Artificial Intelligence (AI) has been part of our everyday life for a long time and includes applications such as the GPS navigation systems in cars, the targeted ads you see online, the autocomplete features in messaging apps, or the automatic transcripts in live videos, just to name a few.
The next generation of AI tools are called ‘Generative AI’. You most likely have heard of ChatGPT, a chatbot whose release to the public stirred both excitement and apprehension.
Generative AI tools (like ChatGPT) are based on a technology called Large Language Models (LLMs). LLMs are trained with ‘big data’: vast amounts of text, sourced from the internet. Within these huge datasets, they identify patterns of how words are being used, in particular which words occur next to which other words and in what frequencies. Then, based on probabilistic calculations, they produce new text by guessing the next likely element in a sentence. This process has resulted in plausible replication of human-like language!
In addition, these models can ‘learn’: they improve their performance by registering feedback they receive from users. Every time a chatbot response receives positive feedback, this increases the likelihood of similar responses in future interactions.
Beyond just the chatbots, other types of Generative AI tools can produce audio and/or visual outputs following verbal prompts. Such models are trained in similar ways to LLMs, relying on huge datasets made up of images and/or audio recordings available online, with assigned labels to categorise them.
Key to remember: Generative AI tools work with best guesses and not logical thinking!
How is generative AI currently used?
Added features in frequently used apps: You may encounter some recent applications of generative AI in your day-to-day use of the internet and common technology, such as web browsing, email or word processing. For example, Bing Chat can now provide you with answers to web searches that summarise key information from the top hits of your search. Grammarly has incorporated an AI feature that makes suggestions for text generation (in addition to suggested edits of existing text).
AI ‘companions’: These are chatbot tools, such as ChatGPT by OpenAI. ChatGPT is currently free for users and responds to conversation-like prompts by providing human-like verbal outputs. Similar applications from other companies include Bing, Bard, Claude, with varied restrictions on access. These chatbots can produce any type of text, from blog posts to poems. Example prompts that users can try include: requesting definitions, summaries, translations, suggestions, narratives, recipes, and more.
Image making: AI tools can create visual design in seconds after a verbal prompt. Audiovisual applications of such technology may appear, for instance, in ads and other marketing materials, posters and presentations.
Voice recognition and speech to text: this type of technology is being incorporated, among others, in disability assistive software.
Applications in academic research: In academic settings, Generative AI brings the promise of sifting through vast amounts of data with maximised time efficiency. For example, sophisticated tools harnessing the generative AI predictive capabilities are being developed in scientific fields such as biotechnology, for the production of new compounds.
Beyond this list, there are increasing attempts to harness the power of this new technology and find new ways to embed it into analytical processes or professional workflows. Therefore, it is important to have a sound understanding not only of the potential gains but also the limitations, when it comes to using generative AI safely and ethically.