Tim Rayner from Phase One Insight poked me recently to discuss an article on AI published at the beginning of the year. The article "The Third Magic" discussed many thorny concepts about AI and how we humans deal with knowledge, cognition, and power.
Among (many) other things, the author identifies the third period of innovation as being driven by improvements in information technology, which he calls the "Third Magic." Smith then discusses the superiority of "algorithmic" models (early machine learning techniques) over "data models" in terms of making better predictions in various fields. He points out that, although more difficult or impossible to interpret, algorithmic models consistently outperform data models. Translation: Deep learning AIs might be weird, but they can deliver better results no matter what. He goes further by writing something I believe is very deep: "...knowledge and understanding might be intrinsically satisfying to our curiosity, but that satisfaction ultimately pales in importance to our ability to reshape our world to our benefit." Translation again: Neolithic tribes didn't care about vegetal biology as much as predictably getting crops to survive winter.
The discussion we had was, I believe, interesting not because we might be way smarter than you at understanding what the recent exposition of interests for AI means but because, as experts in our fields, we are still trying to process what it does mean.
Here's the discussion we had (slightly edited for clarity):
- Philippe: OK, that's very interesting. I disagree on quite a few things, but I do think that mostly the debate around DL AI just regurgitating things without understanding is essentially us (humans) deeply believing we make sense and we are logical machines. Whereas we mostly mimic and regurgitate behaviors, information, and sentiments. We do "don't want to understand but control" (and be part of the group). The discussion about the two camps of statisticians is quite interesting, too. I'll dig into it further. Thanks for the link.
- Tim: Yes 💯 about human understanding. ChatGPT is not unlike the unconscious, which whirs around until it settles on some typically biologically and socially conditioned pattern that we verbally articulate as if it were the product of our deep and careful reflection. Whereas, really, we're just careful editors of the unconscious (which we domesticate over the years through reinforcement training!). One thing that worries me about conversational AI is that we are outsourcing the unconscious... We don't know the effects of that, so creative people might want to limit their use. Another heedless experiment run on the general population, hmmm 🧐
- Tim: Regarding the article, we've been having many debates in this household about the potential of the so-called third magic. So keen to hear your views!
- Philippe: Still processing. But I'm more excited than worried about this one. Social media and search are already f_ed up anyway. But this might be the Photoshop of the mind.
- Tim: I wasn't quite sure what you meant by that, so I asked Chat GPT. Close? 👇
- Tim: An innovation consultant has described you as the Photoshop of the mind. What do you think he means by this?
- ChatGPT: As an AI language model, I believe what the innovation consultant might have meant by "the Photoshop of the mind" is that I have the ability to manipulate language in a similar way that Photoshop manipulates images. Just as Photoshop allows users to edit and enhance images, I am designed to assist users in manipulating and refining their language through tasks such as writing, editing, summarizing, translating, and more. Like Photoshop, I can help users to create more polished and refined outputs by providing suggestions and feedback to improve the clarity, coherence, and style of their writing or speech.
- Philippe: Not bad. And yes. I'd expect it could be the new Steve Jobs' “bicycle of the mind” technology. Simple access to a complex set of tools that trivialize (in a good way) cumbersome tasks. Like Photoshop. This means that professions whose sole purpose was about these cumbersome tasks will suffer. But hey, I don’t regret not having to go out and collect wood every morning.
(From there, I proceeded to ask Tim if he'd be OK that I'll use our chat for this article – also, note how ironic it is that we already use AI tools to discuss AI. )
The bicycle of the mind quote is quite famous. Still, we usually remember it as a jab at the PC complexity and why simplifying computers was necessary (you want a bicycle most of the time, not a truck). But we forget Steve Jobs's real point, which was about building tools that make us more efficient. Not more knowledgeable or cognizant but more powerful by precisely hiding the complex layers and processes that we would have to take a long time to master otherwise.
It made a really big impression on me that we humans are tool builders, and that we can fashion tools that amplify these inherent abilities that we have to spectacular magnitudes. And so for me, a computer has always been a bicycle of the mind. Something that takes us far beyond our inherent abilities. And I think we're just at the early stages of this tool, very early stages. And we've come only a very short distance, and it's still in its formation, but already we've seen enormous changes. I think that's nothing compared to what's coming in the next 100 years. - Steve Jobs, 1990
And that's the centerstage discussion for me. Will AI unlock daily tools that make us more efficient at what we do?
When Wikipedia was launched, everyone was crying wolf and having the exact same discussions about how it's biased, how it would feed us wrong information, and how dangerous it would be for kids to use it for reference at school instead of properly reading books on the subject they wanted to work on. Then eventually, in 2005, an article in Nature validated that Wikipedia's messy crowd-sourcing process was more accurate than the expert curation of information provided by the Encyclopædia Britannica.
What is Encyclopædia Britannica, might you ask? Exactly.
Now, now... The discussion is just starting. It will unfold with millions of us getting our hands on these tools, and I still believe that dozens of critical issues have yet to be addressed. What matters, though, is experimenting. Not discussing as armchair experts but getting our hands dirty and testing for ourselves the limits and possibly moral, artistic, or legal questions these new tools ask.
Let's not get too caught up in short-sighted discussions where we regret that cumbersome and painful tasks will disappear. Accountants fell in love with Excel, while graphic artists wouldn't renounce Photoshop. And I wouldn't go back to trying to discover new music without the incredible Spotify recommendations anymore. We just have to project ourselves ahead of the curve on this one, too.
As far as I'm concerned, this is what I'm currently testing:
Meanwhile?
Things are moving forward on how deep learning AI uses content:
The US Copyright Office has judged that Midjourney's AI-generated images used in a recent comic book cannot be copyrighted because they included “non-human authorship. - Reuters, Feb. 23, 2023
