wreeportage

Since the last couple years, AI has been taking over a lot of aspects of our daily lives. Every self-proclaimed expert on social media is talking about AI. Like most other issues being talked on social media, AI obsessed crowd has put themselves into a binary of AI-promoters and AI-haters. The regular social media thrives on polarisation as it increases user engagements on their platforms. This also leaves out the people on whom AI is forced upon but their opinions are not being heard or valued. It also leaves out the nuances of using or not using AI.

I have had quite a few in person discussions regarding AI in my personal day to day life, and I want to highlight a few of those.

My local grocery and cigarette shop is run by a millennial. He is quite friendly a lot of young people frequent his shops and engages in conversations on various topics. A few days back, he asked about my opinion on AI. I mentioned AI is a marketing term and an actual ‘intelligence’ and asked further about why he is asking about AI. He mentioned about a guy who is very specifically pro-AI was talking about how there will be no need for coders and artists anymore because AI is getting so good, it will just replace those people. There were other people in the shop, and all of them got into a debate about whether AI will replace the people or not. The shopkeeper, Pritam and a few other customers vehemently denied that AI would replace any humans because they saw AI as a tool, that needs other humans to maintain it. If something made by AI is also breaks down, we will also require actual humans to fix it. I agreed. This is an argument you can’t deny. But he was also lamenting that the guy was quite adamant about the fact and was dismissing everyone’s opinion about it. We laughed at his naïveté and engaged more in the conversation of what actually is AI.

People intuitively know that there is nothing intelligent about AI but they find it hard to express how or why. I decided to show him this nifty new tool by google called https://quickdraw.withgoogle.com

This tool by google is actually a veiled effort by google to train their neural network to recognise human doodling. There is a quite a bit of gamification going on. The tool is like an online Pictionary, where the user is given a prompt and they have to draw it. The neural network accesses a database full with other doodles and tries to guess what the user is drawing. The user is given six prompts and after it, the tool shows the user the prompts and the drawings together along with other doodles users across the globe has done.

We played one game or one round, and I purposefully drew a few completely haphazard drawings so that the algorithm could not guess what I drew. It provided random guesses during the rounds. But after it shows the answers, We could see the algorithm was trying to pick a few lines from the haphazard doodles we have done and trying to calculate the probability of that shape be a labelled doodle from it’s database. Even if it has nothing to do with the prompt. Pritam realised by seeing the answers that the AI was trying to match our drawing with already existing doodles made by other humans. The database the AI have access to has doodles with their corresponding labels on it. When a user draws something, it checks the database and tries to match the drawing with a doodle already existing on the database. Since quite a few objects in the real world would look the same in a doodle format, the AI calculates the probability of a user-submitted drawing with various doodles it matches in the database. At the end it would return the label of the doodle with the highest probability match with the user-submitted drawing.

Here is an example of a correct guess:

Here is its reasoning:

Here is an example of a wrong guess:

Here is the reasoning for it: