Building artificial intelligence with purpose

Earlier this week Parmy Olson wrote about real AI usage on Forbes and we decided to share our thoughts about solutions made with artificial intelligence.

The first question: From where can we call it artificial intelligence?

Artificial intelligence basically means that the computer can produce a solution for a problem. Well is a light that can be controlled via smartphone an AI solution?

Well, the lights computing module reads the data that has been sent to it and by that makes the adjustments. The thing is that this can be described as artificial intelligence, but most of the times we are thinking more about deep learning or machine learning solution, even we are speaking about artificial intelligence.

Machine learning means that the computer collects data and learns by that to solve the problem. Deep learning instead are focused on building a model from the data and making analysis on it, so it can learn all the dependencies.

Simple picture to get the picture about AI,ML and DL.

Most of the times AI solutions are described as stupid AI, weak AI or narrow AI. Meaning that they are solving only the problem that they are programmed to do. Artificial General Intelligence or “strong AI” is the next step for artificial intelligence and it’s estimated to start rising in the next 10 years.

The second question: Can we get any value from using artificial intelligence?

This is the question that companies should ask themselves. First of all they should be able to tell how their AI solution ended in the result, so they will not suffer from the “black box” problem. In second they should ask is it giving any value for their business and what is the purpose actually.

As described earlier we can build a simple AI solution to do the tasks, but we can also build a heavier deep learning solution. What is more efficient for business and for the service? Those are things that should be thought about when looking for an AI solution to take care of something.

So we would say that companies can get a lot of value using AI, or loose it. The most important thing is that the team delivering the solutions should answer to the customers problem, instead of delivering what they are asking, because most of the times they really don’t know what they need to solve their problem.

Gartner’s Hype Cycle is full of different AI technologies and it’s telling when we are seeing popular use of the technologies. Companies should be looking at this to see what’s coming in the near future, so they can get ready for them and understand what they actually mean.

Gartner Hype Cycle for Emerging Technologies

The third question: Why there is so many AI startups without AI?

This is easy. It’s being part of the hype-train. Meaning that it’s popular at the time and all of the people can’t really tell if it’s AI actually. As Olson wrote on the article, most the startups described as AI startups are getting more invests. But there is also other aspects for this. For example AI solutions are nowadays easier to build and more cost-efficient so people without high invests can start building those solutions.

Our way to deliver solutions.

Optimux is a platform that we have been using to deliver smart solutions. Our Optimux is used to control road light dimming in Ring 1, a road that has about 90 000 cars daily. The system reads real-time data from different sources and by that makes it’s adjustments to the controlling values and sends the steering data to lights. In our opinion this was the most efficient way to adjust the lights.

However there is a deep learning solution integrated. If Optimux can’t get reliable data from it’s sources or it looses internet connection, it will change to use a prediction model of the values. The system has an offline model that contains predicted steering values for the next two weeks. So we are using artificial intelligence to secure the system and also to produce data that we are comparing to the real-time data, so we can tweak our models.

Optimux is also capable of controlling traffic flows, as it’s capable of tracking problem points causing jams and switching traffic lights so it can make drivers choose different routes as they see that some other lane is going faster for example.

This one example how we have delivered solutions using deep learning and we are now building our data collectors, which are named Optimux Sense. Those can be paired and connected to the main Optimux to create valuable information. They are also capable to be connected to other devices and take control of them. Safely.