- 29 Jan 2023
- 5 min read
AI State Of The Art: It Feels Like Falling Behind
Why, Where, and How modern AI tools matters
By 2023 we had so many powerful AI tools that when you find something like ChatGPT for the first time, it feels like you're falling behind. As of me, I have always been curious about Artificial Intelligence tools, especially from a real-world appliance perspective. By now, there are some many to benefit from than I can't stand sharing my thoughts with you.
WHY AI tools
First, let's clear a common misconception. The term Artificial Intelligence (AI) is heavily overused. There is no complete Artificial Intelligence (at least not in public access). A much better word for everything we know today is "AI tool", and here is why:
Artificial Intelligence, by definition, consists of 7 main branches. Today there are tools successfully combining only 4 of them.
Some AI fields of study (components) are much more developed than others, and some even have their subcomponents. "Vision", for example, has two branches where Image Recognition is the ability of a program to recognize objects or people in an image. And Machine Vision is the application of computer algorithms to extract information from an image.
While we have at least some tools in every branch of Artificial Intelligence, the most advanced ones succeeded in their combination.
Timeline of the most significant AI tools and their components (approximate year of success):
I see considerable potential in Google Duplex as a tool, but there are many things for Google to fix before it becomes stable. That is why it is greyed out.
WHERE AI tools are used
Here are only some of the examples to give you an idea of why AI tools are an integral part of life in the twenty-first century:
- Recommendations engines, fraud/abuse detectors, and author rights trackers are today's most common AI tools. They are everywhere, from Spotify and YouTube music to online shops and social networks.
- With AI predictions, big online shops (like Alibaba and Amazon) know what customers might want to buy and starts delivery upfront. This is the way of achieving one-day delivery even with complex logistics.
- Deep Voice from Baidu adds audio to online books with the author's voice. The author needs to read a sample text for a tool to imitate his voice for the whole book — the same idea for VALL-E from Microsoft.
- The AI algorithms in Amazon Go shops identify the shopper's actions, pair them with sensor information, debit the Amazon shopper's account as they exit the store, and send them a receipt.
HOW can I benefit
There is so much I can show you here, but not to make it boring, I selected only the most exciting ideas from my perspective and divided them into categories.
# Creating — if you have some coding skills
Machine learning must be the oldest and most developed branch of AI, hence the easiest to start from. You can create your first Deep learning model with just a few lines of Python code to predict customer behaviour, stock market trends or even weather patterns.
# MultiLayer Perceptron (MLP) neural network model = Sequential() # fully-connected network layer with number of nodes representing the number of input variables model.add(Dense(x_train.shape, input_dim=x_train.shape, activation='relu')) # fully-connected network layer with 1 node for representing output model.add(Dense(1, activation='sigmoid')) # compile the keras model with loss function for a binary classification problems model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
Of course, the most challenging part of machine learning is model training. The accuracy and reliability of AI-based predictions often depend on the quality of the data used for the projections. There are lots of good books about it, but if you want to understand the idea without spending much time on theory, I recommend you complete this manual and then explore this blog. That is how I found many answers to my questions.
# Evaluating — if you want a ready-to-use tool
There are lots of startups and even big companies that take complex logic of AI behind the scene and give you only beautiful wrappers to use:
- Grammarly — to fix your grammar in any writing you do
- PimEyes Reverse Image Search — to find a person all over the Internet by photo
- Dall-E and Midjourney — to create or edit images by just describing what you want
- Deep Nostalgia — to make your old photos live
...I am sure this list can go on and on. Let me know if I missed your favourite AI tool.
# Applying — if you do not have a specific need
Even without knowing much about AI and being willing to learn something new, you can use one of the most powerful AI tools in your daily life for free. Meet Merlin — GPT3-based Google Chrome plugin.
Here is an example of automating my daily email routine by asking Merlin for the perfect polite response, even without reading a full email. Merlin is basically ChatGPT but in handy wrapping.
Harry up, this tool is soo good that I bet it won't be free very soon.
By exploring modern AI tools and knowing the main AI components, you can see that we are less than halfway from real AI. Four out of seven main AI fields of study are successfully combined in the last couple of years. Now you probably understand why some people (including Elon) are even scared by how fast AI technologies are growing today.
2023 is predicted to be an exciting and innovative year for Artificial Intelligence (AI), and if you still are not using at least Merlin, then you're definitely falling behind.
Originally published at https://medium.com