On Slowing Down AI To Stay in Control

I had a discussion about AI yesterday with my wife. She came back from a two-day conference in Toronto. One of the sessions was about the place of AI in society and how it is time to engage in promoting and organizing some AI regulations.

The more I read and learn about AI capabilities as exposed in tools like ChatGPT, the more I think we will eventually need some regulation. For example, one thing we discussed (and on which we couldn’t agree) is the introduction of a delay in AI training. What I’d like to see is that AI companies are imposed a 2 or 3 years delay for their model training. And why would this be necessary? How would it change the game?

Remember that current training is lagging simply because we lack the processing power to digest all the digital information produced daily. But, eventually, it will come, just like Google replaced Yahoo when index content was initially entered manually by a group of people and then by a community. Sooner or later, ChatGPT or similar tools will digest the web in near real-time. And this is where this is going to be even more scary and could really get out of control.

Imposing delay on models training would help public knowledge and content to settle down and let consensus emerge in any research field, for example. Short-term noise would be reduced. In my opinion, it would be more challenging for ChatGPT to be infected by bad actors who will eventually try to influence results with toxic data.

My wife and I couldn’t agree on the effectiveness of this simple measure. She thinks that it would make ChatGPT useless or less relevant. My take goes the opposite side where, like in real life, things like encyclopedias are still helpful even though they were written and got frozen as soon as they were printed. So there is a need for them, like there is a need for more dynamic knowledge content like Wikipedia.

More than ever, we need to define what makes us unique, how we protect how uniqueness and consider slowing things down a bit, so we can have more time to understand what is going on and where things could go if we let things go without proper framing.

Like Anything Else, The World is Hybrid

DHH wrote “In defence of the office”:

I salute Apple, for example, for sticking to their in-person culture now that the pandemic is long gone. They’re making that choice knowing that some, talented portion of their workforce will leave as a consequence, yet have the confidence that others will fill those chairs. Isn’t this what we wanted? The freedom to choose how we’d like to work by picking between a plentitude of companies offering the style of our preference?

We’re better served by diverse choices because of the diversity of people (profile, aspiration, culture, etc.).

Hot AI Summer (or, my practical uses for AI generators)

This is the sort of empowering interaction I love to experience with these new technologies: I’m not being replaced, I’m getting help, much in the way spellcheck and grammar checks in word processors have been doing for decades.

Interesting parallel here with spell checkers and generative AI. At what point the help we’re getting becomes troubling or questionable? The same goes in real life. Suppose there is something that needs your attention in your house, and a fix is required. Either you know from the get-go what to do and fix it yourself, either you go o YouTube and find a visual guide on how to fix it, or you call someone to your house to fix it for you. What’s wrong with the latter approach?

If there is something that generative AI brings to the table is the inevitable discussion in society of what is the human touch and when is asking for help crosses the lines and what those lines are.

Wait what? Even millennials don’t like algorithms 

I don’t fight the future — we need all the help — but as someone who has made the transition from no technology to some technology to always technology — no matter what I do, there is a tiny bit of me that is still holding on to the analog world.

Om Malik is probably my age. And I feel the same all the time. I know what it was like before the Internet. I prefer chronological over algorithmic timelines. I enjoy human curation. I certainly remember and value the “human touch”.

So that sales pitch of “it’s just a coffee per month” really doesn’t hold water when you think that most people will subscribe to multiple services. Source: It Only Costs a Coffee per Month - Kev Quirk

This is why I maintain a spreadsheet of my monthly (and yearly) spending. And frankly, even with my recent subscriptions cleanup, you wouldn’t want to know my monthly spending on apps and services. Experimenting is not a free ride. Far from it.