PDA

View Full Version : Why is Artificial Intelligence still so bad at suggesting ads and products?



Maciamo
29-12-20, 15:41
I have been using Google for 20+ years, performing dozens of searches per day, and seeing hundreds of ads on various websites almost every day. Yet, I think I have only clicked twice on a Google ad in my life, as the ads never seem to be relevant to my interests. It's all the stranger since I accepted from the start that Google customise the ads displayed based on my search history.

Amazon doesn't do much better. I have also been a regular customer for some 20 years. I have bought hundreds of books on Amazon, first paper ones, then on Kindle. Yet the Kindle recommendations rarely show anything I might be interested in. I always have to actively search for new titles as the AI isn't doing its job. Prime Video is even worse in that regard. It's as if they were trying to push the same recommendations on everyone regardless of their personal preferences.

Netflix is marginally better, but not for the notifications in the top-right corner (bell icon), which are always irrelevant unless it is to announce a new season of a series I have watched.

It's not that I want Google to serve me ads I like all the time, but I wonder how their business model works (and apparently it does work) if other people also get irrelevant ads. For Amazon and Netflix it is more annoying as I could be saving time if the AI was doing its job, but instead it keeps browsing indefinitely for new books, movies and series.

Have you had similar experiences?

don_joe
01-01-21, 14:14
Maybe AI does a good job statistically for average users. You don't seem to be average, your preferences are more refined. The AI categorises you wrongly. Users like you have lower significance, since the profit comes from the middle part of the Gauss bell curve.

Making the AI model work for the extremes would be overfitting in IT jargon. Not worth the effort.

don_joe
08-01-21, 13:33
Recently I read an article about how developers from the Facebook team called their own ad algorithms useless. The article provides statistics that advertising hits the target audience in less than 10 percent of cases.
It is unlikely that such algorithms can be called AI. Also, I don't think that other algorithms used by such companies have gone far from them.
Of course you are right about the fact that ads are shown for which you paid more, which means that this is a more massive product. But I think there is more of a problem with what we call AI.
You will not say that a rocket programmed to launch a satellite into orbit is an AI rocket?


I agree with you about the term AI. But if you hit the target in near 10% of cases, your marketing team is good. :)

don_joe
13-01-21, 20:06
Hmm)) I was called AI on another forum)) And my thread turned into a discussion of artificial intelligence. It's cool. The guys even provided a link to a YouTube video called A robot wrote this movie. Judging by these videos, AI is really progressing lately.
Perhaps in a few years time there will indeed be some good practical use for this rather impressive technology.
But to create real AI, you need to explore the human brain to the end. Now they think that the brain has been researched by 5% or so. Although no one has proven it.

You're right about the brain. AI development takes another course probably than mimicing the human brain. For the reasons that you mention among others. I've read a book from Mathlab "Deep Learning", Phil Kim. I'm a layman, I've read it cause a friend of mine recommended it as an "easy read". I didn't understand most of the details but it is obvious that AI is basically only a bunch of concepts and logical theoretical solutions even thought it has a longer history than we think. This guy wrote a book with the goal to shed some light on this topic. He himself, as an IT expert actually, as he started to deal with this subject, couldn't find any standardised documentation or agreed nomenclature in this field. So he wrote a book himself to help the other guys having the same problem. It's all highly theoretical. Some genious inventions like back propagation algorithm have solved big theoretical problems. AI works well as an idea, the training of AI is still suboptimal and the main reason why it's not broadly in application. The dumb thing just won't learn. :) For certain things like face recognition it actually works well. Take my interpretations with a grain of salt, I'm not able to grasp a lot in this subject. But even smart people from IT industry have difficulties with it, to my comfort. :)