Machine learning …to tweet?

Have you heard about Tay, the intelligent twitter chatbot built by Microsoft that went from a typical 19-years-old girl to a nazi extremist in less than a day and then got yanked off the Internet? The experimental bot tried to mimic a normal teen and interact with other humans on the platform. Trolls found a way to make it say anything, and disaster ensued. It is an interesting anecdote but behind the comical aspect of the matter, how can you build such a software?

 

How do they do it

The technology behind Tay is called Machine Learning. It is a subfield of computer science that get computers to solve problems without being explicitly programmed to by an expert. Even if algorithms and their implementation greatly differ, they more or less all work on the same principles:

  • They require quantifiable data, and a lot of it. We’re talking Big Data and multiple thousands of data points. An algorithm doesn’t do much until it consumes massive amounts of information.
  • It needs to be told what the data is. You cannot tell a computer to predict if a message is happy or sad by referring to its concept of happiness or sadness (which it do not have, at least for now). But give it a million messages known to be happy, and a million message known to be sad, and from this point you can start predicting other messages based of the features encountered in this training data.
  • It has to get feedback on his guesses in order to refine its predictions. Computers will never be perfect at guessing (neither will humans). By allowing feedback, computers can get better or adapt to evolving situations. For example, if a product recommendation algorithm gets from sale statistics that black dresses are never added to users’ carts when these declared themselves males in their profiles, the algorithm will stop suggesting them to men and try something else.

The algorithms learn from experience and training, pretty much like a human would do. Machine learning is teaching computers to recognize and apply patterns. Once an algorithm is correctly trained, it can emit extremely accurate predictions at blazing speeds (tried Google Images or Gmail spam filter recently?).

What is intelligence?

A lot of what we consider intelligence is just pattern recognition; i.e., the ability to go from particular cases to general rules on which we will classify other encounters. We can see a million dogs, and we will all think of them as dogs. Not because we have a unique brain model of each of them (you will probably forget them as soon as they are out of your view, even if they are really cute), but because they all fit our internal pattern of ‘’what a dog should look and like and how it should behave’’.

How do we know that the big brown furry thing coming at us is a mascot and not a bear? Because its fur looks like a carpet rug and that it walks like a human. Patterns. Knowing that when we say an orange car, we’re talking about the color and not the fruit? Because it is used with the word car. Also another pattern. Natural language is a big complex problem of pattern recognition. And it is this problem that Microsoft, and a lot of other players in the field are trying to tackle.

Wanna see how your brain can be tricked with its internal patterns? Check out the concept of the uncanny valley. When something fits enough our natural being pattern, but not completely, our brains have a really hard time accepting what they see, and a feeling of creepiness and unease is often the result.

How machine learning can help you, concretely

Success or failure, Tay stays impressive. But you’re probably not Microsoft, and probably don’t have the same ressources. So let’s see briefly how machine learning can be used wisely to help your company.  

  1. Recommendation engine

    If you have a retail website, product recommendations are a great tool to help increase your average purchase cart. With machine learning, the recommendations that are made to shoppers are based on buyers’ patterns and feedbacks from the sales’ systems. This way, your suggestions become even more relevant, creating a positive feedback loop that increasing the average cart.

  2. Monitoring on social media

    It’s not always easy to keep track of everything that is said on your brand on social media or posted on our social media walls. You might have someone responsible of your social media accounts or might not. You might have a backup plan when this person is not at work or might not. You might monitor what is said about your brand or might not. With machine learning you can set an automated monitoring of consumers’ feelings and automated awareness for unwanted posts. In this case, the machine learning is based on the language used in the messages and posts containing keywords related to your brand. This way, you have better control over the conversation going on about your brand and over your brand’s image.

  3. Insights discovery

    Machine learning combined with statistics and a huge amount of data is called data mining. It consists, among other things, in discovering hidden information and drawing conclusions from the analysis of your data, leading to the discovery of useful insights. Many companies (maybe yours) have huge amount of data, but do not know how to create value out of it. Machine learning, once more, can be of useful help here.  

 

In conclusion

When talking about artificial intelligence and natural language, the Turing test comes to mind. The Turing test is a test, developed by Alan Turing in 1950, of a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. Short version: If a computer is able to fool a human into thinking that he is talking to another human, and not to a computer, the test is considered a success. Do that enough times, and you’ve got an intelligent computer (well at least, on natural language).

So even if Tay turned out not to be the young hip twitter girl her creators hoped for, at being indistinguishable from a troll, Tay did an almost perfect score. In an unorthodox and unplanned way, she’s giving us a small preview of what’s to come in the artificial intelligence (AI) field. The line between man and machine will get blurry.

To have a better idea on how machine learning and predictive algorithms could revolutionize your business processes and marketing efforts, get in touch with us! We promise you will have a human talk to you.

 

Author

Jean-François Grenier, Project manager