the best results.
Some business owners are hesitant about innovations like AI because why fix what's not broken?. One must acknowledge that the industry one may be part of will change due to AI's impact, if it's not already happening.
Many industries are being reshaped by it at this moment. Using AI in a meaningful way really is a competitive advantage, and companies doing it will be winning over their competition. It's a way of staying ahead.
Forecasting, planning, logistics are just a few examples of the internal company processes that can improve with the right AI solutions.
It's important to establish a business result as the primary objective. Not doing AI for the sake of it, but making sure it will have a real and sustained impact on the company.
Building software takes time. Building machine learning models takes time too. Having said that, many techniques aim to provide quick-to-deploy solutions to common problems. Again, it's necessary to separate academic work, R&D, and the discovery of new models from AI applications that are impact-driven.
Our approach to getting results as quickly as possible is by leveraging the expertise and the technologies that are already proved and work in their industry or similar use cases.
In those cases, the investment poured into the project paid off very quickly and continued in the company for years. So, is it possible to get great results fast? If you pick the right problem, the answer is yes.
Check our price optimization use case for more insights on getting impressive results in a couple of months.
An RPA is a tool to automate interactions or actions performed by a user in a certain software or device.
It is a charming tool for automating a task that will take you one step forward in a heavily manual scenario, but it won't solve most problems when trying to scale.
That's something that one tool alone can't do. For most scenarios, going for a more in-depth solution is what will get impactful results.
An RPA can help automate a small part of an existing process, but analyzing and re-thinking the process usually leads to finding opportunities for AI to create more value and eliminate bottlenecks.
Here's an excellent example of what we did for a retail leader from the US, where we automated their pricing processes. It's a case study where an RPA would have made little difference. Instead, studying the process, proposing changes, and using the right models and solutions for each piece of the puzzle created the high value they were looking for.
Finally, there's this concern that probably comes from social networks and big tech. Where is our data going? Why do models need to know the private information of users? And especially for companies, there are concerns about data governance and how models will use sensible data from users, operators, or internal processes.
That is a genuine concern, and it's natural for companies to take the necessary measures to keep their data safe. At the same time, It's not true that AI always needs sensible data to work or that models will end up exposing data disregarding security practices.
It is popularly known nowadays that most modern AI techniques rely on having as much data as possible to have better outputs. That's not incorrect and leads to companies being sometimes "tempted" to feed their algorithms with as much as they can. That's not always the right way to go.
Most problems in traditional industries are solved without taking advantage of private data. Even in scenarios where all data is sensitive, anonymization techniques can protect users' privacy and ensure no confidential information is leaked through the models. External data and even publicly available data sources are often great sources for machine learning models.
And in terms of data governance, there are many options to avoid moving data from local servers and accommodating security practices, like running solutions over a private cloud or on-premises if needed.
Protecting individual identification is an everyday task. We shared before some of our experience on anonymizing personal data, and even the caveats of some privacy-preserving models and datasets. Last but not least, we built a face mask recognition system that's not using any facial recognition and doesn't store data from pedestrians. Believe us, it's possible.
I've heard many times these questions and concerns. They are super valid and come both from decision-makers and employees or operators that see their work threatened.
In my experience, it doesn't matter how many stories we tell, statistics, or the theory behind it. Companies need time and trial experiences to finally trust and embrace AI as an amazing tool and a friend in this journey.
So we always give the same advice to business owners: start small, but START. Allow AI to show a small win or solve a "low hanging fruit" problem, so the company as a whole starts building trust and giving these technologies a chance. That, if well managed, can lead to a path of great success and impact in the AI journey.
From our side, I think the best we can do as ambassadors of these technologies is to inform and enable these first steps with AI. That way, these questions and concerns will slowly disappear, and it will lead to a future with very positive and fruitful uses of AI in the workplace.
Let me know what you think! Can you think of any other myths and concerns?