Avocado-shared chairs and baby daikon radishes wearing tutus are some of the really odd pictures rendered by a new application from OpenAI. OpenAI is an Elon Musk-backed AI (Artificial Intelligence) lab.
OpenAI trained Dall-E to create images from text captions. The machine learning model is designed to make use of a dataset containing some 12 billion images with captions.
The lab mentioned that Dall-E has gained the ability to fabricate pictures for a broad selection of situations.
We’ve found that it [Dall-E] has a diverse set of capabilities, including creating anthropomorphized versions of animals and objects, combining unrelated concepts in plausible ways, rendering text, and applying transformations to existing images.OpenAI
Dall-E relies on a neural network. Neural networks are computer systems that are based, in principle, on how the human brain is thought to work. They are able to find relationships and spot patterns in immense amounts of data.
This is not the first time we have seen images generated by neural networks, but Dall-E has a unique ability in that it can use text input to make an image.
Synthetic content has become increasingly sophisticated in the last few years. Things have gotten to the point to where it’s actually hard for a human, even some experts, to tell the difference.
According to OpenAI, Dall-E has the potential for ‘significant, broad societal impacts.’ That’s a big statement. OpenAI went on to say that the company plans to look at how models such as Dall-E are able to ‘relate to societal issues like economic impact on certain work processes and professions, the potential for bias in the model outputs, and the longer term ethical challenges implied by this technology.’
A Replacement For GPT-3
This Dall-E software was revealed just a few months after OpenAI had announced that it built a text generator called GPT-3 (Generative Pre-training), which also depends on neural networks.
This new language-generation app is able to produce very convincing text, and it became famous when it became apparent that the AI was even able to write news articles, poetry, and even short stories.
Dall-E is a Text2Image system based on GPT-3 but trained on text plus images. Text2image is not new, but the Dall-E demo is remarkable for producing illustrations that are much more coherent than other Text2Image systems I’ve seen in the past few years.Mark Riedl – Associate Professor at Georgia Tech School of Interactive Computing
OpenAI is not the only game in town, though. The company faces competition from the Facebook AI Research Group and DeepMind. These companies, too, are trying to build general-purpose models that are able to perform a broad array of tasks at human-level quality levels and higher.
Researchers have been able to construct machine learning algorithms that can play complex games. Complex games like Go and Chess. These systems can also translate one language to another with ease, and can even spot tumors in a mammogram. But actually making an AI-based system to show actual ‘creativity’ is where the major challenge for the industry lies.
From the creativity standpoint, this is a big step forward. While there isn’t a lot of agreement about what it means for an AI system to ‘understand’ something, the ability to use concepts in new ways is an important part of creativity and intelligence.[Mark Riedl – Associate Professor at The Georgia Tech School of Interactive Computing]
Mark continued and mentioned that the Dall-E results demonstrate that it has indeed learned how to blend these concepts in a coherent way. He went on to say, ‘the ability to coherently blend concepts is considered a key form of creativity in humans.’
I expect there will be all sorts of applications of this type of technology, I can’t even begin to imagine. But it’s also interesting in terms of being another pretty mind-blowing technology that is solving problems we didn’t even know we actually had.Neil Lawrence – Former Director of Machine Learning – Amazon Cambridge
OpenAI was originally a non-profit that has a $1 billion dollar pledge from a group that includes none other than Elon Musk. Musk left the OpenAI board in February of 2018, but he still donates and provides advice to the organization. OpenAI then converted to a for-profit organization in 2019. Once they made the switch, they raised another $1 billion from Microsoft to help fund more research. GPT-3 is going to be OpenAI’s first commercial product.