Artificial Intelligence is on everyone’s lips at the moment: thanks to AI, Tesla is making cars that will never do an accident again, IBM is working at vanishing cancer, Amazon is creating new-generation shops with no cashier, no scanning, no in-shop payment, no nothing except goods and consumers, and Minority Report is about to become real in some of our biggest cities.
Hearing and seeing all of this, it’s like AI is massively changing the world although… AI was unknown to pretty much everyone only one year ago… so what happened? Did we miss the birth of AI? Was AI created like overnight?
A bit of history
Actually, AI is more a concept than a technology, and it heavily relies on Machine Learning, which is the ability for a computer to learn by itself how to “recognize something”. This “something” can be a cat on a picture, it can be an emotion in a voice, it can be a correlation between multiple events that systematically lead to a situation.
There is no magic in Machine Learning, it is based on a set of algorithms that apply algebra, probability and statistics principles. These are called “Support Vector Machine”, “Logistic Regression”, “Bayesian Linear Regression”, “Neural Network”, etc. and what’s especially interesting with them is that they’ve been invented many, many years ago. The Neural Networks for example which is heavily used today in Machine Learning gets its origins in the “Perceptron” which is an algorithm invented in… 1958, 60 years ago!
The first big, worldwide, popular usage of Neural Networks and thus of Machine Learning and thus of Artificial Intelligence is already 20 years old: that’s the PageRank algorithm of Google Search.
But why not earlier?
So, why now? Why not 20 years ago? Why not 60 years ago? Well, that’s because the algorithm is only the first of the three essential elements of Machine Learning: to give results, ML needs algorithms of course, but also data – tons of data – and it needs computing power, a lot. Those two last elements have been lacking for decades.
Today, on the data point of view, I’m pretty sure it’s unuseful to remind you about Big Data, and to remind you that we live in a world in which we have 400 hours of new video uploaded on Youtube, 2.5 millions Instagram likes and 4 millions Google searches… every single minute. That’s a big shift compared to the situation we had only 10 years ago.
On the computing power side, Moore’s law has proved to be true and the processors are today fast enough to support Machine Learning. Mix this with the cloud war raging between Amazon, Microsoft, Google that translates into cheaper than ever access to computing power and you get the perfect combination for enabling Artificial Intelligence.
AI, accessible to all
But, does that mean that AI is accessible to us, poor ordinary people?
Well, Yes. Actually you need to see AI as a “toolbox”, as the mobile is, as the web is, as the whole computing industry is: you can make 1-day AI projects as well as 1000-days AI projects.
How? Well, there are mainly two ways:
- Microsoft, Amazon, Google, IBM, … not only are they offering cheap cloud solutions for what we call “Infrastructure-as-a-Service”, but they are also offering cheap cloud solutions for what we could call “AI-as-a-Service”: when you need to do face recognition, voice emotion analysis, natural language processing or even near real-time recognition of objects in videos, they have an out-of-the-box solution for you!
As an example, for the image recognition, the pricing is around 1-1.50$ for 1.000 processed images. - The cloud solutions are very efficient, but of course you need to stick to the cases for which they were built. If you need more flexibility, or need to use AI for cases that are not covered by the cloud providers, you will need to experiment AI and ML by yourself.
The good news is that Microsoft and Google are also fighting on this field: they’re both open-sourcing parts of their technologies so that you can build AI/ML much more easily. Microsoft does that with its Cognitive Toolkit, while Google does this with TensorFlow and Sonnet. And Microsoft/Google are not the only way to follow, you can also find many open-sourced (or not) solutions in the community.
A few cases
Let’s go through a bunch of cases where AI can enhance your everyday life:
Computer Vision
Computer Vision is the ability for computers to interpret images and videos thanks to AI.
If you’re a used goods marketplace, by associating an image recognition API and your app, you can remove from your users the burden of filling the form describing the object: just “recognize” the picture of the object and automatically suggest a description that fits.
If you’re selling fashion and shoes, how much of your customers are discovering the items they want in physical stores before searching for them on your e-commerce? What if your customers could just send you the picture of the item they wish to order? Actually, Zalando already did it in its app!
If you’re a cooking website, your editorial team might be struggling with all these users that don’t upload pictures along with their recipe: no picture with the recipe means less attractiveness and less SEO performances. What if your CMS was automatically suggesting replacement pictures that fit with the text of the recipe?
All of these cases are today totally accessible at a more than reasonable cost.
Natural Language Processing & Understanding
Also widely available on most of the clouds, a bunch of services can help you in extracting the important elements in a naturally written sentence.
It can extract emotions from human sentences: “No, thanks”, “No :)” and “NO.” do not mean the same thing.
It can also help you understand that “Book a table for 3 at Huggy’s Bar tonight” and “I’ll have dinner at Huggy’s Bar with 2 other people tonight” have the same meaning: it is about a restaurant booking, 3 as a number of persons, “Huggy’s Bar” as a location and 19.00 as a date/time information.
Combine this with the ability for these services to handle both text and vocal inputs, and you get pretty powerful cases.
During the 2016 US Presidential Election for example, nearly 20 percents of all tweets exchanged on the subject were generated by robots!
If you’re trying to optimize the load of your customer support department, automating some parts of the conversation to reduce the number of low-value questions/answers being handled by humans could be a solution as well: think about the volume of repetitive answers your customer agents can give. Or think about the guys at 911 or 112 that need to do high-speed triage between emergencies and non-emergencies when the service gets overcrowded.
Maximizing conversions on your e-commerce
As an e-commerce, one of your challenge is to minimize your shopping cart abandonment rate, which you might tackle by re-engaging your consumers through notifications, e-mails, etc… at the exact right moment so that you maximize the recapture of the abandoned carts.
However the best engagement moment as well as the best message depends on a lot of factors: user’s socio-demographic profile, the kind of product, the total amount, the day, the time and a bunch of external factors.
Once again, AI and Machine Learning can help you: do some campaigns, have some data scientists work on the data, put them in a Machine Learning system and let the system determine user-by-user the best way and the best moment to contact them.
E-commerce conversions is also dependent on how good you are at suggesting the right products to the right consumers, Amazon’s example being of course the state-of-the-art in the recommendation engine field. This kind of engine is often heavily based on AI and Machine Learning as well.
Hint: don’t forget about the data!
Whatever the AI/ML project you want to lead, the base element is the data: so, in any case, collect and store your data, any data that you can have. The more data you collect, the more you’ll be able to tap into past patterns to improve the user experience.
Be careful that you need to collect raw data to have a maximum flexibility with your projects, which means that you should go beyond collecting data with standard analytics tools: surface level analytics, such as the number of clicks per hour, won’t be enough. Think broader than that: useful data can come from numerous sources and in multiple formats: online and offline, photo and video, clicks and contextual data, etc.
To achieve this, setting up a Big Data platform is a must!
As a conclusion, is AI for me?
There’s no doubt AI can enhance any existing service!
What you actually need to remember from all of this is:
- AI is now accessible to anyone.
- AI is a toolbox, as the mobile, the web, the entire computing industry is!
- You don’t need to revolutionize your service, AI is also very good at enhancing existing processes and services.
- Collect data! Any data you can have!
- Iterate! Fail fast, learn fast.
