We use cookies to provide and improve our services. They allow us to remember some of our preferences and improve the overall site performance. View our Privacy policy.
IMG 0023

Three Questions with Prof. Ed Hovy, Executive Director of Melbourne Connect

1. How has the development of AI tools progressed over the past five years?

I think the pace of development is speeding up rapidly. At first, AI systems were built by hand and nobody would willingly interact with them except out of curiosity. In the 1990s, AI machine learning began to address many tasks — from image recognition and language translation to online games and robot soccer.

In the 2000s, when the vast knowledge trove of the web became accessible, neural network algorithms (which had been hamstrung since the1970s) came to their own with a vengeance.

2. What are some of the new AI tools that we should be aware of?

By far, the two most prominent tools announced over the past 6 months are DALL-E and ChatGPT, both produced by OpenAI in San Francisco.

DALL-E produces images in response to a user’s prompts such as “paint the Mona Lisa as an old woman in a factory” or “draw a picture of a dog on a unicycle eating a carrot”. ChatGPT conducts a chat dialogue with the user in over a dozen languages (including Latin), about almost any topic you can imagine — from quantum physics to how to dye your hair. ChatGPT can write poems and songs, create new storylines for plays and books, argue about complex ethical questions and write software programs. Generally, it is the closest thing to humans in intelligence.

These kinds of systems come about by having a neural network algorithm skim billions of images with captions or documents, and breaking them down into micro-characteristics then building up a gigantic store of this knowledge.

3. How are they going to change the way we live, learn, and work this year and beyond?

Just as a calculator is faster and more accurate with multiplication and division, ChatGPT has the potential to help you write almost anything, in many languages and using its own knowledge (which vastly outstrips your own). When correct, its answers are far more to the point than the results of a Google search. The possibilities for increasing your performance are almost limitless.

Unfortunately, these new tools call into question the veracity of almost anything you see or read. DALL-E, as a successor to the deep fake images of past years, can create images so realistic that a normal user cannot tell they are fake. ChatGPT’s messages, especially when it is just a paragraph or sentence, look so realistic that university professors worry about how to ensure test answers, essays, and software programs are actually produced by their students. Many journalists use ChatGPT routinely to write their first drafts, using a cycle of questions and refinements.

I expect we are going to see an increasing number of near-human performance AI systems appearing every few months. AI is no longer just robot factories, clever systems that play chess and operate television quiz shows and self-driving cars. I think we will all start using, and even owning, AI-driven products every day.