How to realize a scientific Artificial Intelligence project

Intelligent machines are without any doubt the future, and that a scientific project is needed to understand AI in detail is also clear. But how exactly can an Artificial Intelligence project be made in reality? What are the keysteps in booting up a research laboratory? This will be answered in the following tutorial.

At first, it is important to know that a science project has to do with buying something. The desired goods can be hardware, software and ideas. But let us make a small example. Suppose, the idea is to start a deeplearning research project. The first step is, that the lab needs some infrastructure:

• fast internet connection

• Workstation PC (5000 US$ each), router hardware

• 3d printer, beamer for conferences, haptic devices for input

• motion tracking hardware, Mocap hardware

• nvidia deeplearning cluster, cloud based server for running a wiki

To get in touch with such technology is simple, if enough money is available. Most of the devices are available in normal computer retail stores, others (for example deeplearning hardware and mocap tracking markers) are available in special store. It sounds surprising, but at least 10% of a research project is about finding the right hardware, and decide which of them has to be bought.

Suppose, all the hardware is there and the lab has around 1 million US$ less money. What’s next? Also the step 2 has to do with buying something. Buying another hardware makes no sense, because if everything is available then on some point it makes no sense to buy the 100th workstation. The next buying decision has to do with scientific ideas. Usually, each of the devices contains a manual, and also the AI literature provides some instructions. The question is now, how to use the equipment in the right way. For example it is possible to install Windows 10 on the workstation PCs or a Linux distribution. It is possible to focus on LSTM neural network or on Convolutional networks. It is possible to investigate research topic A or B. From a certain point of view, this can also be called a buying decision, because the scientists has to decide to get in touch with a certain idea and stay away from another one.

Perhaps one example. It is possible to use a out-of-the-box deeplearning software like Tensorflow or program everything from scratch. The question is not, how to use Tensorflow or how to program with C/C++, the question goes only about the decision itself. That means, to compare what is better and what the researcher want’s. Usually such decisions are made after reading lots of information and discussion the market with other experts. This takes also a lot of time.

Suppose, the research lab has bought computer hardware, and has bought some key ideas. The next step is to do write down a first description of the project. That means, to talk about the own buying decision and communicate it to the outside. In most cases, the people who gave the money are interested in such kind of feedback, because they want to know, for what the 1 million US$ was spend and about which topic in detail, the deeplearning project is about. Until now, the project took around 80% of the overall time. That means, the setup and buying decision need most of the time. With the remaining 20% of the time, the researcher can try to realize their own ideas. In most cases, they will fail, but that isn’t a problem, if the failure is documented well. Writing down, that the own attempt to boot up the workstation and make something useful with Tensorflow wasn’t successful is in case of doubt a real science project. For the beginners that’s sound crazy, because nobody has done real research. All what the project was about is to buy thinks, recognize what’s wrong with them and write down, that’s unclear what the problem is. But, that is the essence of a real AI project. If such a project took 12 months and costs 10 million US$ it can be called a great one and a template for copying the principle.

The most important discovery is, that apart from buying something, there is no way to make science. If somebody is on the standpoint, that he do not buy any product out there, and don’t want to buy any idea available he is not a scientists. Apart from “going shopping” there is nothing what a scientist can do. The only difference to the normal understanding of shopping is, that the products are different. Buying new shoes in a deeplearning project is out of the scope, except it is about image recognition for an online store. Scientific research is modern form of go hunting. At first, the prey is circled and then the meat gets slaughtered. Communication with the outside and other researchers is important because this will increase the success probability. Apart from this archaic ritual there is no other possibility to make science. If somebody isn’t interested in this social game, he isn’t part of the scientific community.

Sometimes, tigers are called hunters but they can be called customers too. They searching for food like humans go to the supermarket and search what’s inside deep cooling rack. And scientists are doing basically the same. Most of the time they are hanging around in computer stores, in libraries and on conferences, because they are searching for intellectual food. That the currency they are focussed on.

Youtube has a dedicated category under the name “shopping haul”. That are videos about people who are visiting a store and putting everything in the basket what is really good. Usually a haul takes time, because it is not possible to buy everything, so a decision has to be made. A haul can took place in many locations: in a computer store, in a supermarket, in a clothing store and so on. For the case of a science, the only difference is the store. Which has mostly to do with an academic book store or a computer store. If the scientists is a good one, he will take some time for doing a carefully decision and he can explain the details.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.