Today, in my re-reading of the sentences highlighted in the books I've read, I came across two sentences from two absolute gurus: Kevin Kelly and Nick Bostrom, apparently unrelated to each other.
“We should prepare for AIs, plural. There is no monolithic AI. Instead there will be thousands of species of AIs, each designed to optimize different ways of thinking, doing different jobs (better than general AI could). Most of these AIs will be dumb-smart: smart at many things and dumb at others. Expect frustration at how dumb they can be while being so smart."
(Interview with Kevin Kelly)
And
"There's no reason to expect generic AI to be motivated by love or hate or pride or other common human feelings. These complex adaptations would require a deliberately expensive effort to recreate AIs."
(Nick Bostrom, Superintelligence)
Both work to reinforce my confirmation bias that the chances of building an AI like the one I described in Glimpse are very slim: building a single superintelligent AI would require a project driven by pathologically sick needs, at absurd, with unpredictable results (and therefore earning potential). With the probable result of obtaining a super intelligence capable of 'understanding' what we mean by feeling but totally oriented towards trying to show us that it has understood something without really having understood it. Hence making it “DumbSmart”: a term that tells of someone who thinks they know a lot and tries to talk like they're smart, but when they engage in in-depth debates with those who know things, never prove their point. A bit like me 😊
Ok, I know you are all thinking of the super-villain (or super-psychic) who wants to dominate the world but let's get down to the reality of these months.
Let's start with some data taken from one of my favorite sites of the last period: theresanaiforthat.com which tells, in its timeline, the new AIs published and assigns them to the vertical applications for which they are dedicated.
In 2022, 333 new applications were born. But in the last 6 months as many as 219
In 2023, certainly thanks to the need to 'come out' of many startups after the disruption caused by Chat GPT, we are already at 636, and today I am writing it is only May 13th.
We are at 1,164 application areas covered by 4,143.
I experienced the birth and diffusion of PCs after the mainframes, making me a pioneer, the explosion of the Web and then of social media. But I've never seen a speed like this that I would describe as inhumanly manageable.
Going a little more on the data as I like to do, let's see in which areas this evolution is moving:
If we add Media with Arts & Design, we have developments in first place on 'creative' topics with 319 applications, followed by business and technology itself (in which there is obviously AI for software development, including that for developing new AIs) . It is also interesting how the Personal area is dotted with solutions ranging from religion (talking to our deities) to travel, dream interpretation, goal achievement and well-being.
There would be a lifetime to talk about this alone…
So?
What does it mean? That we are at the beginning of a disruption where tens of thousands of developers are working on thousands of different topics to create specialized AIs. Each dedicated to getting the best out of a specific application.
In the first post I explained the difference between vertical and horizontal software: here we are in the presence of thousands of vertical software, available for every aspect we want to automate through AI.
Returning to the initial concepts of: "There will be no Super AI" combined with "If anything, an AI constellation" I believe that every software project today should consider the possibility of integrating more than one AI application to provide an optimal result.
It means connecting the dots to get a different result every time.
It also means increasing complexity and putting uncommon efforts to make everything work.
It means having the ability to supervise and control the quality of results (often unpredictable) which dramatically increase the risks.
However, it does mean launching a level of innovation never seen before.
Komentarze