Artificial Intelligence (AI) is marching onward at an alarming pace. Generally speaking, there are two applications for AI and they are generative AI and Discriminative AI. Let’s take a look a both here.
When the AI is able to create pictures, writing samples, audio, and 3D printers built from computer-controlled systems that is what is termed generative AI.
Being able to distinguish actual people in photographs or certain written words or words in speech from fake is part of discriminative AI.
Both forms of AI can learn from basic items, words, speech patterns, or rudimentary knowledge and grow their understanding from there in leaps and bounds. Both are founded on neural networks that respond to input for the creation of output. Both outputs come from feedback based on their prior responses or predictions of responses. They are both trained on extremely large volumes of data.
ChatGPT and Google Bard are both applications of AI being used in real-time today. Both take generative AI and Discriminative AI and blend them into what is called generative adversarial networks or GANs. These GANs help to train both types of AI and as such improve their performance in response to queries.
AI is able to take a series of class pictures, some real, and intersperse those with fake pictures. The generative form of AI gets better and better at creating real photos while the discriminative counterpart becomes more successful at pointing out the fake ones.
Some industries are keen to use this new technology. For example, genetic research hopes to be able to more easily understand how genes combine and turn on or off or become more active from the use of AI. They also want to learn how responses to various changes in genes will be expressed. This might lead to a better understanding of how best to develop gene therapies or predict medications to target specific responses in a person’s genetics.
Manufacturing is interested in how to create machined parts or assemblies via 3D printing and computer-controlled machining with less waste, more speed, and increased efficiency. Optimization of design is also a by-product of these AI applications. Truly doing more with less is the future of manufacturing.
Arts and entertainment can also increase their creative outputs via the use of ChatGPT or Dall-E. These platforms are already generating background music for games, guiding conceptual scenarios and environmental development. With any creative venture comes the possibility of copyright and intellectual property infringements. This may lessen the use of those tools in that context. Time will be the ultimate determinant in that regard.
GANs have assisted with the production of unique text since ChatGPT has been available for mainstream use. Open AI has not unleased its third version of text-generating responses from text prompts. That is the chatbot. The responses to this have been incredible pushing for larger language models (LLMs). Bloom, Flamingo, and Jasper are a few newcomers. They can all create memos, starter codes, condolence cards, school essays, and meeting minutes. Impressive right?
There are some hitches and those are mostly the generation of syntactically incorrect replies or self-contradictory statements. Sometimes the programs simply make up a reply. The professionals call these hallucinations. The rule is not to simply count on the AI-generated information as gospel, check, check, check.
As this AI learns more and more they will become better at text-to-speech replies and also image and space synthesis making them able to answer phone calls using realistic voice intonations, cadence, and volume; translation of arbitrary text into better quality images; and design buildings and spaces using the specific codes required and materials needed.
No matter where you sit on the spectrum of AI it is here to stay.
References: