It has become very popular to associate a product with AI. Every day, you can stumble upon a new startup using “.ai” in their URL or in their marketing communication. This hype wave helps marketers a lot! Why? Because it sells and it helps to attract potential investors. However, are they really using AI… ? or just some advanced analytics?
Based on my experience, many AI companies are actually using basic data analysis, logic, Robotic Process Automation based bots or “pseudo-AI”. Through a specific communication, all these technologies can appear as AI if you only know a little bit about them.
In this article, I assume that AI means weak AI and that Machine Learning (ML) is an advanced solution! I know too well that ML algorithms are generally just memorizing and running statistical models. Despite this, these ML solutions are often considered as the best of artificial intelligence.
Back to our topic, organizations want to become more data-driven. As a consequence, many SaaS companies or AI development teams are proposing “augmented analytics” or other solutions… which most of the time can’t be considered as AI! For me, AI systems can get smarter with the more data they analyze and become increasingly capable with experience. It is not just a compilation of algorithms or just a bot created to automate a task by replicating a human action.
In my last project, I had to make sure that an image recognition model was actually able to learn and find new ways to help the end-user.
For instance, I see many decision-making tools using the term “augmented analytics”, even though all they really do is use data analytics and visualization to highlight data in a more clever way. The tool doesn’t get more intelligent over time, it is not learning and adapting to data.
As Vance Reavie said, “Machine learning is a continuation of the concepts around predictive analytics, with one huge difference: The AI system is able to make assumptions, test and learn autonomously. “ (source)
Machine learning is also about predictions and has this capacity to recalibrate models in real-time automatically. This feature is very important for most organizations. Meanwhile, predictive analytics must be refreshed with “change” data.
When it comes to analytics and AI, here is what you should keep in mind:
- Predictive analytics is making assumptions and testing based on past data to predict future (what/ifs).
- Data analysis is about reviewing data from past events.
- AI analyzes data, makes assumptions, learns and provides predictions at a scale and depth of detail impossible to replicate by humans.
A predictive analytics tool can go through a specific set of defined data inputs and a single-purpose model, and support some kind of decision. As a result, it may lead to better productivity, higher sales or better understanding, but often actually fails at making precise recommendations and supporting decisions that really can have an impact on the long-term.
This is because the models aren’t designed to look at the various ‘ripple’ effects. They can’t easily find patterns in data and they simply do not have the capability to process that much data.
The word “algorithm” can also become a marketing asset. An algorithm may be quite clever and elaborate but it can also not learn by basically doing what you taught it to do. In more complex scenarios, it may do things you can’t even explain… that’s AI. In a recent Machine Learning project, I had to challenge data scientists into explaining to me how will the AI learn from data, identify hidden patterns and actually make unique recommendations to the end-user.
AI is able to test many times data to predict every possible outcome, at a speed and capability no human could attain.
In all my projects, we had to help the company either create or prepare a data set. I can remember myself taking hundreds of pictures to help the company create a data set and then do the labeling…
I highly recommend you to know more about the data set that will be used in your project. I noticed that most impressive AI startups have unique datasets. In general, if you decide to use a SaaS AI solution or outsource your personalized AI development, you should expect a lot of questions related to your data, if not… it’s a red flag. If your technological partner ignores the importance of data, they probably intend to just rely on non-AI technologies.
I strongly recommend you to not underestimate the data aspect. Indeed, it may take considerable time and resources to prepare the data if it is not already available. In my opinion, an organization considering a particular AI application should first consult with AI experts to determine if the required data is available. If not, you need time to create a real data culture within your organization.
Robotic Process Automation
During my past projects, I also realized that people tend to confuse a lot Robotic Process Automation (RPA), which is a software robot (bot) that mimics human actions and AI, which is the simulation of human intelligence by machines.
To keep it simple, I’ll say that RPA is associated with the automation of a given task whereas AI is more about automation and learning related to this same task to improve it.
Another distinction would be that RPA tends to be more process-centric while AI is data-driven. Nevertheless, RPA and AI are just different ends of a continuum called Intelligent Automation.
Let’s not forget that AI is an umbrella term . One company could use it to mean simple automation, image recognition model and another company may mean deep learning. All decision makers want to become data-driven but confuse business intelligence (BI) with AI.
I always remind my partners that if you want to have a data-driven strategy for your organization, BI and analytics (enhanced with data visualization techniques) are perfect solutions, but if you need personalized recommendations, the ability to predict the results of some business decisions and so on, ML is where you need to start your AI transformation journey.
In general, whenever you are interested in an AI solution, you must ask how the data will be used and how is it different from a more traditional analytics solution. An AI solution learns from data so if your interlocutor can’t explain this part clearly and what it means for your business, then there is an issue.
Furthermore, I invite you to ask about the training of the AI; training is how you help it learn about your environment and desired outcomes. Training is a fundamental part of any AI project. It’s absolutely crucial that everyone involved in the development of your model understands how it works.
Steps to follow in an AI project
Usually, in many AI projects, we have 3 important steps:
The first step is about training the AI then we have the validation step. In the second one, we will test the model with new data. The algorithm should do better than when it encountered the training data for the first time. The goal here is to look at results and evaluate them. It’s possible that we have overfitting issues (it happens when the model has been trained too specifically to only recognize examples in the training data).
Finally, the last phase is testing. If the model has aced the validation process, it’s ready to be tested against data without tags. This simulates the state of the data it will be expected to perform against in the real world.
I recommend you to follow these steps when conducting an AI project.
The AI transformation journey is not easy. Many companies are promoting the idea of easy AI but the reality is different. In order to change your organization, you must ask specific questions, study AI case studies and perhaps learn a little bit more about AI yourself.
For more information: