A simple formula for complex research
How to explain the role of AI to non-geeks
‘Hi Siri, would you like to come into my house?’
Artificial intelligence is now a permanent resident of our very own private space and it is important to set the common terms of our relationship with AI. There is a lot of information available about scientific topics and focusing on the relevant information can be complex even for people that work on them.
There is no stupid question about AI. Taking the right decisions as a society is more important than ever and requires that each of us grasps the basics.
We are all curious but afraid of being perceived as stupid when asking questions. Also, you may not feel as the legitimate person to either master the topic or consider their level of knowledge about the topic. From previous life experience (e.g. at school) there can be traumatisms about certain topics (maths, computers). Artificial intelligence is also a frightening topic for most of us since it is related to the loss of control over our lives. Several movies and stories exemplified how AI can turn bad (e.g. The Terminator). But this can be a hook to attract interest !
A definition of Artificial intelligence describes it as the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages. How to translate this in simpler terms?
Imagine you are trying to explain what Artificial intelligence is to your grandmother. She will probably be biased about listening to her grandchild. So let’s take the grandmother of one of your friends (this should keep it unbiased). The best approach here would be using something that both of you know quite well, and use them as building blocks to explain a more complex idea.
“We work with computers. A computer is a kind of calculator that works with electricity and transforms everything we see in a combination of ones and zeros. Ones represent electricity on and zeros represent the electricity off. In this way, we can represent images, text and keep them stored in memory. Artificial intelligence takes care of making computers see, hear, speak and take decisions as humans would do. To do so, people in our field spend a lot of time writing instructions for the calculator”.
This is our recipe for explaining complex ideas:
- Basic rules of popular science are general to most writing tasks. Know your audience, adapt your message and content to the reader. Present in a clear and attractive way.
- Place the work in its context: social, economic, cultural. This motivates people and helps them following your explanation. News and pop culture is what creates excitement and is also common knowledge! What are you talking about, why did you pursue this research, who will be impacted by the results?
- Link your work with the audience, with its background and knowledge. For example, in our work, we may need to explain why do we want to develop a more robust algorithm for recognizing tumors in scans. The result will be used in hospitals and will have a direct impact on the lives of the patients.
- Use simple analogies. Use metaphors. A popular analogy is the one between the human brain and a deep neural network. The algorithm learns from seeing images, as a child’s brain would. An example of a metaphor in AI is the “black-box” metaphor, even used among researchers. This refers to the fact that we do not exactly know how predictions (e.g. an image recognition) are made by deep learning algorithms.
Metaphors, however, can be misleading and the use of the “black box” term in the machine learning community refers to the difficulty to explain the predictions, it is not a rigorous definition.
Practice and feedback: Despite all your efforts, your explanation may be clear to you but not to others. Practice your speech with various people, let various people read and give you feedback for improving the readability.
So, do you want to give it a try? Use our formula to explain AI to your grandmother. If you fail, build an AI system to explain itself.
Thanks to all the authors that participated in writing this post Ojimzt, Mara Graziani, Oreiller Valentin, Adrien Depeursinge
