Why AI Is Not the Next Television - Tech Insiders News, April 2023
Welcome to the Digital Explorers' Diary #35!
A curated collection of thought-provoking topics about interactive technologies, AI, web3, sense-making, entrepreneurship, and psychology.
Unintended consequences: AI is not the next TV
You probably have recently seen posts on social networks or news websites saying something along these lines:
When the TV was released, many people said that it will harm our society greatly. It didn't. Now the same thing is happening with AI. You’ll see, AI will be safe and beneficial to our society’s progress.
(This is not an exact quote, but you get the gist.)
Well, first, this is a fallacy named False Analogy. It’s not because these two things are related in one aspect (they are both technologies) that one can draw the same conclusion.
Let’s dive in a bit. Oh, and you can replace AI with many other emergent technologies: CRISPR, weaponized drones, and many biotechnologies…
As I mentioned many times here: technology is not values neutral. TV maybe didn’t harm our society greatly, but it deeply changed the dynamics of our world. It became the main source of information and entertainment. It allowed amazing things: the world could witness events happening live worldwide. It generated a whole new market for advertisers, and perhaps, it allowed the easier spread of misinformation and lobbying from governments.
But it was not easy to start a new TV channel, and it did cost quite a bit for a household to buy one. Timelines vary between parts of the world, but after World War 2, it took ~10 years for TV to become mainstream. As a reminder, ChatGPT reached 100 million users in… two months! Speaking of WW2, before the Nuclear Bomb, humanity didn’t have any means to destroy itself at scale in a matter of days. The bomb is not easy to make though; non-state actors cannot create one.
Even though it still takes a lot of money to train a large AI model, anyone can now use them for a few bucks a month; and more models are becoming available for offline usage, the last example being Stability AI gradually releasing its LLM.
So yeah, ease of access and spread velocity, this is what differs from the TV, or the Walkman, or the Radio, or any other technology-that-will-make-our-youth-dumb.
I’m not saying that AI is entirely bad; there is too much uncertainty right now, but as a society, we need to be careful about the exponential spread of high technologies.
Imagine if, in a few years, it becomes relatively easy and cheap to train a GPT-4-like model and remove all securities; what do you think ill-intentioned groups will do?
In Other News
- Direct access to GPU is arriving in Chrome, and they focus the press release on the improvement it will bring to 3D on the Web. But do you know what else works better on GPUs? Humm? Yeah… AI. Here again, another step to make the technology more accessible.
- A few weeks back, I mentioned how I believe in AR Glasses becoming the next way to interact with the Internet of Objects, and releases such as this new segmentation model from Meta and this If Seen Then That demo is just the beginning!
- Snapchat is targeting retail and physical spaces. The major announcement is called AR Mirrors, a set of technologies that allow in-store virtual try-on, and it’s already in use by Coca-Cola, Nike, and Princess Polly, amongst other big names.
I’m really curious to see if it will succeed. Many companies have been trying for more than 10 years. Will Snapchat be able also to catch offline users?
From the Podcast
Every week, I’m teaming up with Guillaume Brincin and Sébastien Spas on the Lost In Immersion podcast. This week:
- We expand on The Sims AI agents I mentioned last week by debating its interest: do we want our virtual worlds to be populated by AI agents, or should we focus on having «human-only» interactions?
- Seb does a recap of his visit to Laval Virtual. He tested all the major MR Headsets: Pico, XR Elite, Varjo, Meta Quest Pro, and Lynx. He feels that if Lynx can fix the tracking issues he saw, it can become the best headset on the market.
The HapitX gloves didn’t convince him, as the touch didn’t give him a real feeling, but he nuances that by saying that it might be a calibration issue as other users were impressed by it. - We then discuss the various initiatives to automate AI instances, notably AutoGPT and its variant ChaosGPT, which anonymous developers created and gave as a goal to destroy humanity. Fantastic. How can we regulate this type of misuse? We also mentioned the concerns around privacy; some countries and companies are starting to ban ChatGPT usage. We highlight how fast generative AI moved to paid subscriptions: Adobe Firefly, OpenAI, Midjourney, and all the leaders are now making users pay for the optimal service.
Tune in at 10 am UTC on Wednesdays on Twitch, and listen to the replays on Youtube!