Amir Mamaghani, Turing Technology Founder is here to talk about the concept of a ‘master algorithm’, something often discussed in the AI development field…
My initial thoughts regarding this question would allow me to say no, there is no master algorithm in machine learning. However, just saying that is not going to help you, so I’ve tried to explain further.
If you’re talking about Artificial General Intelligence, the concept is that General AI acts as a human equivalent, providing the intelligent performance on a level equivalent or similar to humans. This can be found in development by many companies, like GoodAI.
It is true that there is some research where the artificial intelligence has passed the Turing Test, in which a programme is trying to provide that it is not a machine. However, simply beating the Turing Test is not an indicator of a ‘Master Algorithm’.
Artificial General Intelligence is not the end, Artificial Super Intelligence is the end, or, at least it is one end. The concept of Super AI is the creation of machines or machine that have the power of multiple humans. Right now, we have not yet matched the capabilities of one human, so we may be a way off, but the near future holds promising signs for Super AI, which can be considered far superior to the ‘Master Algorithm’.
Master Algorithm has different definitions to different programmers and experts, but the most generally accepted notion is that the Master Algorithm can learn to process absolutely any given task. This is the point at which an algorithm is the equivalent of one human brain, or, as they say in the movies, it is when robots become more intelligent than humans!
Machine learning is a category of AI algorithms. Whilst machine learning can perform very powerfully, as we have seen in many applications, the significant drawback is that they can only do the specific tasks that they are designed to do.
We have, for example, facial recognition systems built using AI (which are being used by Facebook). These systems perform better than humans for the sole task of facial recognition.
From the category of deep learning, algorithms were able to:
‘there is no one else in the world.
there is no one else in sight.
they were the only ones who mattered.
they were the only ones left.
he had to be with me.
she had to be with him.
i had to do this.
i wanted to kill him.
i started to cry.
i turned to him.’
Aiva, an AI composer, created these songs (though the Soundcloud account was set up by a human). I’ve listened to these songs and they’re actually very, very good. Whoever thought that Pianists could be at threat from robots!
Recreate a movie
The extract below is taken from the blog of the Terence Broad, who recreated Blade Runner’s opening 15 minutes using neural networks…
‘The type of neural network used is an ‘autoencoder’. An autoencoder is a type of neural net with a very small bottleneck, it encodes a data sample into a much smaller representation (in this case a 200 digit number), then reconstructs the data sample to the best of its ability. The reconstructions are in no way perfect, but the project was more of a creative exploration of both the capacity and limitations of this approach.’
According to VentureBeat: ‘Google’s DeepMind uses a deep learning technique referred to as deep reinforcement learning. Researchers used this method to teach a computer to play the Atari game Breakout. The computer wasn’t taught or programmed in any specific way to play the game. Instead, it was given control of the keyboard while watching the score, and its goal was to maximize the score. After two hours of playing, the computer became an expert at the game.’
Every computer game is at threat from AI learning to play, and then dominating the game. Anything from chess and pong, to World of Warcraft and Call of Duty.
‘Stanford researcher Timnit Gebru took 50 million Google Street View images and explored what a deep learning network can do with them. The computer learned to localize and recognize cars. It detected over 22 million cars, including their makes, models, body types, and years.’ (source)
Neural networks are able to study the patterns found throughout pieces of art, such as the strokes, colours, and shading. Using this analysis, they can transfer the original style into a new artwork. Here are two masterful examples — Gene Kogan and DeepArt.
Web design modifications
Website builders such as Wix are now able to help you construct, update and modify your sites with help from AI. The underlying technology behind these systems can take historical data and indicators, such as web conversions, and make changes to increase them.
It can be said that a master algorithm will eventually exist, but right now, it has not been found.