Faking It
Unveiling the Illusions: A Conversation with Toby Walsh
On Tuesday, I had the pleasure of discussing artificial intelligence with Prof. Toby Walsh, one of the top AI scientists in the world and a “rock star of AI”, according to The Australian. The occasion was the Queensland launch of his most recent book, Faking It! Artificial Intelligence in a Human World.
Queensland AI Hub hosted the launch, which unofficially kicked off Australia’s AI Month (an official launch in Canberra took place the following day). In this newsletter, you can access the recording of our conversation and a summary of some of the points we covered.
On the Genesis of the Book
As always, Toby brought his sharp insights and wit to the conversation, captivating the audience with his perspectives on the evolving realm of artificial intelligence. What made him write this book, number four in his collection so far? The questions he kept getting after writing the previous book. But also the imminence of generative AI. He started writing the book before ChatGPT was released, expecting that generative AI algorithms might create new challenges for our society.
"I realized that we were going to be soon very fooled by all of this fake AI, and AI fakes that were coming into our lives. And perhaps we should prepare ourselves for it." – Toby Walsh
On What Inspired Toby to Become an AI Scientist
Toby recounted the awe he felt as a young boy watching “2001: A Space Odyssey” during the Apollo era, which seeded his passion for AI—a field that has now brought the futuristic visions of his childhood to life.
On AI’s Creative Deceptions
Toby’s fourth book delves into AI's deceptive capabilities. He shared the curious origin of HAL, the AI from “2001: A Space Odyssey,” whose name is eerily (and intentionally) similar to IBM’s. This segued into a broader discussion on the development and ethical considerations surrounding AI, particularly large language models like GPT-3 and its successors.
"We've actually overestimated human intelligence. That there's a huge amount of communication that humans do that requires minimal intelligence."
We discussed the potential of generative AI to deceive society and its negative and positive implications.
“That picture that most of you saw the Pope wearing a white puffer jacket, fake generated by AI, but the problem is, you can see that that guy is His Holiness, the Pope was being fashionable, right? The only person who missed out on that was the Pope. He should have gone out and bought a white puffer jacket and owned it.”
I won’t share the picture of the Pope here: I don’t want to amplify the reach of that particular image. And—curiously—I was banned for a few hours by MidJourney today. My sin was trying to generate a Puffer Jacket Pope. Perhaps The Holy See intervened. So here’s a related image generated just one day after the Puffer Jacket Pope became popular. An imaginary Vatican x Balenciaga colab. Jude Law would wear it.
On The Irony of Progress
As Toby elaborated, large language models have been trained on extensive datasets, including the contentious Books3 dataset, which encompasses a vast array of literature without authors' consent—raising crucial questions about intellectual property in the age of AI.
"There has been no attempt to either seek consent, or no attempt also, to try and return value back to the creators of all of this intellectual property, which seems to me ultimately, not a sustainable place to be."
What’s the irony here? Some of Toby’s books are in this dataset, and some LLMs are now trained on Toby’s words, which caution against irresponsible use of human intellectual property.
On Faking It Through History
Toby guided us through a history of fakery, from the Mechanical Turk to modern-day AI, illustrating how deceit has long been intertwined with technological progress. He beautifully wove this theme into the larger narrative of AI’s journey and its potential pitfalls.
"The Mechanical Turk lives on today. It's a service offered by Amazon where they get humans to pretend to be machines, which is exactly what too many startups have been using to fake artificial intelligence."
The Final Thought from AI
In a bold move, Toby ended his book with a paragraph generated by GPT-3, challenging us to ponder this technology's profound and sometimes disquieting implications. The last sentence he shared encapsulated the eerie wisdom that AI can offer.
“And so I will leave everyone with one universal thought: the smartest person in the room is the one that closes the door.”
Buy “Faking It” in your favourite bookstore. I enjoyed reading it. I am sure you will, too.