Have you ever seen a photo of an avocado kettle or read a clever article that slightly differs from the topic? If so, you may have discovered the latest trend of artificial intelligence (AI).
DALL-E, GPT and PALM Machine machine learning systems create a name as innovative tools that are able to perform creative tasks.
These systems are assessed as “foundation models” and are not the noise and party tricks. How does this new AI approach work? Does this mean that human creativity will find its end and a deep nightmare will begin?
1. What is the foundation model?
Foundation models work, creating a single large database of general information, and then adapting the scheme to new challenges. Previous models tended to start from scratch for every new challenge. To compare photos (such as a cat shutter for animals) with a signature (“Mr. Fuzzyboots the Tabby Cat relaxes in the sun”) required scanning of hundreds of millions of examples.
After the training, this model is able to say what cats (and other things) look like in the pictures. The model can also be used for several other useful AI tasks, such as creating new photos from the signature itself (“Show me a koala immersing basketball”) or editing photos based on written instructions (“Make this monkey pay taxes”).
2. How does it work?
Foundation models are based on “deep neural networks”, which are loosely inspired by how the brain works. This includes sophisticated mathematics and significant computing power, but it boils down to the complicated form of matching patterns.
For example, a deep neural network can associate the word “cat” with pixel patterns, which often appear in the paintings of cats, such as soft, blurred, hairy dots of the texture. The more examples see the model (the more appropriate the results), the larger the model (the more layers or “depths”), the more complicated these patterns and correlations can be.
In a sense, foundation models are only an extension of “deep learning” models, which dominated AI research over the last decade. However, they have unforgettable or “rising” behaviors that can be both surprising and innovative.
For example, Google's palm language system seems to be able to explain to difficult metaphors and jokes. It goes beyond simply imitating the types of information to which it was originally designed for processing.
3. For now, access is limited.
The scale of these AI systems is overwhelming. Palm has 540 billion parameters, which means that even if everyone on the planet remembered 50 numbers, we still would not have enough storage to recreate the model.
The models are so high that their training requires significant amounts of calculation resources and others. One estimates put the cost of teaching the OPENAI GPT-3 language model at around USD 5 million.
As a result, only the main technology companies, such as Opeli, Google and Baidu, can now afford the construction of foundation models. These companies limit who can use these services, which has economic sense. Use limits can give us hope that these systems will not be used for vile purposes (such as creating false messages or defamatory materials) in the near future. However, independent researchers are also unable to listen to these models and report their arrangements in a transparent and responsible way. So we don't know the implicit use of their use yet.
4. What will these models introduce to “creative” industries?
More foundation models will be created in the near future. Smaller models are already published in Open Source versions. Software companies are starting to experiment with the licensing and commercialization of these services, while AI researchers work hard to make the software more effective and available.
The extraordinary creativity demonstrated by Palm and Dall-E 2 indicates that this technology can affect creative competitions earlier than expected.
As he says, robots would first take over the work of a “blue collar”. Competitions requiring creativity and education, known as the work of a “white collar”, were to be relatively safe before automation.
However, AI deep learning models already stand out in tasks, such as X -rays analysis and the determination of macular degeneration. Foundation models can soon offer cheap and “good enough” creativity in areas such as advertising, copywriting, spare illustration or graphic design.
The future of creative jobs can be slightly different than we expected.
5. What does this mean for legal facts, messages and media?
Because we will not be able to say that creative content is the result of human activity, foundation models will eventually affect the regulations in areas such as intellectual property and evidence.
We will also have to deal with disinformation and disinformation, which are generated by these applications. We have to deal with many disinformation problems, as we see in the developing Russian invasion of Ukraine and the emerging edition of deep false photos and video. Foundation models are prepared to increase these challenges.
Time to plan!
As scientists who examine the influence of artificial intelligence on society, we believe that foundation models will cause serious transformations. They are strictly controlled (for now), so we can have some time to consider their consequences before they become a big problem. Genas are not yet beyond the bottle, but the foundation models are a large bottle, and inside there is a very clever genie.
