AI innovations are quickly penetrating our society. From criminal justice to music recommendations, the applications are already widespread. Often these applications have focused on replacing or augmenting humans in their daily work, although many innovative new uses are being thought up every day. However AI still has a long way to go before replacing humans.
AI still isn’t as intelligent as we think.
The over-hyped GPT releases by Open AI are a case in point. These AI text generators are so good at creating coherent texts that they have caused much concern over potential malicious applications for such technology. Capable of stringing together long, grammatically correct texts, the examples that have been shown of the internet are certainly frightening.
This has caused many to raise concerns about its potential application for such things as generating massive amounts of fake news. What is often overlooked however is the amount of human editing that goes on before the samples are published. While the content that is published is most likely to have been written by GPT itself, it is filtered and selected to show only the best.
This is largely because AI lacks a model of the world that a human naturally gains from living in it. While its texts might resemble perfect English, it’s still liable to say that the sun is cold and green.
Where AI excels
The great advantage AI has over humans is to handle more parameters and inputs than a human can possibly deal with. It is through this brute force strength that Open AI’s text generators achieve their strength, using literally billions of parameters to decide what the next character in a sentence should be. But this brute force strength is more usefully applied to existing multi-parameter problems.
This is most visible in everyday consumer related applications that are being implemented by large companies like Facebook, YouTube and others. Here these companies are taking thousands of data points to better understand their users and customers, delivering recommendations and targeted advertisements almost better than their own family could.
This has seen innovations being made in matching peoples tastes with product, film and music recommendations. Giants like Spotify, Amazon and Netflix are famous for this, and provide their users with an endless stream of highly appealing new suggestions. A brief look at the comments
section of many YouTube videos illustrates just how good AI is at this, with many users commenting that they would never have thought of watching such a video if it weren’t recommended to them.
Small companies innovating as AI becomes more accessible
Advances in hardware speed and affordability have leveled the playing field somewhat. This means AI is no longer the exclusive preserve of gigantic corporations with data centers full of hardware. Now it’s new startups are now driving some of the most adventurous and creative innovations in the field.
One such innovator is Palate Club and their wine tasting app. Using AI for what it’s best at—taking thousands of data points and using them to make seemingly simple recommendations—Palate Club pairs wine drinkers with wines they’ll love.
For every wine that goes into its database, Palate Club collects countless datapoints about the wine. Everything from the colour, right down to the particular type and balance of acid in the wine. Yes, there’s different types of acid in wine!
Then, combining this data with every like and dislike registered by users of their app and any other data they can get their hands on, their AI algorithms then find the best wine recommendations for their users. Surely this will see many users exclaiming to themselves “I would never have drunk merlot before, much less in the summer. But somehow I found myself here and I love it!”