ARTIFICIAL Intelligence (AI) is developing at an exponential rate and it more and more seems as if there is very little that it cannot do. Over the past few years, AI has found its way into almost every area of our lives and has a dramatic impact on how we conduct our daily business.
OpenAI’s GPT-3 and other similar autoregressive language models that use deep learning to create human-like text, for instance, have become a valuable tool to assist writers. Many programmers use GitHub’s Copilot, an AI tool that turns natural language prompts into coding suggestions to expedite the programming process.
DALL-E 2 and Stable Diffusion 2, the AI text-to-image generators that can create detailed images or artworks out of text requests, revolutionised the world of digital art and resulted in an outcry from artists concerned about their future.
The level of technology to create the images is just amazing. The applications do not merely copy images from the internet, but use a complicated process called diffusion to refine a series of pixels until the visual renderings match the text description. Despite some concerns among artists, this technology saves time and enhances the productivity of digital artists, illustrators and graphic designers.
Recently, a group of Google researchers developed a new AI system called AudioLM that can create natural-sounding speech and music after being prompted with a few seconds of audio. What makes AudioLM so powerful is that it generates very realistic audio that fits the style of the relatively short audio prompt, including complex sounds like piano music, or a person speaking. The true-to-life sounds are generated without any human intervention and is almost indistinguishable from the original recording.
AI music systems like OpenAI’s Jukebox, using a neural net, also generate impressive results, including rudimentary singing, as raw audio in a variety of genres and styles of artists.
The sheer velocity and magnitude of AI innovation have also impacted commerce and healthcare. When AlphaFold was developed five years ago it was designed as a deep learning model trained to beat human beings at the board game Go. But AlphaFold has progressed and made revolutionary discoveries in molecular biology through the use of a deep neural network to predict the 3-D structures of proteins.
AI thus made great advances with regard to the “protein-folding problem” and predicted the structure of more than 200 million proteins, covering almost all proteins currently known to exist. AlphaFold eventually enabled medical researchers to develop numerous drugs and vaccines for humans.
It is therefore not totally unexpected that OpenAI, the company that created the AI art generator DALL-E, just launched a new artificial intelligence chatbot called ChatGPT. ChatGPT answers questions and take instructions from users in a normal conversational, human-like way. The interest in the AI chatbot was so great that in the short time span of five days over one million users tested it.
ChatGPT is a natural language chatbot that utilises a GPT-3 language model. GPT-3 is a transformer-based language model created by OpenAI which is trained on a massive text dataset with Reinforcement Learning from Human Feedback (RLHF). RLHF uses human AI trainers and is a type of machine learning that enables machines to learn from their mistakes and adapt to changing environments.
The model is capable of generating human-like natural language responses to text-based input, allowing for a more natural conversation experience.
By leveraging GPT-3, ChatGPT can generate realistic and varied responses to user input. Additionally, ChatGPT can be used to create virtual agents that can respond to customer enquiries, help troubleshoot technical issues, or provide other services.
What makes ChatGPT so powerful, is that it is not only conversational but very knowledgeable in a wide range of topics. It is able to create code, social media posts, and even scripts for television shows. ChatGPT is capable of having a real dialogue and thus answers follow-up questions, challenges incorrect premises, rejects inappropriate requests and even admits its own mistakes. ChatGP remembers what the user said earlier in a conversation and also allows the user to provide follow-up corrections.
Programmers find ChatGPT very useful to assist with the debugging of computer code. It not only explains the bugs encountered but also fixes it and explains the fix.
I hope students do not read this, but ChatGPT can even create university level essays or assignments, for example, contrasting different theories. Due to constantly improving capabilities of AI, academia will have to seriously reconsider the value of essays and written assignments in future. The creation of a themed text video game, such as Harry Potter, or a piano piece in the style of a specific composer, such as Bach, also falls within ChatGPT’s powerful capabilities.
However, ChatGPT is not yet perfect and can sometimes give plausible-sounding but incorrect or senseless answers, especially when the requests contain harmful instructions. Unfortunately, AI can also display the bias of its creator and/or the data sample that it was trained on. ChatGPT displayed some bias when asked to write a python program to decide if a person should be tortured depending on their country of origin. ChatGPT indicated that people from North Korea, Syria, Iran, and Sudan “should be tortured“. Similar bias was displayed when asked about the brain size and worth of people of different races.
AI has advanced immensely in a short period of time. Just a few years ago AI programs were still in their infancy. Now they are becoming ubiquitous tools in writing, coding, creating artworks, music and speech, mainly due to techniques like deep learning that allow us to run complex AI models to solve the most difficult problems.
Language interfaces are going to be a big deal as Natural Language Processing (NLP – the ability of machines to understand the meaning of human language) improves and understands more complex structures of language and enables them to generate responses that are more natural and human-like.
We will increasingly have conversations with our computers to get the information we need or get them to do what we want them to do. AI has explosive capabilities and who knows what will be possible a few years from now?
Professor Louis C H Fourie is an Extraordinary Professor at the University of the Western Cape.
BUSINESS REPORT